O que é ML Model Selector?

Choosing a machine learning model is often harder than fitting one. This ML model selector (dataset → model recommender) profiles your CSV in the browser — rows, column types, missingness, class imbalance, text-heavy vs tabular signals, and likely task type — then ranks model families with transparent heuristic scores. You get top picks for classification, regression, clustering, anomaly detection, time series, and NLP paths, each with pros/cons, preprocessing steps, evaluation metrics, library suggestions (scikit-learn, XGBoost, LightGBM, CatBoost), and a starter code snippet. Prefer accuracy, speed, interpretability, small data, or production simplicity and the ranking updates. Nothing is trained on your data and the CSV never leaves your device, so it is safe for sensitive datasets. Use it as a decision-support checklist before you open a notebook — not as an AutoML trainer.

Como usar ML Model Selector

1. Upload a CSV (or pick a sample) — or switch to Manual mode and describe rows, features, and target type.
2. Confirm or override the task (classification, regression, clustering, etc.) and set your priority.
3. Select the target column if needed and tick constraints like imbalance or explainability.
4. Review the best pick, ranked alternatives, dataset insights, and preprocessing/metrics checklists.
5. Copy or export the Markdown/JSON report, then try the suggested starter code in your own environment.

Casos de uso comuns

• Picking a first model family after receiving a new tabular CSV
• Checking whether boosting, linear models, or clustering fits a described dataset
• Explaining model choice trade-offs (accuracy vs interpretability) to stakeholders

Perguntas frequentes

Does this tool train models on my data?

No. It only profiles the dataset and applies transparent recommendation rules. You still train models locally with scikit-learn, XGBoost, or similar.

Is my CSV uploaded to a server?

No. Parsing and profiling run entirely in your browser. Your file is not stored on our servers.

What if I have fewer than 200 rows?

The selector warns against deep learning and high-capacity models, and boosts simpler options like logistic/linear regression or shallow trees.

How does it handle imbalanced classification?

It detects class imbalance from the target column (or your manual flag), suggests F1/PR-AUC, and favors models/settings that support class_weight or resampling notes.

Can I use it without uploading a file?

Yes. Manual mode lets you enter approximate rows, feature count, target type, and traits like categoricals or text, then still returns ranked recommendations.