featureforge-studio 0.1.0

Interactive EDA and feature engineering studio for reproducible ML preprocessing workflows

FeatureForge Studio

FeatureForge Studio is an interactive EDA and feature-engineering workspace for machine learning preprocessing. It gives you a Streamlit app for visual workflows and an importable Python library for scripts, notebooks, and reusable preprocessing pipelines.

Install name: featureforge-studio
Import name: featureforge

Highlights

• Upload CSV datasets and inspect schema, missing values, duplicates, memory usage, and column types.
• Generate smart data-quality warnings for missingness, skew, high cardinality, duplicate rows, and likely ID columns.
• Explore data with interactive Plotly charts: histograms, boxplots, count plots, scatter plots, correlation heatmaps, and missing-value views.
• Queue preprocessing steps for imputation, scaling, encoding, and outlier handling.
• Export cleaned datasets, transformation config JSON, preprocessing code, and transformation reports.
• Use the same backend logic from Python via featureforge.

Installation

Install from a local wheel:

pip install dist/featureforge_studio-0.1.0-py3-none-any.whl

For local development:

git clone https://github.com/aarush07/featureforge.git
cd featureforge/featureforge
python3 -m venv .venv
source .venv/bin/activate
pip install -e .[dev]

After publishing to PyPI:

pip install featureforge-studio

Run The App

From an installed package:

featureforge-app

From the repository:

cd featureforge
streamlit run app.py

If the default Streamlit port is busy:

streamlit run app.py --server.port 8502

Library Usage

import pandas as pd
from featureforge import ProfilingEngine, TransformationManager

df = pd.DataFrame(
    {
        "age": [21, None, 35],
        "city": ["Pune", "Delhi", None],
        "salary": [40000, 52000, 61000],
    }
)

profile = ProfilingEngine().generate_summary(df)

steps = [
    {
        "id": "missing_values_1",
        "type": "missing_values",
        "method": "median",
        "columns": ["age"],
        "params": {},
        "label": "Missing Values - median",
    },
    {
        "id": "encoding_2",
        "type": "encoding",
        "method": "onehot",
        "columns": ["city"],
        "params": {"handle_unknown": "ignore"},
        "label": "Encoding - onehot",
    },
]

processed_df, metadata = TransformationManager().apply_steps(df, steps)

Core API

• DatasetManager: CSV loading, column cleanup, basic validation.
• ProfilingEngine: dataset summaries, missing-value tables, type detection, smart warnings, optional ydata-profiling reports.
• EDAEngine: Plotly figure generation from declarative chart configs.
• TransformationManager: validates and applies transformation queues.
• PipelineBuilder: builds sklearn preprocessing pipelines from transformation config.
• CodeGenerator: exports standalone sklearn preprocessing code.
• ExportEngine: creates CSV, JSON, Python, and Markdown export payloads.

Supported Transformations

Area	Methods
Missing values	mean, median, mode, constant, drop rows
Scaling	StandardScaler, MinMaxScaler, RobustScaler
Encoding	OneHotEncoder, OrdinalEncoder, label-style category codes
Outliers	IQR filtering, Z-score filtering

Build A Distributable Package

python -m pip install build twine
python -m build

This creates:

dist/featureforge_studio-0.1.0-py3-none-any.whl
dist/featureforge_studio-0.1.0.tar.gz

Publish To PyPI

First test the package on TestPyPI:

twine upload --repository testpypi dist/*

Then publish to PyPI:

twine upload dist/*

Note: the PyPI name featureforge is already taken by another project, so this package uses the distribution name featureforge-studio while keeping the clean import name featureforge.

Development

Run tests:

pytest -q

Rebuild after metadata changes:

rm -rf dist build *.egg-info
python -m build

License

MIT License. See LICENSE.