ML Models | MQL Functions

ML Models

ml algos ml train ml train ml fit ml fit ml model predict ml model predict_proba ml model transform ml model inverse_transform ml model score ml model evaluate ml model importance ml model predict_interval ml model permutation_importance ml model partial_dependence ml model explain ml model rules ml model pmml ml model pmml ml model info ml model show ml model exist ml model delete ml model delete_all ml model save ml model load ml cross_validate ml grid_search ml compare ml compare ml automl ml automl

ml algos

Description

Get the list of the algorithms with their kind, the function to use (ml train or ml fit), the parameters with their default values and a description

admin
ml algos

mentdb
[{"algo": "linear", "kind": "regressor", "function": "ml train", "defaults": {"fit_intercept": true}, "description": "Ordinary least squares with the full statistics (coefficients, std errors, t, p-values, R2, F, AIC, BIC, Durbin-Watson)"}, {"algo": "ridge", "kind": "regressor", "function": "ml train", "defaults": {"alpha": 1, "fit_intercept": true}, "description": "Linear regression with L2 penalty"}, {"algo": "lasso", "kind": "regressor", "function": "ml train", "defaults": {"alpha": 1, "max_iter": 1000, "tol": 1e-6}, "description": "Linear regression with L1 penalty (variable selection)"}, {"algo": "elastic_net", "kind": "regressor", "function": "ml train", "defaults": {"alpha": 1, "l1_ratio": 0.5, "max_iter": 1000, "tol": 1e-6}, "description": "Linear regression with L1 + L2 penalties"}, {"algo": "huber", "kind": "regressor", "function": "ml train", "defaults": {"epsilon": 1.35, "max_iter": 100, "tol": 1e-8}, "description": "Huber robust regression (IRLS): the outliers get a lower weight"}, {"algo": "ransac", "kind": "regressor", "function": "ml train", "defaults": {"residual_threshold": "MAD of y", "max_trials": 100, "min_samples": "n_features + 1", "seed": 42}, "description": "RANSAC robust regression: best linear model on random subsets, refit on the inliers"}, {"algo": "wls", "kind": "regressor", "function": "ml train", "defaults": {"weights": [1, 1, 1]}, "description": "Weighted least squares with the statistics (one weight per row, ex: 1 \/ variance)"}, {"algo": "isotonic", "kind": "regressor", "function": "ml train", "defaults": {"increasing": true}, "description": "Isotonic regression (monotonic, one feature, pool adjacent violators)"}, {"algo": "poisson", "kind": "regressor", "function": "ml train", "defaults": {"max_iter": 100, "tol": 1e-8}, "description": "Poisson regression (counts, log link): coefficients, rate ratios, deviance, AIC"}, {"algo": "logistic", "kind": "classifier", "function": "ml train", "defaults": {"alpha": 0, "max_iter": 100, "tol": 1e-8}, "description": "Logistic regression (IRLS): odds ratios and p-values for 2 classes, one-vs-rest beyond"}, {"algo": "naive_bayes", "kind": "classifier", "function": "ml train", "defaults": {"var_smoothing": 1e-9}, "description": "Gaussian naive Bayes"}, {"algo": "lda", "kind": "classifier", "function": "ml train", "defaults": {"shrinkage": 0}, "description": "Linear discriminant analysis"}, {"algo": "qda", "kind": "classifier", "function": "ml train", "defaults": {"reg_param": 0}, "description": "Quadratic discriminant analysis"}, {"algo": "knn_classifier", "kind": "classifier", "function": "ml train", "defaults": {"k": 5, "metric": "euclidean", "weights": "uniform", "scale": false}, "description": "k nearest neighbors (metrics: euclidean, manhattan, chebyshev, minkowski + p, canberra, cosine)"}, {"algo": "knn_regressor", "kind": "regressor", "function": "ml train", "defaults": {"k": 5, "metric": "euclidean", "weights": "uniform", "scale": false}, "description": "k nearest neighbors regression"}, {"algo": "decision_tree_classifier", "kind": "classifier", "function": "ml train", "defaults": {"max_depth": 0, "min_samples_split": 2, "min_samples_leaf": 1, "max_features": "all", "criterion": "gini", "seed": 42}, "description": "CART decision tree (gini or entropy), readable rules (max_depth 0 = no limit)"}, {"algo": "decision_tree_regressor", "kind": "regressor", "function": "ml train", "defaults": {"max_depth": 0, "min_samples_split": 2, "min_samples_leaf": 1, "max_features": "all", "seed": 42}, "description": "CART regression tree, readable rules"}, {"algo": "random_forest_classifier", "kind": "classifier", "function": "ml train", "defaults": {"n_trees": 100, "max_depth": 0, "min_samples_leaf": 1, "max_features": "sqrt", "criterion": "gini", "bootstrap": true, "seed": 42}, "description": "Random forest, out-of-bag accuracy, feature importance"}, {"algo": "random_forest_regressor", "kind": "regressor", "function": "ml train", "defaults": {"n_trees": 100, "max_depth": 0, "min_samples_leaf": 1, "max_features": "all", "bootstrap": true, "seed": 42}, "description": "Random forest regression, out-of-bag R2"}, {"algo": "extra_trees_classifier", "kind": "classifier", "function": "ml train", "defaults": {"n_trees": 100, "max_depth": 0, "min_samples_leaf": 1, "max_features": "sqrt", "bootstrap": false, "seed": 42}, "description": "Extremely randomized trees"}, {"algo": "extra_trees_regressor", "kind": "regressor", "function": "ml train", "defaults": {"n_trees": 100, "max_depth": 0, "min_samples_leaf": 1, "max_features": "all", "bootstrap": false, "seed": 42}, "description": "Extremely randomized trees regression"}, {"algo": "gradient_boosting_classifier", "kind": "classifier", "function": "ml train", "defaults": {"n_estimators": 100, "learning_rate": 0.1, "max_depth": 3, "subsample": 1, "seed": 42}, "description": "Gradient boosted trees (log loss)"}, {"algo": "gradient_boosting_regressor", "kind": "regressor", "function": "ml train", "defaults": {"n_estimators": 100, "learning_rate": 0.1, "max_depth": 3, "subsample": 1, "seed": 42}, "description": "Gradient boosted trees (squared loss)"}, {"algo": "svm_classifier", "kind": "classifier", "function": "ml train", "defaults": {"type": "c", "C": 1, "nu": 0.5, "kernel": "rbf", "gamma": "scale", "degree": 3, "coef0": 0, "probability": true, "scale": true, "seed": 42}, "description": "Support vector machine (libsvm, kernels rbf, linear, poly, sigmoid; type c or nu; C auto = cross validated C and gamma)"}, {"algo": "linear_svm", "kind": "classifier", "function": "ml train", "defaults": {"C": 1, "penalty": "l2", "multiclass": "ovr", "tol": 0.1, "max_iter": 1000, "scale": true}, "description": "Linear SVM (liblinear), very fast on many rows and columns; penalty l1 selects the variables; multiclass ovr or crammer_singer; C auto"}, {"algo": "liblinear_logistic", "kind": "classifier", "function": "ml train", "defaults": {"C": 1, "penalty": "l2", "tol": 0.01, "max_iter": 1000, "scale": true}, "description": "Regularized logistic regression (liblinear) with probabilities; penalty l1 gives a sparse model (C = inverse of the regularization)"}, {"algo": "linear_svr", "kind": "regressor", "function": "ml train", "defaults": {"C": 1, "epsilon": 0.1, "loss": "squared", "tol": 0.01, "max_iter": 1000, "scale": true}, "description": "Linear support vector regression (liblinear)"}, {"algo": "svm_regressor", "kind": "regressor", "function": "ml train", "defaults": {"type": "epsilon", "C": 1, "epsilon": 0.1, "nu": 0.5, "kernel": "rbf", "gamma": "scale", "probability": false, "scale": true}, "description": "Support vector regression (epsilon-SVR or nu-SVR; probability = Laplace prediction intervals; C auto)"}, {"algo": "mlp_classifier", "kind": "classifier", "function": "ml train", "defaults": {"hidden": [10], "activation": "relu", "epochs": 500, "learning_rate": 0.01, "batch_size": 32, "alpha": 1e-4, "tol": 1e-6, "seed": 42}, "description": "Multilayer perceptron (backpropagation, Adam, softmax + cross entropy, stops when the loss is stable)"}, {"algo": "mlp_regressor", "kind": "regressor", "function": "ml train", "defaults": {"hidden": [10], "activation": "relu", "epochs": 500, "learning_rate": 0.01, "batch_size": 32, "alpha": 1e-4, "tol": 1e-6, "seed": 42}, "description": "Multilayer perceptron regression (backpropagation, Adam)"}, {"algo": "bagging_classifier", "kind": "classifier", "function": "ml train", "defaults": {"base": "decision_tree_classifier", "n_estimators": 10, "max_samples": 1, "seed": 42}, "description": "Bagging of any classifier (base = algorithm name or {\"algo\": ..., \"params\": {...}}), out-of-bag accuracy"}, {"algo": "bagging_regressor", "kind": "regressor", "function": "ml train", "defaults": {"base": "decision_tree_regressor", "n_estimators": 10, "max_samples": 1, "seed": 42}, "description": "Bagging of any regressor, out-of-bag R2"}, {"algo": "adaboost_classifier", "kind": "classifier", "function": "ml train", "defaults": {"base": {"algo": "decision_tree_classifier", "params": {"max_depth": 1}}, "n_estimators": 50, "learning_rate": 1, "seed": 42}, "description": "AdaBoost (SAMME, multiclass) on any classifier, weighted resampling"}, {"algo": "adaboost_regressor", "kind": "regressor", "function": "ml train", "defaults": {"base": {"algo": "decision_tree_regressor", "params": {"max_depth": 3}}, "n_estimators": 50, "learning_rate": 1, "seed": 42}, "description": "AdaBoost.R2 (Drucker) on any regressor, weighted median"}, {"algo": "stacking_classifier", "kind": "classifier", "function": "ml train", "defaults": {"estimators": ["naive_bayes", "knn_classifier"], "final": {"algo": "logistic", "params": {"alpha": 1}}, "k": 5, "passthrough": false, "seed": 42}, "description": "Stacking: out-of-fold probabilities of the estimators feed a final classifier"}, {"algo": "stacking_regressor", "kind": "regressor", "function": "ml train", "defaults": {"estimators": ["linear", "knn_regressor"], "final": "ridge", "k": 5, "passthrough": false, "seed": 42}, "description": "Stacking: out-of-fold predictions of the estimators feed a final regressor"}, {"algo": "voting_classifier", "kind": "classifier", "function": "ml train", "defaults": {"estimators": ["naive_bayes", "logistic"], "voting": "soft", "weights": [1, 1]}, "description": "Vote of several classifiers: mean of the probabilities (soft) or majority (hard), optional weights"}, {"algo": "voting_regressor", "kind": "regressor", "function": "ml train", "defaults": {"estimators": ["linear", "knn_regressor"], "weights": [1, 1]}, "description": "Weighted mean of several regressors"}, {"algo": "one_r", "kind": "classifier", "function": "ml train", "defaults": {"min_bucket": 6}, "description": "OneR (Holte): the best single feature cut in intervals, a very readable baseline"}, {"algo": "ripper", "kind": "classifier", "function": "ml train", "defaults": {"folds": 3, "min_weight": 2, "optimizations": 2, "seed": 42}, "description": "RIPPER (Cohen, JRip in Weka): ordered if-then rules grown by FOIL gain, pruned, stopped by description length, optimized"}, {"algo": "m5", "kind": "regressor", "function": "ml train", "defaults": {"min_instances": 4, "pruned": true, "smoothed": "auto"}, "description": "M5 model tree (M5P in Weka): a tree with a linear model in each leaf, pruned, smoothed (auto = if it lowers the cross validated error), readable rules"}, {"algo": "gaussian_process", "kind": "regressor", "function": "ml train", "defaults": {"kernel": "rbf", "length_scale": "auto", "signal_variance": "auto", "noise": "auto", "restarts": 3, "seed": 42}, "description": "Gaussian process regression (kernels rbf, matern32, matern52; hyperparameters by maximum likelihood), prediction intervals; up to 1000 rows"}, {"algo": "bayes_net", "kind": "classifier", "function": "ml train", "defaults": {"structure": "tan", "discretize": "mdl", "bins": 5, "max_parents": 2, "alpha": 0.5}, "description": "Bayesian network classifier (BayesNet in Weka): structure naive, tan (tree augmented) or k2 (score search), features discretized (mdl = Fayyad-Irani or frequency)"}, {"algo": "cost_sensitive_classifier", "kind": "classifier", "function": "ml train", "defaults": {"base": "naive_bayes", "cost": [[0, 1], [5, 0]], "method": "minimize_cost", "seed": 42}, "description": "Cost-sensitive classification of any classifier: cost[true][predicted]; minimize_cost = class of minimum expected cost, reweight = rows resampled by the cost of their class"}, {"algo": "quantile_regression", "kind": "regressor", "function": "ml train", "defaults": {"quantile": 0.5, "max_iter": 500, "tol": 1e-8}, "description": "Linear quantile regression (median by default): a conditional quantile instead of the mean, robust to the outliers, prediction bands"}, {"algo": "pipeline_classifier", "kind": "classifier", "function": "ml train", "defaults": {"steps": ["scaler_zscore", {"algo": "pca", "params": {"n_components": 2}}], "final": "logistic"}, "description": "Pipeline: transformers (scalers, pca, ica, nmf, kernel_pca) then a classifier, saved as one model (the new rows get the same preparation)"}, {"algo": "pipeline_regressor", "kind": "regressor", "function": "ml train", "defaults": {"steps": ["scaler_zscore"], "final": {"algo": "knn_regressor", "params": {"k": 5}}}, "description": "Pipeline: transformers then a regressor, saved as one model"}, {"algo": "calibrated_classifier", "kind": "classifier", "function": "ml train", "defaults": {"base": "naive_bayes", "method": "sigmoid", "k": 5, "seed": 42}, "description": "Probability calibration of any classifier on out-of-fold predictions: sigmoid (Platt) or isotonic"}, {"algo": "kmeans", "kind": "clustering", "function": "ml fit", "defaults": {"k": 3, "n_init": 10, "max_iter": 300, "tol": 1e-4, "seed": 42}, "description": "k-means (k-means++ initialization)"}, {"algo": "fuzzy_cmeans", "kind": "clustering", "function": "ml fit", "defaults": {"k": 3, "m": 2, "max_iter": 300, "tol": 1e-5, "seed": 42}, "description": "Fuzzy c-means: membership degree of each point to each cluster"}, {"algo": "gmm", "kind": "clustering", "function": "ml fit", "defaults": {"k": 2, "max_iter": 200, "tol": 1e-6, "reg_covar": 1e-6, "seed": 42}, "description": "Gaussian mixture (EM, full covariances), AIC and BIC"}, {"algo": "som", "kind": "clustering", "function": "ml fit", "defaults": {"rows": 5, "cols": 5, "epochs": 100, "learning_rate": 0.5, "sigma": 2.5, "standardize": true, "seed": 42}, "description": "Self-organizing map (Kohonen): predict = node index, transform = [row, col] on the grid"}, {"algo": "ica", "kind": "transformer", "function": "ml fit", "defaults": {"n_components": "all", "max_iter": 500, "tol": 1e-6, "seed": 42}, "description": "Independent component analysis (FastICA): separate mixed signals into independent sources"}, {"algo": "nmf", "kind": "transformer", "function": "ml fit", "defaults": {"n_components": 2, "max_iter": 500, "seed": 42}, "description": "Non-negative matrix factorization X = W H (spectra, topics, mixtures): transform = W"}, {"algo": "kernel_pca", "kind": "transformer", "function": "ml fit", "defaults": {"n_components": 2, "kernel": "rbf", "gamma": "1 / n_features", "degree": 3, "coef0": 1}, "description": "Kernel PCA (rbf, poly, sigmoid): non linear components, up to 1500 rows"}, {"algo": "em_mixed", "kind": "clustering", "function": "ml fit", "defaults": {"k": 2, "categorical": [], "max_iter": 200, "tol": 1e-6, "seed": 42}, "description": "EM clustering of mixed data: Gaussian for the numeric columns, categorical distributions for the columns listed in categorical; transform = membership probabilities, BIC"}, {"algo": "pca", "kind": "transformer", "function": "ml fit", "defaults": {"n_components": "all", "standardize": false}, "description": "Principal component analysis (n_components: integer or fraction of variance, ex: 0.95)"}, {"algo": "scaler_minmax", "kind": "transformer", "function": "ml fit", "defaults": {"min": 0, "max": 1}, "description": "Scale each column to [min, max]"}, {"algo": "scaler_zscore", "kind": "transformer", "function": "ml fit", "defaults": {}, "description": "Standardize each column: (x - mean) \/ sample standard deviation"}, {"algo": "scaler_robust", "kind": "transformer", "function": "ml fit", "defaults": {}, "description": "Robust scaling: (x - median) \/ IQR"}, {"algo": "scaler_maxabs", "kind": "transformer", "function": "ml fit", "defaults": {}, "description": "Scale each column by its maximum absolute value"}, {"algo": "isolation_forest", "kind": "anomaly", "function": "ml fit", "defaults": {"n_trees": 100, "sample_size": 256, "contamination": "auto", "seed": 42}, "description": "Isolation forest: anomaly score in [0, 1], 1 = anomaly in predict"}, {"algo": "one_class_svm", "kind": "anomaly", "function": "ml fit", "defaults": {"nu": 0.5, "kernel": "rbf", "gamma": "scale", "scale": true}, "description": "One-class SVM novelty detection"}]

ml train <id> <algo> <json_X> <json_y>

Description

Train a supervised model (classifier or regressor) with the default parameters and keep it in memory.
Returns the model information: classes, training score and the model specific results (coefficients, p-values, importance...).

Parameters

id: The model id - String - required
algo: The algorithm (see ml algos) - String - required
json_X: The rows (JSON array of arrays of numbers) - String - required
json_y: The labels or the target values - String - required

admin
ml train "lin" "linear" "[[1, 1], [2, 1], [3, 2], [4, 3], [5, 5], [6, 5], [7, 8], [8, 8]]" "[6.1, 8.0, 11.2, 14.1, 18.3, 19.9, 24.8, 26.1]"

mentdb
{"id": "lin", "algo": "linear", "kind": "regressor", "n_samples": 8, "n_features": 2, "params": {}, "train_r2": 0.998640910735476, "intercept": 3.398113207547425, "coef": [1.9028301886791894, 0.9943396226415189], "coefficients": [{"name": "intercept", "estimate": 3.398113207547425, "std_error": 0.30864164553951284, "t": 11.00989855600155, "p_value": 1.0758381408000759E-4, "ci_low": 2.6047245998026414, "ci_high": 4.1915018152922086}, {"name": "x0", "estimate": 1.9028301886791894, "std_error": 0.21292726609787951, "t": 8.936526653211649, "p_value": 2.9234207021793917E-4, "ci_low": 1.3554832261362733, "ci_high": 2.4501771512221056}, {"name": "x1", "estimate": 0.9943396226415189, "std_error": 0.18297645880949256, "t": 5.434248914374191, "p_value": 0.0028620361043391294, "ci_low": 0.523983661276775, "ci_high": 1.4646955840062628}], "r2": 0.998640910735476, "adj_r2": 0.9980972750296664, "rmse": 0.25693219133342576, "sigma": 0.32499637153272, "f_statistic": 1836.9671087888566, "f_p_value": 6.809574171207089E-8, "log_likelihood": -0.47996365881899195, "aic": 8.959927317637984, "bic": 9.277693484357327, "durbin_watson": 1.2160256021354434, "df_residual": 5}

Decision tree with string labels

admin
ml train "tree" "decision_tree_classifier" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "[\"low\", \"low\", \"low\", \"low\", \"mid\", \"mid\", \"mid\", \"mid\", \"high\", \"high\", \"high\", \"high\"]"

mentdb
{"id": "tree", "algo": "decision_tree_classifier", "kind": "classifier", "n_samples": 12, "n_features": 2, "params": {}, "classes": ["high", "low", "mid"], "train_accuracy": 1, "depth": 2, "leaves": 3, "nodes": 5}

Gaussian naive Bayes

admin
ml train "nb" "naive_bayes" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "[0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2]"

mentdb
{"id": "nb", "algo": "naive_bayes", "kind": "classifier", "n_samples": 12, "n_features": 2, "params": {}, "classes": [0, 1, 2], "train_accuracy": 1, "priors": [0.3333333333333333, 0.3333333333333333, 0.3333333333333333], "means": [[1.125, 1.975], [5.675, 8.025], [8.875, 1.125]], "variances": [[0.06687501022854166, 0.02187501022854169], [0.21687501022854172, 0.12687501022854167], [0.06687501022854152, 0.06687501022854166]]}

ml train <id> <algo> <json_X> <json_y> <json_params>

Description

Train a supervised model with parameters (JSON object, the missing parameters keep their default values).
Algorithms: linear, ridge, lasso, elastic_net, huber, ransac, wls, isotonic, poisson, logistic, liblinear_logistic, naive_bayes, lda, qda, knn_classifier, knn_regressor, decision_tree_classifier, decision_tree_regressor, random_forest_classifier, random_forest_regressor, extra_trees_classifier, extra_trees_regressor, gradient_boosting_classifier, gradient_boosting_regressor, svm_classifier, svm_regressor, linear_svm, linear_svr, mlp_classifier, mlp_regressor, one_r, ripper, m5, gaussian_process, bayes_net, cost_sensitive_classifier, quantile_regression, pipeline_classifier, pipeline_regressor, calibrated_classifier.
Meta-learning on any of them: bagging_classifier, bagging_regressor, adaboost_classifier, adaboost_regressor, stacking_classifier, stacking_regressor, voting_classifier, voting_regressor (an estimator = an algorithm name or {"algo": ..., "params": {...}}).

Parameters

id: The model id - String - required
algo: The algorithm (see ml algos) - String - required
json_X: The rows - String - required
json_y: The labels or the target values - String - required
json_params: The parameters, ex: {"k": 3} - String - required

admin
ml train "lr" "logistic" "[[1, 5], [2, 6], [3, 5], [4, 7], [5, 6], [6, 8], [7, 7], [8, 8], [2, 8], [6, 5], [3, 7], [7, 5]]" "[0, 0, 1, 0, 0, 1, 1, 1, 1, 0, 0, 1]" "{}"

mentdb
{"id": "lr", "algo": "logistic", "kind": "classifier", "n_samples": 12, "n_features": 2, "params": {}, "classes": [0, 1], "train_accuracy": 0.75, "positive_class": "1", "coefficients": [{"name": "intercept", "estimate": -6.218136168495791, "std_error": 4.617212423942465, "z": -1.346729497705535, "p_value": 0.17806736840192083, "ci_low": -15.267706228393905, "ci_high": 2.831433891402323}, {"name": "x0", "estimate": 0.4795158249919381, "std_error": 0.3569246786270205, "z": 1.3434650325566955, "p_value": 0.1791214413054445, "ci_low": -0.22004369031055526, "ci_high": 1.1790753402944314}, {"name": "x1", "estimate": 0.6358343543902452, "std_error": 0.6172951160487594, "z": 1.0300330228759194, "p_value": 0.3029945039933202, "ci_low": -0.5740418408977963, "ci_high": 1.8457105496782866}], "odds_ratios": [1.6152921287500903, 1.888597243672302], "log_likelihood": -6.43133138284943, "mcfadden_r2": 0.22679584230413063, "aic": 18.862662765698857, "bic": 20.317382715062863, "iterations": 5}

Random forest of 50 trees, out-of-bag accuracy

admin
ml train "rf" "random_forest_classifier" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "[0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2]" "{\"n_trees\": 50, \"seed\": 1}"

mentdb
{"id": "rf", "algo": "random_forest_classifier", "kind": "classifier", "n_samples": 12, "n_features": 2, "params": {"seed":1,"n_trees":50}, "classes": [0, 1, 2], "train_accuracy": 1, "n_trees": 50, "oob_accuracy": 1}

Poisson regression of counts

admin
ml train "poi" "poisson" "[[1, 1], [2, 1], [3, 2], [4, 3], [5, 5], [6, 5], [7, 8], [8, 8]]" "[1, 0, 2, 3, 5, 4, 8, 11]" "{}"

mentdb
{"id": "poi", "algo": "poisson", "kind": "regressor", "n_samples": 8, "n_features": 2, "params": {}, "train_r2": 0.9531490862599854, "coefficients": [{"name": "intercept", "estimate": -0.43571097260094405, "std_error": 0.6818892931589502, "z": -0.6389761167570917, "p_value": 0.5228384690230521, "ci_low": -1.7721894286359612, "ci_high": 0.9007674834340731}, {"name": "x0", "estimate": 0.21499415748072437, "std_error": 0.3318684726954716, "z": 0.6478294118586155, "p_value": 0.5170952900075826, "ci_low": -0.43545609660671425, "ci_high": 0.865444411568163}, {"name": "x1", "estimate": 0.1373246804668291, "std_error": 0.26727890066955234, "z": 0.513787957533577, "p_value": 0.6074002483263744, "ci_low": -0.386532338672952, "ci_high": 0.6611816996066102}], "rate_ratios": [1.239854653070191, 1.1472005616733991], "deviance": 3.0935653440320134, "null_deviance": 24.183266296319964, "pseudo_r2": 0.8720782665945017, "log_likelihood": -12.817267075558089, "aic": 31.634534151116178, "bic": 31.872858776155685, "dispersion": 0.4017473787791058, "iterations": 5}

Lasso: variable selection

admin
ml train "las" "lasso" "[[1, 1], [2, 1], [3, 2], [4, 3], [5, 5], [6, 5], [7, 8], [8, 8]]" "[6.1, 8.0, 11.2, 14.1, 18.3, 19.9, 24.8, 26.1]" "{\"alpha\": 0.5}"

mentdb
{"id": "las", "algo": "lasso", "kind": "regressor", "n_samples": 8, "n_features": 2, "params": {"alpha":0.5}, "train_r2": 0.9934205340911614, "intercept": 4.288383592733339, "coef": [1.7921718623612617, 0.8992346731250868], "r2": 0.9934205340911614, "non_zero": 2, "iterations": 208}

k nearest neighbors, k = 3, weighted by the distance

admin
ml train "knn" "knn_classifier" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "[0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2]" "{\"k\": 3, \"weights\": \"distance\"}"

mentdb
{"id": "knn", "algo": "knn_classifier", "kind": "classifier", "n_samples": 12, "n_features": 2, "params": {"k":3,"weights":"distance"}, "classes": [0, 1, 2], "train_accuracy": 1, "k": 3, "metric": "euclidean", "weights": "distance", "scale": false}

Support vector machine, linear kernel

admin
ml train "svm" "svm_classifier" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "[0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2]" "{\"kernel\": \"linear\", \"probability\": false}"

mentdb
{"id": "svm", "algo": "svm_classifier", "kind": "classifier", "n_samples": 12, "n_features": 2, "params": {"kernel":"linear","probability":false}, "classes": [0, 1, 2], "train_accuracy": 1, "support_vectors": 4}

Gradient boosting regressor

admin
ml train "gb" "gradient_boosting_regressor" "[[1, 1], [2, 1], [3, 2], [4, 3], [5, 5], [6, 5], [7, 8], [8, 8]]" "[6.1, 8.0, 11.2, 14.1, 18.3, 19.9, 24.8, 26.1]" "{\"n_estimators\": 50, \"max_depth\": 2}"

mentdb
{"id": "gb", "algo": "gradient_boosting_regressor", "kind": "regressor", "n_samples": 8, "n_features": 2, "params": {"max_depth":2,"n_estimators":50}, "train_r2": 0.9999429024591229, "n_estimators": 50, "learning_rate": 0.1, "train_loss_first": 39.53680468750001, "train_loss_last": 0.0027733613827626365}

Multilayer perceptron

admin
ml train "mlp" "mlp_classifier" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "[0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2]" "{\"hidden\": [8], \"seed\": 1}"

mentdb
{"id": "mlp", "algo": "mlp_classifier", "kind": "classifier", "n_samples": 12, "n_features": 2, "params": {"hidden":[8],"seed":1}, "classes": [0, 1, 2], "train_accuracy": 1, "layers": [2, 8, 3], "activation": "relu", "epochs": 500, "loss": 8.300241470744723E-4}

Huber robust regression: the outlier 40 gets a small weight

admin
ml train "hub" "huber" "[[1], [2], [3], [4], [5], [6], [7], [8]]" "[2.1, 3.9, 6.2, 8.1, 9.8, 12.2, 40, 15.9]" "{}"

mentdb
{"id": "hub", "algo": "huber", "kind": "regressor", "n_samples": 8, "n_features": 1, "params": {}, "train_r2": 0.3393020074348725, "intercept": 0.03405084631088329, "coef": [2.008728814037127], "scale": 0.21591965275445937, "outliers": 1, "weights": [1, 1, 1, 1, 1, 1, 0.011252393273285454, 1], "iterations": 12}

RANSAC: the outlier is out of the inliers

admin
ml train "ran" "ransac" "[[1], [2], [3], [4], [5], [6], [7], [8]]" "[2.1, 3.9, 6.2, 8.1, 9.8, 12.2, 40, 15.9]" "{\"seed\": 1}"

mentdb
{"id": "ran", "algo": "ransac", "kind": "regressor", "n_samples": 8, "n_features": 1, "params": {"seed":1}, "train_r2": 0.3337168464253085, "intercept": 0.09139344262294458, "coef": [1.9848360655737718], "residual_threshold": 4.1499999999999995, "n_inliers": 7, "inliers": [1, 1, 1, 1, 1, 1, 0, 1]}

Weighted least squares (weights = 1 / variance of each measure)

admin
ml train "wls" "wls" "[[1], [2], [3], [4], [5]]" "[2.2, 3.9, 6.1, 8.3, 9.7]" "{\"weights\": [1, 2, 1, 2, 1]}"

mentdb
{"id": "wls", "algo": "wls", "kind": "regressor", "n_samples": 5, "n_features": 1, "params": {"weights":[1,2,1,2,1]}, "train_r2": 0.9942840144445739, "intercept": 0.1071428571428612, "coef": [1.9833333333333307], "coefficients": [{"name": "intercept", "estimate": 0.1071428571428612, "std_error": 0.30560708697603484, "t": 0.35059022420924146, "p_value": 0.7490688862847927, "ci_low": -0.8654352875422783, "ci_high": 1.0797210018280006}, {"name": "x0", "estimate": 1.9833333333333307, "std_error": 0.09336450726550884, "t": 21.24290473352097, "p_value": 2.282313012639175E-4, "ci_low": 1.686205802181373, "ci_high": 2.2804608644852884}], "weighted_r2": 0.9933958671557132, "sigma": 0.3234241404149898}

Isotonic regression: a dose-response that can only increase

admin
ml train "iso" "isotonic" "[[1], [2], [3], [4], [5], [6]]" "[1, 3, 2, 4, 3.5, 5]" "{}"

mentdb
{"id": "iso", "algo": "isotonic", "kind": "regressor", "n_samples": 6, "n_features": 1, "params": {}, "train_r2": 0.9387755102040817, "x": [1, 2, 3, 4, 5, 6], "fitted": [1, 2.5, 2.5, 3.75, 3.75, 5]}

Sparse logistic regression (liblinear, L1 penalty): some coefficients are exactly 0

admin
ml train "l1" "liblinear_logistic" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "[0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2]" "{\"penalty\": \"l1\"}"

mentdb
{"id": "l1", "algo": "liblinear_logistic", "kind": "classifier", "n_samples": 12, "n_features": 2, "params": {"penalty":"l1"}, "classes": [0, 1, 2], "train_accuracy": 1, "solver": "L1R_LR", "coef": [[-0.6046643542974789, -0.17067637207652772], [0, 0.7637938652724102], [0.5026806018134624, -0.32857901666305933]], "non_zero": 5}

Linear SVM (liblinear)

admin
ml train "lsvm" "linear_svm" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "[0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2]" "{\"C\": 1}"

mentdb
{"id": "lsvm", "algo": "linear_svm", "kind": "classifier", "n_samples": 12, "n_features": 2, "params": {"C":1}, "classes": [0, 1, 2], "train_accuracy": 1, "solver": "L2R_L2LOSS_SVC_DUAL", "coef": [[-0.2631636089281483, -0.1377868762759679], [0.03437428428150138, 0.29954708238910843], [0.23324530865192009, -0.17995374594713365]], "non_zero": 6}

Linear support vector regression (liblinear)

admin
ml train "lsvr" "linear_svr" "[[1, 1], [2, 1], [3, 2], [4, 3], [5, 5], [6, 5], [7, 8], [8, 8]]" "[6.1, 8.0, 11.2, 14.1, 18.3, 19.9, 24.8, 26.1]" "{\"C\": 100, \"epsilon\": 0.01}"

mentdb
{"id": "lsvr", "algo": "linear_svr", "kind": "regressor", "n_samples": 8, "n_features": 2, "params": {"epsilon":0.01,"C":100}, "train_r2": 0.9986394655160965, "solver": "L2R_L2LOSS_SVR_DUAL", "coef": [1.889891667496855, 1.003682953870743], "non_zero": 2}

nu-SVM: nu bounds the fraction of support vectors

admin
ml train "nusvm" "svm_classifier" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "[0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2]" "{\"type\": \"nu\", \"nu\": 0.3}"

mentdb
{"id": "nusvm", "algo": "svm_classifier", "kind": "classifier", "n_samples": 12, "n_features": 2, "params": {"nu":0.3,"type":"nu"}, "classes": [0, 1, 2], "train_accuracy": 1, "support_vectors": 8}

C and gamma chosen by cross validation

admin
ml train "auto" "svm_classifier" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "[0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2]" "{\"C\": \"auto\"}"

mentdb
{"id": "auto", "algo": "svm_classifier", "kind": "classifier", "n_samples": 12, "n_features": 2, "params": {"C":"auto"}, "classes": [0, 1, 2], "train_accuracy": 1, "support_vectors": 12, "auto": {"C": 0.01, "gamma": 0.05, "cv_accuracy": 1}}

Crammer-Singer multiclass linear SVM

admin
ml train "cs" "linear_svm" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "[0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2]" "{\"multiclass\": \"crammer_singer\"}"

mentdb
{"id": "cs", "algo": "linear_svm", "kind": "classifier", "n_samples": 12, "n_features": 2, "params": {"multiclass":"crammer_singer"}, "classes": [0, 1, 2], "train_accuracy": 1, "solver": "MCSVM_CS", "coef": [[-0.1494330511466499, -0.07459847128757915], [0.010281329851293455, 0.17177552522258244], [0.13915172129535644, -0.09717705393500306]], "non_zero": 6}

Bagging of 10 k nearest neighbors models, out-of-bag accuracy

admin
ml train "bag" "bagging_classifier" "[[1, 5], [2, 6], [3, 5], [4, 7], [5, 6], [6, 8], [7, 7], [8, 8], [2, 8], [6, 5], [3, 7], [7, 5]]" "[0, 0, 1, 0, 0, 1, 1, 1, 1, 0, 0, 1]" "{\"base\": {\"algo\": \"knn_classifier\", \"params\": {\"k\": 3}}, \"n_estimators\": 10}"

mentdb
{"id": "bag", "algo": "bagging_classifier", "kind": "classifier", "n_samples": 12, "n_features": 2, "params": {"n_estimators":10,"base":{"params":{"k":3},"algo":"knn_classifier"}}, "classes": [0, 1], "train_accuracy": 0.9166666666666666, "method": "bagging", "n_estimators": 10, "base": "knn_classifier", "oob_accuracy": 0.75}

AdaBoost of decision stumps (SAMME)

admin
ml train "ada" "adaboost_classifier" "[[1, 5], [2, 6], [3, 5], [4, 7], [5, 6], [6, 8], [7, 7], [8, 8], [2, 8], [6, 5], [3, 7], [7, 5]]" "[0, 0, 1, 0, 0, 1, 1, 1, 1, 0, 0, 1]" "{\"n_estimators\": 20}"

mentdb
{"id": "ada", "algo": "adaboost_classifier", "kind": "classifier", "n_samples": 12, "n_features": 2, "params": {"n_estimators":20}, "classes": [0, 1], "train_accuracy": 0.9166666666666666, "method": "adaboost", "n_estimators": 7, "estimator_weights": [1.0986122886681096, 0.9555114450274362, 1.203972804325936, 0.6410908186038923, 0.3315811388877808, 0.568963888449741, 0.19405414471959614], "base": "decision_tree_classifier"}

Stacking: naive Bayes and k nearest neighbors, logistic regression on their out-of-fold probabilities

admin
ml train "stk" "stacking_classifier" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "[0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2]" "{\"estimators\": [\"naive_bayes\", {\"algo\": \"knn_classifier\", \"params\": {\"k\": 3}}], \"k\": 3}"

mentdb
{"id": "stk", "algo": "stacking_classifier", "kind": "classifier", "n_samples": 12, "n_features": 2, "params": {"estimators":["naive_bayes",{"params":{"k":3},"algo":"knn_classifier"}],"k":3}, "classes": [0, 1, 2], "train_accuracy": 1, "method": "stacking", "n_estimators": 2, "final_estimator": "logistic"}

Soft vote of three classifiers

admin
ml train "vote" "voting_classifier" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "[\"low\", \"low\", \"low\", \"low\", \"mid\", \"mid\", \"mid\", \"mid\", \"high\", \"high\", \"high\", \"high\"]" "{\"estimators\": [\"naive_bayes\", \"lda\", \"decision_tree_classifier\"]}"

mentdb
{"id": "vote", "algo": "voting_classifier", "kind": "classifier", "n_samples": 12, "n_features": 2, "params": {"estimators":["naive_bayes","lda","decision_tree_classifier"]}, "classes": ["high", "low", "mid"], "train_accuracy": 1, "method": "voting", "n_estimators": 3, "voting": "soft"}

Vote of two regressors

admin
ml train "vr" "voting_regressor" "[[1, 1], [2, 1], [3, 2], [4, 3], [5, 5], [6, 5], [7, 8], [8, 8]]" "[6.1, 8.0, 11.2, 14.1, 18.3, 19.9, 24.8, 26.1]" "{\"estimators\": [\"linear\", \"huber\"]}"

mentdb
{"id": "vr", "algo": "voting_regressor", "kind": "regressor", "n_samples": 8, "n_features": 2, "params": {"estimators":["linear","huber"]}, "train_r2": 0.9986031328029731, "method": "voting", "n_estimators": 2, "voting": "soft"}

OneR: the best single feature

admin
ml train "one" "one_r" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "[\"low\", \"low\", \"low\", \"low\", \"mid\", \"mid\", \"mid\", \"mid\", \"high\", \"high\", \"high\", \"high\"]" "{\"min_bucket\": 3}"

mentdb
{"id": "one", "algo": "one_r", "kind": "classifier", "n_samples": 12, "n_features": 2, "params": {"min_bucket":3}, "classes": ["high", "low", "mid"], "train_accuracy": 1, "rules": 3, "feature": 0, "training_accuracy": 1}

RIPPER: ordered if-then rules

admin
ml train "rip" "ripper" "[[2, 1], [3, 2], [6, 1], [7, 2], [8, 1], [6, 6], [7, 8], [2, 7], [3, 5], [1, 2], [9, 2], [8, 7], [4, 4], [6.5, 2.5], [2.5, 8]]" "[\"no\", \"no\", \"yes\", \"yes\", \"yes\", \"no\", \"no\", \"no\", \"no\", \"no\", \"yes\", \"no\", \"no\", \"yes\", \"no\"]" "{}"

mentdb
{"id": "rip", "algo": "ripper", "kind": "classifier", "n_samples": 15, "n_features": 2, "params": {}, "classes": ["no", "yes"], "train_accuracy": 1, "rules": 2, "default_class": "no", "training_accuracy": 1}

M5 model tree: a linear model in each leaf

admin
ml train "m5" "m5" "[[1], [2], [3], [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24]]" "[1.5, 3.5, 6.3, 7.1, 9.9, 11.1, 14.4, 16.1, 18.5, 19.8, 22.2, 23.9, 22.7, 22.1, 20.5, 19.9, 19.3, 18.0, 16.8, 16.9, 15.0, 13.8, 13.1, 11.8]" "{}"

mentdb
{"id": "m5", "algo": "m5", "kind": "regressor", "n_samples": 24, "n_features": 1, "params": {}, "train_r2": 0.9954985725202457, "leaves": 5, "smoothed": false}

Gaussian process: the length scale, the signal variance and the noise by maximum likelihood

admin
ml train "gp" "gaussian_process" "[[0], [0.5], [1], [1.5], [2], [2.5], [3], [3.5], [4], [5], [6]]" "[0.02, 0.5, 0.83, 1.01, 0.9, 0.62, 0.13, -0.33, -0.78, -0.95, -0.3]" "{}"

mentdb
{"id": "gp", "algo": "gaussian_process", "kind": "regressor", "n_samples": 11, "n_features": 1, "params": {}, "train_r2": 0.9995909376674038, "kernel": "rbf", "length_scale": 1.1456293313027177, "signal_variance": 3.6036512533026195, "noise": 8.606243576114696E-4, "log_marginal_likelihood": 6.644204256653743}

Bayesian network, tree augmented naive Bayes (TAN) on features discretized by MDL

admin
ml train "bn" "bayes_net" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "[\"low\", \"low\", \"low\", \"low\", \"mid\", \"mid\", \"mid\", \"mid\", \"high\", \"high\", \"high\", \"high\"]" "{}"

mentdb
{"id": "bn", "algo": "bayes_net", "kind": "classifier", "n_samples": 12, "n_features": 2, "params": {}, "classes": ["high", "low", "mid"], "train_accuracy": 1, "structure": "tan", "network": [{"feature": "x0", "parents": ["class"], "cuts": [3.25, 7.35]}, {"feature": "x1", "parents": ["class", "x0"], "cuts": [1.65, 4.85]}], "log_score": -13.183347464017316}

Bayesian network searched by K2

admin
ml train "k2" "bayes_net" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "[\"low\", \"low\", \"low\", \"low\", \"mid\", \"mid\", \"mid\", \"mid\", \"high\", \"high\", \"high\", \"high\"]" "{\"structure\": \"k2\", \"discretize\": \"frequency\", \"bins\": 3}"

mentdb
{"id": "k2", "algo": "bayes_net", "kind": "classifier", "n_samples": 12, "n_features": 2, "params": {"bins":3,"discretize":"frequency","structure":"k2"}, "classes": ["high", "low", "mid"], "train_accuracy": 1, "structure": "k2", "network": [{"feature": "x0", "parents": ["class"], "cuts": [3.25, 7.35]}, {"feature": "x1", "parents": ["class"], "cuts": [1.65, 4.85]}], "log_score": -13.183347464017316}

Cost-sensitive: missing a 1 costs 5 times more than a false alarm

admin
ml train "cs5" "cost_sensitive_classifier" "[[1, 5], [2, 6], [3, 5], [4, 7], [5, 6], [6, 8], [7, 7], [8, 8], [2, 8], [6, 5], [3, 7], [7, 5]]" "[0, 0, 1, 0, 0, 1, 1, 1, 1, 0, 0, 1]" "{\"base\": \"logistic\", \"cost\": [[0, 1], [5, 0]]}"

mentdb
{"id": "cs5", "algo": "cost_sensitive_classifier", "kind": "classifier", "n_samples": 12, "n_features": 2, "params": {"cost":[[0,1],[5,0]],"base":"logistic"}, "classes": [0, 1], "train_accuracy": 0.5833333333333334, "method": "minimize_cost", "base": "logistic", "cost": [[0, 1], [5, 0]]}

Quantile regression: the 90th percentile

admin
ml train "q90" "quantile_regression" "[[1], [2], [3], [4], [5], [6], [7], [8]]" "[2.1, 3.9, 6.2, 8.1, 9.8, 12.2, 40, 15.9]" "{\"quantile\": 0.9}"

mentdb
{"id": "q90", "algo": "quantile_regression", "kind": "regressor", "n_samples": 8, "n_features": 1, "params": {"quantile":0.9}, "train_r2": -0.9254942234459915, "quantile": 0.9, "intercept": -4.216666901234568, "coefficients": [6.31666664197531]}

Pipeline: standardization, PCA, then k nearest neighbors, in one model

admin
ml train "pipe" "pipeline_classifier" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "[\"low\", \"low\", \"low\", \"low\", \"mid\", \"mid\", \"mid\", \"mid\", \"high\", \"high\", \"high\", \"high\"]" "{\"steps\": [\"scaler_zscore\", {\"algo\": \"pca\", \"params\": {\"n_components\": 1}}], \"final\": {\"algo\": \"knn_classifier\", \"params\": {\"k\": 3}}}"

mentdb
{"id": "pipe", "algo": "pipeline_classifier", "kind": "classifier", "n_samples": 12, "n_features": 2, "params": {"final":{"params":{"k":3},"algo":"knn_classifier"},"steps":["scaler_zscore",{"params":{"n_components":1},"algo":"pca"}]}, "classes": ["high", "low", "mid"], "train_accuracy": 1, "steps": ["scaler_zscore", "pca", "knn_classifier"], "final": {"k": 3, "metric": "euclidean", "weights": "uniform", "scale": false}}

Calibrated probabilities of a naive Bayes (isotonic)

admin
ml train "cal" "calibrated_classifier" "[[1, 5], [2, 6], [3, 5], [4, 7], [5, 6], [6, 8], [7, 7], [8, 8], [2, 8], [6, 5], [3, 7], [7, 5]]" "[0, 0, 1, 0, 0, 1, 1, 1, 1, 0, 0, 1]" "{\"method\": \"isotonic\", \"k\": 3}"

mentdb
{"id": "cal", "algo": "calibrated_classifier", "kind": "classifier", "n_samples": 12, "n_features": 2, "params": {"method":"isotonic","k":3}, "classes": [0, 1], "train_accuracy": 0.5833333333333334, "method": "isotonic", "base": "naive_bayes"}

Used in use cases (21 recipes)

Process & reactor engineer › Yield regimes with a model tree
Biologist & microbiologist › Monotonic dose-response curve
Epidemiologist › Odds ratios of a case-control study
Geneticist & bioinformatician › Sparse gene signature with an L1 logistic regression
Climate & environment scientist › Temperature between sparse stations
IoT & sensor engineer › Route the maintenance tickets
and 15 more in the Use cases

Often used with

ml model delete · ml model predict · ml cross_validate · ml model predict_interval · ml model predict_proba · ml model rules · file delete · math abs

ml fit <id> <algo> <json_X>

Description

Fit an unsupervised model with the default parameters (clustering, transformer, anomaly detector) and keep it in memory

Parameters

id: The model id - String - required
algo: The algorithm: kmeans, fuzzy_cmeans, gmm, em_mixed, som, pca, ica, nmf, kernel_pca, scaler_minmax, scaler_zscore, scaler_robust, scaler_maxabs, isolation_forest, one_class_svm - String - required
json_X: The rows - String - required

admin
ml fit "sc" "scaler_zscore" "[[1, 1], [2, 1], [3, 2], [4, 3], [5, 5], [6, 5], [7, 8], [8, 8]]"

mentdb
{"id": "sc", "algo": "scaler_zscore", "kind": "transformer", "n_samples": 8, "n_features": 2, "params": {}, "center": [4.5, 4.125], "scale": [2.449489742783178, 2.850438562747845]}

Principal component analysis

admin
ml fit "pca" "pca" "[[1, 1], [2, 1], [3, 2], [4, 3], [5, 5], [6, 5], [7, 8], [8, 8]]"

mentdb
{"id": "pca", "algo": "pca", "kind": "transformer", "n_samples": 8, "n_features": 2, "params": {}, "components": [[0.6501176375186445, 0.7598335721637841], [0.7598335721637841, -0.6501176375186445]], "explained_variance": [13.930893161238439, 0.19410683876156246], "explained_variance_ratio": [0.9862579229195355, 0.013742077080464599], "cumulative_ratio": [0.9862579229195355, 1], "mean": [4.5, 4.125]}

ml fit <id> <algo> <json_X> <json_params>

Description

Fit an unsupervised model with parameters

Parameters

id: The model id - String - required
algo: The algorithm (see ml algos) - String - required
json_X: The rows - String - required
json_params: The parameters, ex: {"k": 3} - String - required

admin
ml fit "km" "kmeans" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "{\"k\": 3}"

mentdb
{"id": "km", "algo": "kmeans", "kind": "clustering", "n_samples": 12, "n_features": 2, "params": {"k":3}, "centers": [[1.125, 1.975], [8.875, 1.125], [5.675, 8.025]], "sizes": [4, 4, 4], "labels": [0, 0, 0, 0, 2, 2, 2, 2, 1, 1, 1, 1], "inertia": 2.265, "iterations": 2}

Gaussian mixture: weights, means, covariances, AIC, BIC

admin
ml fit "gmm" "gmm" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "{\"k\": 3}"

mentdb
{"id": "gmm", "algo": "gmm", "kind": "clustering", "n_samples": 12, "n_features": 2, "params": {"k":3}, "weights": [0.3333333333333333, 0.3333333333333333, 0.3333333333333333], "means": [[1.125, 1.975], [8.875, 1.125], [5.675, 8.025]], "covariances": [[[0.06687599999999999, -0.006874999999999988], [-0.006874999999999988, 0.021876000000000024]], [[0.06687599999999985, -0.06687499999999992], [-0.06687499999999992, 0.06687599999999999]], [[0.21687600000000004, 0.07562499999999996], [0.07562499999999996, 0.126876]]], "log_likelihood": 7.189060273446542, "aic": 19.621879453106914, "bic": 27.86529249950292, "iterations": 2}

PCA keeping 95% of the variance

admin
ml fit "pca" "pca" "[[1, 1], [2, 1], [3, 2], [4, 3], [5, 5], [6, 5], [7, 8], [8, 8]]" "{\"n_components\": 0.95, \"standardize\": true}"

mentdb
{"id": "pca", "algo": "pca", "kind": "transformer", "n_samples": 8, "n_features": 2, "params": {"n_components":0.95,"standardize":true}, "components": [[0.7071067811865475, 0.7071067811865478]], "explained_variance": [1.971870186196374], "explained_variance_ratio": [0.9859350930981868], "cumulative_ratio": [0.9859350930981868], "mean": [4.5, 4.125]}

Isolation forest

admin
ml fit "iso" "isolation_forest" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "{\"n_trees\": 50}"

mentdb
{"id": "iso", "algo": "isolation_forest", "kind": "anomaly", "n_samples": 12, "n_features": 2, "params": {"n_trees":50}, "n_trees": 50, "sample_size": 12, "threshold": 0.5}

Self-organizing map of 2 x 2 nodes

admin
ml fit "som" "som" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "{\"rows\": 2, \"cols\": 2}"

mentdb
{"id": "som", "algo": "som", "kind": "clustering", "n_samples": 12, "n_features": 2, "params": {"rows":2,"cols":2}, "rows": 2, "cols": 2, "hits": [4, 4, 0, 4], "quantization_error": 0.43191705328649355, "topographic_error": 0.3333333333333333, "weights": [[2.3068076274525606, 1.9952628614088286], [7.542378100764898, 2.0637808444632038], [3.8736967806436353, 4.653618706478728], [6.017605604484293, 6.917178252442712]]}

Independent components (FastICA)

admin
ml fit "ica" "ica" "[[1, 1], [2, 1], [3, 2], [4, 3], [5, 5], [6, 5], [7, 8], [8, 8]]" "{\"n_components\": 2}"

mentdb
{"id": "ica", "algo": "ica", "kind": "transformer", "n_samples": 8, "n_features": 2, "params": {"n_components":2}, "components": [[-0.07599918428766368, -0.3112583949361164], [-1.8515362238256674, 1.5617193307971864]]}

Non-negative factorization in 2 parts

admin
ml fit "nmf" "nmf" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "{\"n_components\": 2}"

mentdb
{"id": "nmf", "algo": "nmf", "kind": "transformer", "n_samples": 12, "n_features": 2, "params": {"n_components":2}, "components": [[0.4950101728290347, 1.255860317472518], [2.8865191891800994, 0.18345922849877425]], "reconstruction_error": 6.550700930819886E-13}

Kernel PCA (rbf)

admin
ml fit "kp" "kernel_pca" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "{\"n_components\": 2, \"gamma\": 0.1}"

mentdb
{"id": "kp", "algo": "kernel_pca", "kind": "transformer", "n_samples": 12, "n_features": 2, "params": {"n_components":2,"gamma":0.1}, "kernel": "rbf", "eigenvalues": [3.903505756413059, 3.7801644327528003]}

EM of mixed data: column 1 is a category

admin
ml fit "mix" "em_mixed" "[[1.2, 0], [1.0, 0], [0.8, 0], [1.1, 1], [5.1, 1], [4.9, 2], [5.3, 2], [5.0, 2]]" "{\"k\": 2, \"categorical\": [1]}"

mentdb
{"id": "mix", "algo": "em_mixed", "kind": "clustering", "n_samples": 8, "n_features": 2, "params": {"categorical":[1],"k":2}, "clusters": [{"weight": 0.5, "means": [5.075, 0], "categories": [{"column": 1, "mode": 2, "probability": 0.6363636363636364}]}, {"weight": 0.5, "means": [1.025, 0], "categories": [{"column": 1, "mode": 0, "probability": 0.6363636363636364}]}], "log_likelihood": -6.917519033881967, "bic": 32.550011942882456}

Used in use cases (7 recipes)

Chemist › Unmix absorption spectra
Electronics & signal engineer › Separate two mixed sources
IoT & sensor engineer › Detect abnormal machines with an isolation forest
Data scientist › From SQL rows to a trained model
Statistician & biostatistician › How many populations? Gaussian mixture and BIC
Machine learning engineer › Map the operating modes of a machine with a self-organizing map
and 1 more in the Use cases

Often used with

ml model delete · ml model transform · ml model predict · json doc · matrix transpose · ml elbow · ml prep impute · ml prep matrix

ml model predict <id> <json_X>

Description

Predict with a model: the labels (classifier), the values (regressor), the cluster index (clustering) or 1 for an anomaly and 0 for a normal row (anomaly detector)

Parameters

id: The model id - String - required
json_X: The rows - String - required

admin
ml train "nb" "naive_bayes" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "[0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2]";
ml model predict "nb" "[[1, 2], [6, 8], [9, 1]]"

mentdb
[0, 1, 2]

Regression

admin
ml train "lin" "linear" "[[1, 1], [2, 1], [3, 2], [4, 3], [5, 5], [6, 5], [7, 8], [8, 8]]" "[6.1, 8.0, 11.2, 14.1, 18.3, 19.9, 24.8, 26.1]";
ml model predict "lin" "[[9, 9], [10, 9]]"

mentdb
[29.4726415094338, 31.37547169811299]

Cluster of new points

admin
ml fit "km" "kmeans" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "{\"k\": 3}";
ml model predict "km" "[[1, 1], [9, 1]]"

mentdb
[0, 1]

Anomalies

admin
ml fit "iso" "isolation_forest" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "{\"contamination\": 0.1}";
ml model predict "iso" "[[5.5, 8], [30, 30]]"

mentdb
[0, 1]

Used in use cases (11 recipes)

Biologist & microbiologist › Monotonic dose-response curve
Epidemiologist › Detect a rare disease with SMOTE
IoT & sensor engineer › Detect abnormal machines with an isolation forest
Industrial quality engineer › Inspection driven by the cost of errors
Data scientist › From SQL rows to a trained model
Machine learning engineer › Tune a gradient boosting with a grid search
and 5 more in the Use cases

Often used with

ml model delete · ml train · ml fit · ml model transform · math abs · math floor · math max · math mod

ml model predict_proba <id> <json_X>

Description

Get the probability of each class (classifier)

Parameters

id: The model id - String - required
json_X: The rows - String - required

admin
ml train "lr" "logistic" "[[1, 5], [2, 6], [3, 5], [4, 7], [5, 6], [6, 8], [7, 7], [8, 8], [2, 8], [6, 5], [3, 7], [7, 5]]" "[0, 0, 1, 0, 0, 1, 1, 1, 1, 0, 0, 1]";
ml model predict_proba "lr" "[[4, 6], [7, 8]]"

mentdb
{"classes": [0, 1], "proba": [[0.618943596231428, 0.381056403768572], [0.09751468455906687, 0.9024853154409331]]}

Used in use cases (2 recipes)

Machine learning engineer › Save a trained model and score with it later
Deep learning engineer › Predictive maintenance with a neural network

Often used with

ml model delete · ml train · file delete · math max · ml cross_validate · ml model info · ml model load · ml model predict

ml model transform <id> <json_X>

Description

Transform the rows: scaled values (scaler), components (PCA), distance to each center (kmeans), membership degrees (fuzzy_cmeans), probability of each component (gmm), anomaly score (isolation_forest, one_class_svm)

Parameters

id: The model id - String - required
json_X: The rows - String - required

admin
ml fit "mm" "scaler_minmax" "[[1, 1], [2, 1], [3, 2], [4, 3], [5, 5], [6, 5], [7, 8], [8, 8]]";
ml model transform "mm" "[[4.5, 4.5]]"

mentdb
[[0.5, 0.5]]

Projection on the principal components

admin
ml fit "pca" "pca" "[[1, 1], [2, 1], [3, 2], [4, 3], [5, 5], [6, 5], [7, 8], [8, 8]]" "{\"n_components\": 1}";
ml model transform "pca" "[[1, 1], [8, 8]]"

mentdb
[[-4.649891644327081], [5.219766823449919]]

Used in use cases (6 recipes)

Chemist › Unmix absorption spectra
Electronics & signal engineer › Separate two mixed sources
IoT & sensor engineer › Detect abnormal machines with an isolation forest
Data scientist › From SQL rows to a trained model
Machine learning engineer › Map the operating modes of a machine with a self-organizing map
Data scientist › Customer segments on principal components

Often used with

ml fit · ml model delete · ml model predict · json doc · matrix transpose · ml elbow · ml prep impute · ml prep matrix

ml model inverse_transform <id> <json_X>

Description

Go back to the original values (scaler, PCA)

Parameters

id: The model id - String - required
json_X: The transformed rows - String - required

admin
ml fit "mm" "scaler_minmax" "[[1, 1], [2, 1], [3, 2], [4, 3], [5, 5], [6, 5], [7, 8], [8, 8]]";
ml model inverse_transform "mm" "[[0, 1], [0.5, 0.5]]"

mentdb
[[1, 8], [4.5, 4.5]]

ml model score <id> <json_X> <json_y>

Description

Get the accuracy (classifier) or the R2 (regressor) on a data set

Parameters

id: The model id - String - required
json_X: The rows - String - required
json_y: The true labels or values - String - required

admin
ml train "knn" "knn_classifier" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "[0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2]" "{\"k\": 3}";
ml model score "knn" "[[1, 2.5], [6, 7], [9, 2]]" "[0, 1, 1]"

mentdb
0.6666666666666666

Used in use cases (1 recipe)

Deep learning engineer › A neural network learns the XOR truth table

Often used with

math abs · math floor · math mod · ml model delete · ml model predict · ml train

ml model evaluate <id> <json_X> <json_y>

Description

Full evaluation on a test set: classification report (precision, recall, f1, kappa, MCC, confusion matrix) or regression metrics (MAE, MSE, RMSE, MAPE, R2...)

Parameters

id: The model id - String - required
json_X: The rows - String - required
json_y: The true labels or values - String - required

admin
ml train "lda" "lda" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "[0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2]";
ml model evaluate "lda" "[[1, 2.5], [6, 7], [9, 2], [5, 1]]" "[0, 1, 2, 2]"

mentdb
{"accuracy": 1, "classes": [{"label": 0, "precision": 1, "recall": 1, "f1": 1, "support": 1}, {"label": 1, "precision": 1, "recall": 1, "f1": 1, "support": 1}, {"label": 2, "precision": 1, "recall": 1, "f1": 1, "support": 2}], "macro": {"precision": 1, "recall": 1, "f1": 1}, "weighted": {"precision": 1, "recall": 1, "f1": 1}, "kappa": 1, "mcc": 1, "confusion": [[1, 0, 0], [0, 1, 0], [0, 0, 2]], "labels": [0, 1, 2]}

Used in use cases (1 recipe)

Machine learning engineer › Random forest with a hold-out test set

Often used with

ml model delete · ml model importance · ml prep split_stratified · ml train

ml model importance <id>

Description

Get the importance of each feature and the ranking (trees, forests, boosting, linear models, PCA)

Parameters

id: The model id - String - required

admin
ml train "rf" "random_forest_classifier" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "[0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2]" "{\"n_trees\": 50}";
ml model importance "rf"

mentdb
{"importance": [0.5198131210107259, 0.48018687898927415], "ranking": [0, 1]}

Used in use cases (1 recipe)

Machine learning engineer › Random forest with a hold-out test set

Often used with

ml model delete · ml model evaluate · ml prep split_stratified · ml train

ml model predict_interval <id> <json_X> <level>

Description

Prediction interval at a level: exact Student interval of the linear regression (OLS), Gaussian interval of gaussian_process (wider far from the data), Laplace interval of svm_regressor trained with {"probability": true}

Parameters

id: The model id - String - required
json_X: The rows - String - required
level: The level, ex: 0.95 - Number - required

admin
ml train "lin" "linear" "[[1, 1], [2, 1], [3, 2], [4, 3], [5, 5], [6, 5], [7, 8], [8, 8]]" "[6.1, 8.0, 11.2, 14.1, 18.3, 19.9, 24.8, 26.1]";
ml model predict_interval "lin" "[[9, 9], [4.5, 4]]" 0.95

mentdb
[{"prediction": 29.4726415094338, "lower": 28.408755856197086, "upper": 30.536527162670513}, {"prediction": 15.938207547169853, "lower": 15.050152056481128, "upper": 16.826263037858578}]

Support vector regression

admin
ml train "svr" "svm_regressor" "[[1, 1], [2, 1], [3, 2], [4, 3], [5, 5], [6, 5], [7, 8], [8, 8]]" "[6.1, 8.0, 11.2, 14.1, 18.3, 19.9, 24.8, 26.1]" "{\"C\": 10, \"probability\": true}";
ml model predict_interval "svr" "[[9, 9]]" 0.95

mentdb
[{"prediction": 23.995092183609525, "lower": 17.179107519201267, "upper": 30.811076848017784}]

Gaussian process: the interval grows outside the data (x = 9)

admin
ml train "gp" "gaussian_process" "[[0], [0.5], [1], [1.5], [2], [2.5], [3], [3.5], [4], [5], [6]]" "[0.02, 0.5, 0.83, 1.01, 0.9, 0.62, 0.13, -0.33, -0.78, -0.95, -0.3]";
ml model predict_interval "gp" "[[1.5], [4.5], [9]]" 0.95

mentdb
[{"prediction": 0.9970255969807764, "lower": 0.9508531931424334, "upper": 1.0431980008191195}, {"prediction": -0.9795884847459152, "lower": -1.0300936413407031, "upper": -0.9290833281511273}, {"prediction": 0.8174345280669167, "lower": -1.0035827482883013, "upper": 2.638451804422135}]

Used in use cases (2 recipes)

Climate & environment scientist › Temperature between sparse stations
Statistician & biostatistician › Prediction interval of a calibration

Often used with

ml model delete · ml train

ml model permutation_importance <id> <json_X> <json_y> <repeats> <seed>

Description

Importance of each feature for any supervised model: drop of the score (accuracy or R2) when the column is shuffled, mean and standard deviation over the repeats, ranking

Parameters

id: The model id - String - required
json_X: The rows (a test set is better) - String - required
json_y: The true labels or values - String - required
repeats: The number of shuffles per column - Number - required
seed: The seed - Number - required

admin
ml train "rf" "random_forest_classifier" "[[1, 5], [2, 6], [3, 5], [4, 7], [5, 6], [6, 8], [7, 7], [8, 8], [2, 8], [6, 5], [3, 7], [7, 5]]" "[0, 0, 1, 0, 0, 1, 1, 1, 1, 0, 0, 1]" "{\"seed\": 1}";
ml model permutation_importance "rf" "[[1, 5], [2, 6], [3, 5], [4, 7], [5, 6], [6, 8], [7, 7], [8, 8], [2, 8], [6, 5], [3, 7], [7, 5]]" "[0, 0, 1, 0, 0, 1, 1, 1, 1, 0, 0, 1]" 10 1

mentdb
{"metric": "accuracy", "base_score": 1, "importance": [0.30000000000000004, 0.325], "std": [0.09781564923143896, 0.1269903272295829], "ranking": [1, 0]}

Used in use cases (1 recipe)

Data scientist › Explain a model and one of its predictions

Often used with

ml model delete · ml model explain · ml model partial_dependence · ml train

ml model partial_dependence <id> <json_X> <feature> <points>

Description

Partial dependence: average prediction (regressor) or average probability of each class (classifier) when one column takes the values of a grid between its 5th and 95th percentiles

Parameters

id: The model id - String - required
json_X: The rows - String - required
feature: The column index - Number - required
points: The number of points of the grid (2 to 200) - Number - required

admin
ml train "gb" "gradient_boosting_regressor" "[[1, 1], [2, 1], [3, 2], [4, 3], [5, 5], [6, 5], [7, 8], [8, 8]]" "[6.1, 8.0, 11.2, 14.1, 18.3, 19.9, 24.8, 26.1]" "{\"n_estimators\": 50}";
ml model partial_dependence "gb" "[[1, 1], [2, 1], [3, 2], [4, 3], [5, 5], [6, 5], [7, 8], [8, 8]]" 0 4

mentdb
{"feature": 0, "grid": [1.35, 3.4499999999999997, 5.549999999999999, 7.65], "average_prediction": [11.534049420829904, 13.580183754043789, 18.217273099253994, 21.316817774378357]}

Used in use cases (1 recipe)

Data scientist › Explain a model and one of its predictions

Often used with

ml model delete · ml model explain · ml model permutation_importance · ml train

ml model explain <id> <json_background> <json_row>

Description

Explain one prediction: the effect of each feature = prediction of the row minus the average prediction when this feature takes the values of a background data set (the other features keep the values of the row)

Parameters

id: The model id - String - required
json_background: The background rows (ex: the training set) - String - required
json_row: The row to explain, ex: [[8, 2]] - String - required

admin
ml train "lin" "linear" "[[1, 1], [2, 1], [3, 2], [4, 3], [5, 5], [6, 5], [7, 8], [8, 8]]" "[6.1, 8.0, 11.2, 14.1, 18.3, 19.9, 24.8, 26.1]";
ml model explain "lin" "[[1, 1], [2, 1], [3, 2], [4, 3], [5, 5], [6, 5], [7, 8], [8, 8]]" "[[8, 2]]"

mentdb
{"prediction": 20.609433962263978, "baseline": 16.062500000000043, "effects": [6.659905660377163, -2.1129716981132276]}

Classifier: effects on the probability of the predicted class

admin
ml train "rf" "random_forest_classifier" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "[0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2]" "{\"seed\": 1}";
ml model explain "rf" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "[[6, 7.5]]"

mentdb
{"class": 1, "probability": 1, "baseline": 0.33666666666666667, "effects": [0.2733333333333331, 0.4866666666666667]}

Used in use cases (1 recipe)

Data scientist › Explain a model and one of its predictions

Often used with

ml model delete · ml model partial_dependence · ml model permutation_importance · ml train

ml model rules <id>

Description

Get the readable rules of a model: decision tree (one rule per leaf), one_r and ripper (ordered rules: the first rule that fires gives the class), m5 (the linear model of each leaf)

Parameters

id: The model id - String - required

admin
ml train "tree" "decision_tree_classifier" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "[\"low\", \"low\", \"low\", \"low\", \"mid\", \"mid\", \"mid\", \"mid\", \"high\", \"high\", \"high\", \"high\"]";
ml model rules "tree"

mentdb
[{"if": "x0 <= 3.25", "then": "low", "probability": 1, "samples": 4}, {"if": "x0 > 3.25 and x1 <= 4.5", "then": "high", "probability": 1, "samples": 4}, {"if": "x0 > 3.25 and x1 > 4.5", "then": "mid", "probability": 1, "samples": 4}]

RIPPER

admin
ml train "rip" "ripper" "[[2, 1], [3, 2], [6, 1], [7, 2], [8, 1], [6, 6], [7, 8], [2, 7], [3, 5], [1, 2], [9, 2], [8, 7], [4, 4], [6.5, 2.5], [2.5, 8]]" "[\"no\", \"no\", \"yes\", \"yes\", \"yes\", \"no\", \"no\", \"no\", \"no\", \"no\", \"yes\", \"no\", \"no\", \"yes\", \"no\"]";
ml model rules "rip"

mentdb
[{"if": "x0 > 5.5 and x1 <= 5", "then": "yes", "probability": 0.8571428571428571, "samples": 5}, {"if": "true", "then": "no", "probability": 0.9166666666666666, "samples": 10}]

M5 model tree

admin
ml train "m5" "m5" "[[1], [2], [3], [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24]]" "[1.5, 3.5, 6.3, 7.1, 9.9, 11.1, 14.4, 16.1, 18.5, 19.8, 22.2, 23.9, 22.7, 22.1, 20.5, 19.9, 19.3, 18.0, 16.8, 16.9, 15.0, 13.8, 13.1, 11.8]";
ml model rules "m5"

mentdb
[{"if": "x0 <= 6.5", "then": "y = -0.23333326144762 + 1.9428571223183675 * x0", "samples": 6}, {"if": "x0 > 6.5 and x0 <= 10.5", "then": "y = 1.390000189719995 + 1.8599999776800005 * x0", "samples": 4}, {"if": "x0 > 10.5 and x0 <= 14.5", "then": "y = 22.725", "samples": 4}, {"if": "x0 > 14.5 and x0 <= 20.5", "then": "y = 32.86666651549524 - 0.8171428485044899 * x0", "samples": 6}, {"if": "x0 > 20.5", "then": "y = 36.599999721900005 - 1.0299999876400001 * x0", "samples": 4}]

Used in use cases (2 recipes)

Process & reactor engineer › Yield regimes with a model tree
Industrial quality engineer › Readable rules of the defects

Often used with

ml model delete · ml train · ml cross_validate

ml model pmml <id>

Description

Export a model to PMML 4.4 (XML) to score it in another tool (JPMML, KNIME, SAS, R...): linear, ridge, lasso, elastic_net, huber, ransac, wls, poisson, logistic, naive_bayes, decision trees, random forests, extra trees, gradient boosting, kmeans; the features are named x0, x1...

Parameters

id: The model id - String - required

admin
ml train "lin" "linear" "[[1, 1], [2, 1], [3, 2], [4, 3], [5, 5], [6, 5], [7, 8], [8, 8]]" "[6.1, 8.0, 11.2, 14.1, 18.3, 19.9, 24.8, 26.1]";
ml model pmml "lin"

mentdb
<?xml version="1.0" encoding="UTF-8"?>
<PMML xmlns="http://www.dmg.org/PMML-4_4" version="4.4">
<Header description="MentDB model lin (linear)"><Application name="MentDB"/></Header>
<DataDictionary numberOfFields="3">
<DataField name="x0" optype="continuous" dataType="double"/>
<DataField name="x1" optype="continuous" dataType="double"/>
<DataField name="y" optype="continuous" dataType="double"/>
</DataDictionary>
<RegressionModel functionName="regression" algorithmName="linear">
<MiningSchema>
<MiningField name="y" usageType="target"/>
<MiningField name="x0"/>
<MiningField name="x1"/>
</MiningSchema>
<RegressionTable intercept="3.398113207547425">
<NumericPredictor name="x0" coefficient="1.9028301886791894"/>
<NumericPredictor name="x1" coefficient="0.9943396226415189"/>
</RegressionTable>
</RegressionModel>
</PMML>

ml model pmml <id> <json_params>

Description

PMML export with parameters: features (the names of the columns), target (the name of the predicted field, y)

Parameters

id: The model id - String - required
json_params: The parameters - String - required

admin
ml train "tree" "decision_tree_classifier" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "[\"low\", \"low\", \"low\", \"low\", \"mid\", \"mid\", \"mid\", \"mid\", \"high\", \"high\", \"high\", \"high\"]" "{\"max_depth\": 2}";
ml model pmml "tree" "{\"features\": [\"length\", \"width\"], \"target\": \"species\"}"

mentdb
<?xml version="1.0" encoding="UTF-8"?>
<PMML xmlns="http://www.dmg.org/PMML-4_4" version="4.4">
<Header description="MentDB model tree (decision_tree_classifier)"><Application name="MentDB"/></Header>
<DataDictionary numberOfFields="3">
<DataField name="length" optype="continuous" dataType="double"/>
<DataField name="width" optype="continuous" dataType="double"/>
<DataField name="species" optype="categorical" dataType="string">
<Value value="high"/>
<Value value="low"/>
<Value value="mid"/>
</DataField>
</DataDictionary>
<TreeModel functionName="classification" splitCharacteristic="binarySplit" algorithmName="decision_tree_classifier">
<MiningSchema>
<MiningField name="species" usageType="target"/>
<MiningField name="length"/>
<MiningField name="width"/>
</MiningSchema>
<Output>
<OutputField name="predicted_species" optype="categorical" dataType="string" feature="predictedValue"/>
<OutputField name="probability_high" optype="continuous" dataType="double" feature="probability" value="high"/>
<OutputField name="probability_low" optype="continuous" dataType="double" feature="probability" value="low"/>
<OutputField name="probability_mid" optype="continuous" dataType="double" feature="probability" value="mid"/>
</Output>
<Node score="high" recordCount="12">
<True/>
<ScoreDistribution value="high" recordCount="4.0" probability="0.3333333333333333"/>
<ScoreDistribution value="low" recordCount="4.0" probability="0.3333333333333333"/>
<ScoreDistribution value="mid" recordCount="4.0" probability="0.3333333333333333"/>
<Node score="low" recordCount="4">
<SimplePredicate field="length" operator="lessOrEqual" value="3.25"/>
<ScoreDistribution value="high" recordCount="0.0" probability="0.0"/>
<ScoreDistribution value="low" recordCount="4.0" probability="1.0"/>
<ScoreDistribution value="mid" recordCount="0.0" probability="0.0"/>
</Node>
<Node score="high" recordCount="8">
<SimplePredicate field="length" operator="greaterThan" value="3.25"/>
<ScoreDistribution value="high" recordCount="4.0" probability="0.5"/>
<ScoreDistribution value="low" recordCount="0.0" probability="0.0"/>
<ScoreDistribution value="mid" recordCount="4.0" probability="0.5"/>
<Node score="high" recordCount="4">
<SimplePredicate field="width" operator="lessOrEqual" value="4.5"/>
<ScoreDistribution value="high" recordCount="4.0" probability="1.0"/>
<ScoreDistribution value="low" recordCount="0.0" probability="0.0"/>
<ScoreDistribution value="mid" recordCount="0.0" probability="0.0"/>
</Node>
<Node score="mid" recordCount="4">
<SimplePredicate field="width" operator="greaterThan" value="4.5"/>
<ScoreDistribution value="high" recordCount="0.0" probability="0.0"/>
<ScoreDistribution value="low" recordCount="0.0" probability="0.0"/>
<ScoreDistribution value="mid" recordCount="4.0" probability="1.0"/>
</Node>
</Node>
</Node>
</TreeModel>
</PMML>

Used in use cases (1 recipe)

Data scientist › Choose a model automatically and export it in PMML

Often used with

ml automl · ml model delete · ml model predict · string length

ml model info <id>

Description

Get the information of a model (the same result as ml train / ml fit, without the training score)

Parameters

id: The model id - String - required

admin
ml fit "km" "kmeans" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "{\"k\": 2}";
ml model info "km"

mentdb
{"id": "km", "algo": "kmeans", "kind": "clustering", "n_samples": 12, "n_features": 2, "params": {"k":2}, "centers": [[3.4, 5], [8.875, 1.125]], "sizes": [8, 4], "labels": [0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1], "inertia": 116.875, "iterations": 3}

Used in use cases (1 recipe)

Machine learning engineer › Save a trained model and score with it later

Often used with

file delete · ml model delete · ml model load · ml model predict_proba · ml model save · ml train

ml model show

Description

Get the models in memory

admin
ml fit "km" "kmeans" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "{\"k\": 2}";
ml model show

mentdb
[{"id": "km", "algo": "kmeans", "kind": "clustering", "n_samples": 12, "n_features": 2, "created": 1793777181007}]

ml model exist <id>

Description

Check if a model exists in memory (1 or 0)

Parameters

id: The model id - String - required

admin
ml model exist "unknown_model"

mentdb
0

ml model delete <id>

Description

Delete a model from the memory

Parameters

id: The model id - String - required

admin
ml fit "km" "kmeans" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "{\"k\": 2}";
ml model delete "km"

mentdb
1

Used in use cases (29 recipes)

Chemist › Unmix absorption spectra
Process & reactor engineer › Yield regimes with a model tree
Biologist & microbiologist › Monotonic dose-response curve
Epidemiologist › Odds ratios of a case-control study
Geneticist & bioinformatician › Sparse gene signature with an L1 logistic regression
Climate & environment scientist › Temperature between sparse stations
and 23 more in the Use cases

Often used with

ml train · ml model predict · ml fit · ml model transform · ml cross_validate · ml model predict_interval · ml model predict_proba · ml model rules

ml model delete_all

Description

Delete all the models from the memory (returns the number of deleted models)

admin
ml fit "km" "kmeans" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "{\"k\": 2}";
ml model delete_all

mentdb
1

ml model save <id> <path>

Description

Save a model in a file (compressed Java serialization). A restricted session can only write in tmp/[user]/.

Parameters

id: The model id - String - required
path: The file path - String - required

admin
ml train "nb" "naive_bayes" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "[0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2]";
ml model save "nb" "tmp/nb.model"

mentdb
1

Used in use cases (1 recipe)

Machine learning engineer › Save a trained model and score with it later

Often used with

file delete · ml model delete · ml model info · ml model load · ml model predict_proba · ml train

ml model load <id> <path>

Description

Load a model from a file and keep it in memory with a new id

Parameters

id: The new model id - String - required
path: The file path - String - required

admin
ml train "nb" "naive_bayes" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "[0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2]";
ml model save "nb" "tmp/nb.model";
ml model load "nb2" "tmp/nb.model"

mentdb
{"id": "nb2", "algo": "naive_bayes", "kind": "classifier", "n_samples": 12, "n_features": 2, "params": {}, "classes": [0, 1, 2], "priors": [0.3333333333333333, 0.3333333333333333, 0.3333333333333333], "means": [[1.125, 1.975], [5.675, 8.025], [8.875, 1.125]], "variances": [[0.06687501022854166, 0.02187501022854169], [0.21687501022854172, 0.12687501022854167], [0.06687501022854152, 0.06687501022854166]]}

Used in use cases (1 recipe)

Machine learning engineer › Save a trained model and score with it later

Often used with

file delete · ml model delete · ml model info · ml model predict_proba · ml model save · ml train

ml cross_validate <algo> <json_X> <json_y> <k> <json_params> <seed>

Description

k-fold cross validation of an algorithm: accuracy (classifier) or R2 (regressor) on each fold, mean and standard deviation

Parameters

algo: The algorithm - String - required
json_X: The rows - String - required
json_y: The labels or values - String - required
k: The number of folds - Number - required
json_params: The parameters - String - required
seed: The seed of the shuffle - Number - required

admin
ml cross_validate "naive_bayes" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "[0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2]" 4 "{}" 1

mentdb
{"metric": "accuracy", "scores": [1, 1, 1, 1], "mean": 1, "std": 0}

Regression

admin
ml cross_validate "linear" "[[1, 1], [2, 1], [3, 2], [4, 3], [5, 5], [6, 5], [7, 8], [8, 8]]" "[6.1, 8.0, 11.2, 14.1, 18.3, 19.9, 24.8, 26.1]" 4 "{}" 1

mentdb
{"metric": "r2", "scores": [0.9988641915349603, 0.9969113423517159, 0.9960183901670927, 0.7491296813649689], "mean": 0.9352309013546845, "std": 0.12407317101685685}

Used in use cases (8 recipes)

Process & reactor engineer › Yield regimes with a model tree
Geneticist & bioinformatician › Sparse gene signature with an L1 logistic regression
Data scientist › Select the useful variables of a regression
Machine learning engineer › Benchmark five classifiers with a cross validation
Deep learning engineer › Predictive maintenance with a neural network
Data scientist › Features that only work together
and 2 more in the Use cases

Often used with

ml model delete · ml train · ml prep columns · ml prep select_k_best · math max · ml model predict · ml model predict_proba · ml model rules

ml grid_search <algo> <json_X> <json_y> <k> <json_grid> <seed>

Description

Try every combination of parameters with a cross validation and give the best one

Parameters

algo: The algorithm - String - required
json_X: The rows - String - required
json_y: The labels or values - String - required
k: The number of folds - Number - required
json_grid: The values to try for each parameter, ex: {"k": [1, 3, 5]} - String - required
seed: The seed of the shuffle - Number - required

admin
ml grid_search "knn_classifier" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "[0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2]" 3 "{\"k\": [1, 3, 5]}" 1

mentdb
{"best_params": {"k":1}, "best_score": 1, "results": [{"metric": "accuracy", "scores": [1, 1, 1], "mean": 1, "std": 0, "params": {"k":1}}, {"metric": "accuracy", "scores": [1, 1, 1], "mean": 1, "std": 0, "params": {"k":3}}, {"metric": "accuracy", "scores": [0.75, 0.5, 0.25], "mean": 0.5, "std": 0.25, "params": {"k":5}}]}

Used in use cases (1 recipe)

Machine learning engineer › Tune a gradient boosting with a grid search

Often used with

ml model delete · ml model predict · ml train

ml compare <json_X> <json_y> <json_estimators>

Description

Compare algorithms like the Weka Experimenter: the same 10 x 10-fold cross validation for each one, then a corrected resampled paired t-test (Nadeau and Bengio) of each algorithm against the first one (better, worse or same at the level 0.05)

Parameters

json_X: The rows - String - required
json_y: The labels or values - String - required
json_estimators: The algorithms: names or {"algo": ..., "params": {...}}, the first one is the baseline - String - required

admin
ml compare "[[1, 5], [2, 6], [3, 5], [4, 7], [5, 6], [6, 8], [7, 7], [8, 8], [2, 8], [6, 5], [3, 7], [7, 5]]" "[0, 0, 1, 0, 0, 1, 1, 1, 1, 0, 0, 1]" "[\"naive_bayes\", \"logistic\", {\"algo\": \"knn_classifier\", \"params\": {\"k\": 3}}]"

mentdb
{"metric": "accuracy", "k": 10, "repeats": 10, "baseline": "naive_bayes", "level": 0.05, "results": [{"algo": "naive_bayes", "params": {}, "mean": 0.485, "std": 0.4738356276391656}, {"algo": "logistic", "params": {}, "mean": 0.42, "std": 0.46449712136629623, "difference": -0.065, "t": -0.6439937299584969, "p_value": 0.5210687016502709, "vs_baseline": "same"}, {"algo": "knn_classifier", "params": {"k":3}, "mean": 0.66, "std": 0.4431293675255983, "difference": 0.175, "t": 1.3283444772484392, "p_value": 0.18711840363489518, "vs_baseline": "same"}]}

ml compare <json_X> <json_y> <json_estimators> <json_params>

Description

Compare algorithms with parameters: k (folds, 10), repeats (10), level (0.05), seed

Parameters

json_X: The rows - String - required
json_y: The labels or values - String - required
json_estimators: The algorithms - String - required
json_params: The parameters - String - required

admin
ml compare "[[1], [2], [3], [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24]]" "[1.5, 3.5, 6.3, 7.1, 9.9, 11.1, 14.4, 16.1, 18.5, 19.8, 22.2, 23.9, 22.7, 22.1, 20.5, 19.9, 19.3, 18.0, 16.8, 16.9, 15.0, 13.8, 13.1, 11.8]" "[\"linear\", \"m5\", {\"algo\": \"knn_regressor\", \"params\": {\"k\": 3}}]" "{\"k\": 4, \"repeats\": 5}"

mentdb
{"metric": "r2", "k": 4, "repeats": 5, "baseline": "linear", "level": 0.05, "results": [{"algo": "linear", "params": {}, "mean": -0.1715513821833993, "std": 0.7654129813551466}, {"algo": "m5", "params": {}, "mean": 0.9551735849400375, "std": 0.09852605532975835, "difference": 1.1267249671234367, "t": 2.408359166194989, "p_value": 0.026342780654552644, "vs_baseline": "better"}, {"algo": "knn_regressor", "params": {"k":3}, "mean": 0.8925861285952831, "std": 0.08130492980481316, "difference": 1.0641375107786826, "t": 2.4166915651125827, "p_value": 0.025889390550725366, "vs_baseline": "better"}]}

Used in use cases (2 recipes)

Geneticist & bioinformatician › Gene dependencies with a Bayesian network
Statistician & biostatistician › Is the best model really better?

Often used with

ml model delete · ml train

ml automl <id> <json_X> <json_y>

Description

AutoML: cross validation of about 15 algorithms with small grids of parameters (classification or regression, detected from y), a leaderboard, then the best one trained on all the rows and saved under the id

Parameters

id: The id of the best model - String - required
json_X: The rows - String - required
json_y: The labels or values - String - required

admin
ml automl "best" "[[1], [2], [3], [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24]]" "[1.5, 3.5, 6.3, 7.1, 9.9, 11.1, 14.4, 16.1, 18.5, 19.8, 22.2, 23.9, 22.7, 22.1, 20.5, 19.9, 19.3, 18.0, 16.8, 16.9, 15.0, 13.8, 13.1, 11.8]";
ml model predict "best" "[[25], [26]]"

mentdb
[10.391571616788148, 9.275131028420626]

Complete: a table with missing values and a text column -> the best model -> predicted price

admin
#House prices (k EUR): surface, rooms (some missing), city;
-> "[rows]" "[{\"surface\": 45, \"rooms\": 2, \"city\": \"Saint-Denis\", \"price\": 162}, {\"surface\": 80, \"rooms\": 4, \"city\": \"Saint-Paul\", \"price\": 315}, {\"surface\": 62, \"rooms\": 3, \"city\": \"Saint-Denis\", \"price\": 215}, {\"surface\": 110, \"rooms\": 5, \"city\": \"Saint-Paul\", \"price\": 430}, {\"surface\": 38, \"rooms\": null, \"city\": \"Le Tampon\", \"price\": 105}, {\"surface\": 95, \"rooms\": 4, \"city\": \"Le Tampon\", \"price\": 252}, {\"surface\": 70, \"rooms\": 3, \"city\": \"Saint-Paul\", \"price\": 280}, {\"surface\": 55, \"rooms\": 2, \"city\": \"Le Tampon\", \"price\": 150}, {\"surface\": 120, \"rooms\": 6, \"city\": \"Saint-Denis\", \"price\": 410}, {\"surface\": 66, \"rooms\": null, \"city\": \"Saint-Denis\", \"price\": 228}, {\"surface\": 88, \"rooms\": 4, \"city\": \"Saint-Paul\", \"price\": 342}, {\"surface\": 75, \"rooms\": 3, \"city\": \"Le Tampon\", \"price\": 196}, {\"surface\": 50, \"rooms\": 2, \"city\": \"Saint-Paul\", \"price\": 205}, {\"surface\": 100, \"rooms\": 5, \"city\": \"Le Tampon\", \"price\": 268}, {\"surface\": 58, \"rooms\": 3, \"city\": \"Saint-Denis\", \"price\": 199}]";
-> "[fields]" "[\"surface\", \"rooms\", \"city\"]";
-> "[enc]" "{\"one_hot\": {\"city\": [\"Le Tampon\", \"Saint-Denis\", \"Saint-Paul\"]}}";
#Matrix, then the missing rooms replaced by the median;
-> "[X]" (ml prep impute (ml prep matrix [rows] [fields] [enc]) "median");
-> "[y]" (ml prep vector [rows] "price");
json load "tuto_a" (ml automl "tuto_price" [X] [y] "{\"k\": 3, \"include\": [\"linear\", \"ridge\", \"random_forest_regressor\", \"gradient_boosting_regressor\"]}");
-> "[best]" (concat (json select "tuto_a" "/leaderboard/[0]/algo") " (cross validated R2 " (math round (json select "tuto_a" "/leaderboard/[0]/score") 3) ")");
json unload "tuto_a";
#Price of a 72 m2 house with 3 rooms in Saint-Paul;
-> "[p]" (ml model predict "tuto_price" (ml prep matrix "[{\"surface\": 72, \"rooms\": 3, \"city\": \"Saint-Paul\"}]" [fields] [enc]));
ml model delete "tuto_price";
concat "best model: " [best] ", predicted price: " [p]

mentdb
best model: ridge (cross validated R2 0.911), predicted price: [280.49471794487886]

ml automl <id> <json_X> <json_y> <json_params>

Description

AutoML with parameters: task (auto, classification or regression), k (folds, 5), seed, time_budget (seconds, 120), include and exclude (lists of algorithm names)

Parameters

id: The id of the best model - String - required
json_X: The rows - String - required
json_y: The labels or values - String - required
json_params: The parameters - String - required

admin
ml automl "best" "[[1, 2], [1.5, 1.8], [1.2, 2.2], [0.8, 1.9], [5, 8], [6, 8.5], [5.5, 7.5], [6.2, 8.1], [9, 1], [8.5, 1.5], [9.2, 0.8], [8.8, 1.2]]" "[0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2]" "{\"k\": 3, \"include\": [\"logistic\", \"naive_bayes\", \"knn_classifier\", \"decision_tree_classifier\"]}"

mentdb
{"task": "classification", "metric": "accuracy", "k": 3, "leaderboard": [{"rank": 1, "algo": "logistic", "params": {}, "score": 1, "std": 0, "seconds": 0.001}, {"rank": 2, "algo": "logistic", "params": {"alpha": 1}, "score": 1, "std": 0, "seconds": 0}, {"rank": 3, "algo": "naive_bayes", "params": {}, "score": 1, "std": 0, "seconds": 0.001}, {"rank": 4, "algo": "decision_tree_classifier", "params": {"max_depth": 4}, "score": 1, "std": 0, "seconds": 0}, {"rank": 5, "algo": "knn_classifier", "params": {"k": 5, "scale": true}, "score": 0.5833333333333333, "std": 0.14433756729740646, "seconds": 0}, {"rank": 6, "algo": "decision_tree_classifier", "params": {"max_depth": 8, "min_samples_leaf": 3}, "score": 0.5, "std": 0, "seconds": 0}, {"rank": 7, "algo": "knn_classifier", "params": {"k": 15, "scale": true}, "score": 0.16666666666666666, "std": 0.14433756729740646, "seconds": 0}], "skipped": [], "seconds": 0.003, "model": {"id": "best", "algo": "logistic", "kind": "classifier", "n_samples": 12, "n_features": 2, "params": {}, "classes": [0, 1, 2], "train_accuracy": 1, "strategy": "one-vs-rest", "coef": [[45.15277690362241, -7.684780588426674, -4.092351151576107], [-47.08001974256776, 0.8937170889558808, 9.049806456315101], [-26.843751983452414, 7.124528336801502, -5.351343285475679]]}}

Used in use cases (1 recipe)

Data scientist › Choose a model automatically and export it in PMML

Often used with

ml model delete · ml model pmml · ml model predict · string length