Easy-ML Studio

Easy-ML Studio is a no-code machine learning studio that runs on your own machine. Upload a dataset in the browser, train several models side by side, compare their metrics, make predictions, explain them with SHAP, and export a trained model as a ready-to-run FastAPI service.

It covers three workflows:

• Tabular classification and regression (CSV / Excel)
• Text classification (TF-IDF or sentence embeddings)
• Image classification (SimpleCNN, ResNet18, MobileNetV2, or your own CNN)

Everything stays local: the UI, the models and your data never leave your computer.

Install

pip install easy-ml-studio

The core install covers tabular ML with scikit-learn models and TF-IDF text classification. Add extras for more:

pip install "easy-ml-studio[all]"
easy-ml download-models   # spaCy English model, needed for lemmatization

Python 3.10 – 3.12 is supported. Features whose extra is not installed are greyed out in the UI with the command to enable them.

Launch

easy-ml

The studio opens in your browser at http://127.0.0.1:5010. Useful options:

easy-ml --port 8080          # use another port
easy-ml --no-browser         # don't open a browser tab
easy-ml --host 0.0.0.0       # expose on your network (see Security below)
python -m easyml             # same as easy-ml

Features

• Train several models in one run and compare accuracy, confusion matrices and classification reports
• Result charts for every model: ROC and precision-recall curves, predicted vs actual, errors, feature importance and top words
• Light or thorough hyperparameter search (GridSearchCV)
• Missing-value imputation, scaling, PCA, one-hot encoding, and SMOTE / Tomek Links for imbalanced classes
• Live progress and training curves while models train; stop a run at any time
• Run history stored in SQLite, so every model can be reused later
• Retrain any run on new data with the same settings, and compare two runs side by side
• Move a run to another computer as a single .zip (download and import)
• Single-record and batch (CSV) predictions
• SHAP waterfall explanations for tabular predictions
• One-click export of a tabular model as a FastAPI app with Dockerfile
• Use any trained model from Python or the command line (easyml.load_model, easy-ml predict)
• Automatic EDA reports with ydata-profiling
• Light and dark themes, works fully offline

Use your models outside the studio

Every run is saved, so you can use its models from any Python program or script on the same machine:

import easyml

model = easyml.load_model(5)                      # run #5, best model by default
model.predict({"age": 45, "city": "NY"})          # -> ["high"]
model.predict("new_customers.csv")                # one prediction per row
model.predict(["great phone", "late delivery"])   # text runs
model.predict("photos/")                          # image runs: a file, a list or a folder

Or from the command line:

easy-ml runs                                        # list runs with their best model
easy-ml predict 5 new_customers.csv -o out.csv      # adds Prediction (and Confidence) columns
easy-ml predict 3 "my laptop got a software update" # text runs print the label
easy-ml predict 7 photos/ -o labels.csv             # image runs: one row per image
easy-ml retrain 5 october.csv                       # run #5's recipe on new data
easy-ml export 5 -o churn.zip                       # move run #5 to another computer...
easy-ml import churn.zip                            # ...and load it there

Pick a specific model with model_name= / --model (for example RandomForestClassifier). The studio's prediction page shows these snippets pre-filled for each run.

Train from Python

from easyml import MLTrainer

run = MLTrainer(
    task="MLClassification",          # or "MLRegression", "NLPTextClassification"
    name="churn",
    dataset_path="churn.csv",
    target_column="churned",
    models=["logistic_regression", "random_forest"],
).run()

print(run.run_id, run.performance)

Runs trained from Python appear in the studio's Run History and can be used for predictions there. The Python guide lists every option.

Where data is stored

Uploads, trained models, the run database and logs live outside the package, so upgrading never deletes them:

Set the EASYML_DATA_DIR environment variable to use another folder. Settings → Clear Data in the studio deletes everything stored there.

Security

Easy-ML Studio is a local, single-user tool with no login.

• By default it only listens on 127.0.0.1, so other machines cannot reach it.
• Requests from other websites are rejected, so a page you visit cannot drive the studio.
• Only use --host 0.0.0.0 on a network you trust: anyone who can reach the port can use the studio and its data.

Docker

docker compose up --build

The image targets base, ml, nlp, cv and full (default) select which extras are installed. Data is kept in the easyml-data volume.

Documentation

• User guide
• Python guide: train and use models from code and the command line
• Developer guide

Development

git clone https://github.com/Yatrik07/easy-ml.git
cd easy-ml
python -m venv venv
venv\Scripts\activate            # Windows; use `source venv/bin/activate` on macOS/Linux
pip install -e ".[all,dev]"
pytest
easy-ml

License

Apache License 2.0