Upload your data, choose a target and compare scikit-learn models directly in your browser.
Choose an example. MyMlLab loads the data and selects the correct target and task automatically.
Example datasets are supplied for demonstration and education.
or select a file from your computer
| Upload a CSV or load the demo dataset to begin. |
Audit data quality, distributions, outliers, dependence, multicollinearity and target relationships before modelling.
Load a dataset, choose the target and run the analysis.
| Column | Type | Missing | Missing % | Unique | Role / warning |
|---|---|---|---|---|---|
| Run statistical analysis. | |||||
| Variable | Count | Missing | Mean | Std. | Variance | SEM | 95% CI low | 95% CI high | Min | Q1 | Median | Q3 | Max | IQR | CV % | Skew | Kurtosis | Shapiro p |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Run statistical analysis. | ||||||||||||||||||
| Variable | Count | Missing | Unique | Mode | Mode count | Mode % | Entropy |
|---|---|---|---|---|---|---|---|
| Run statistical analysis. | |||||||
| Variable | IQR count | IQR % | MAD count | MAD % |
|---|---|---|---|---|
| Run statistical analysis. | ||||
| Variable | VIF | Tolerance | Assessment |
|---|---|---|---|
| Run statistical analysis. | |||
| Variable A | Variable B | Pearson r |
|---|---|---|
| Run statistical analysis. | ||
| Feature | Association | ||||
|---|---|---|---|---|---|
| Run statistical analysis. | |||||
Explore scikit-learn model families and select up to three algorithms.
Configure and compare up to three leakage-safe pipelines with holdout or cross-validation.
Pipeline A is always active. Enable B and C to compare preprocessing under identical folds.
Regression ranking metrics
Python and scikit-learn execute locally through WebAssembly.
Load a dataset and configure the experiment.
The MVP does not send your CSV to a training server. Computation happens in this browser tab.
Cross-validation stability and performance on the untouched final test set.
Train models to populate the leaderboard.
| Rank | Model | Score | |||
|---|---|---|---|---|---|
| No results yet. | |||||
| Model | Run an experiment | ||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| No results yet. | |||||||||||||||||||
| Class | Precision | Recall / sensitivity | Specificity | F1 | Support |
|---|
Cross-validation, parameter optimization, feature importance, downloadable models, equation discovery and publication-ready reports.
This MVP contains the free browser engine. Authentication, subscriptions and protected PRO features are the next layer.
Using a web application should not automatically mean uploading your ML dataset to someone else's training server.
In the current MVP, a CSV selected in Studio is read by the browser and processed by the browser-based Python environment. The MVP does not provide a dataset-upload endpoint for model training.
The product prefers computation on the user's device whenever practical. The website and browser libraries are delivered over the internet, but the goal is to keep training data local unless a future feature explicitly requires otherwise.
Future versions may use external services for authentication, subscription status, billing, optional project synchronization or optional cloud computation. Cloud computation should be a clearly distinguishable mode.
This page describes the intended architecture of the current MVP. Before commercial launch, a formal Privacy Policy and Terms should be completed for the actual services in use.
The free version is a real ML tool. PRO is for deeper validation, optimization, explainability and reproducible outputs.
The MVP provides a guided browser workflow for tabular regression and classification using scikit-learn.
Open Studio and upload a CSV dataset. Free currently accepts up to 5,000 rows.
Select the column to predict. MyMlLab attempts to detect regression versus classification.
Run the audit, distributions, dependence, outlier and target diagnostics.
Compare up to three models and three leakage-safe preprocessing pipelines.
Rank combinations with holdout or cross-validation, then score the winner on an untouched final test set.
Use the leaderboard and visualization to compare selected models.
The Statistics step provides column auditing, descriptive statistics, distributions, Q-Q plots, Pearson/Spearman/Kendall correlations, IQR and MAD outliers, VIF/tolerance, target relationships and automatic data-quality recommendations.
Models are organized into scikit-learn-style families. A selected algorithm is shown with a purple highlight, glow, check mark and selected-model chip. Free experiments compare up to three models and up to three independently configured preprocessing pipelines.
Pipeline A can pass clean numerical predictors through unchanged as Original Data. Custom preprocessing includes optional numeric and categorical imputation, optional one-hot or ordinal encoding, five scaling choices, Yeo-Johnson and quantile transforms, variance/F-score/mutual-information selection and PCA. Validation supports an untouched final test set plus holdout, 3-fold, 5-fold or 10-fold cross-validation.
Regression reports R-squared, adjusted R-squared, MAE, MSE, RMSE, median and maximum error, MAPE, sMAPE, explained variance, mean error and conditional MSLE/RMSLE. Classification reports accuracy, balanced accuracy, macro and weighted precision/recall/F1/Jaccard, specificity, MCC, Cohen's kappa, losses, ROC-AUC, PR-AUC, Brier score, a counts/row-percent confusion matrix and per-class metrics.
V13.2 includes 33 regression algorithms and 26 classification algorithms across baselines, linear and regularized models, robust and generalized linear models, discriminant analysis, Naive Bayes, trees, ensembles, nearest neighbors, cross decomposition, neural networks, kernel methods and Gaussian processes. SVM and SVR are deliberately excluded.
Studio includes a 500-row sample of the UCI Combined Cycle Power Plant dataset for regression, the scikit-learn Breast Cancer Wisconsin Diagnostic dataset for binary classification, and the scikit-learn Iris dataset for multiclass classification. Selecting an example configures its target, task and recommended scaler automatically.
A modern desktop browser is recommended. The first load can take longer because the Python runtime and scientific packages must be downloaded.
MyMlLab is a local-first machine-learning studio focused on turning tabular datasets into reproducible experiments without forcing users to configure Python environments or upload data to a remote training service.
Many experiments repeat the same setup work: importing data, identifying feature types, preprocessing, splitting, configuring models, calculating metrics and comparing results. MyMlLab turns that workflow into a guided interface.
The first releases focus on classical tabular ML. Planned directions include stronger validation, hyperparameter optimization, AutoML, explainability, symbolic regression/classification, reproducible export and later browser-based deep learning where hardware allows it.