Chokkhu

An End-to-End, Research-Grade ML and Data Science Pipeline Toolkit for Tabular and Computer Vision Datasets.

"Minimalistic Code. Maximum Output. Zero Heavy Dependencies."

Overview

Chokkhu is a high-performance Python toolkit designed to streamline the complete Machine Learning lifecycle including raw data loading, statistical Exploratory Data Analysis, advanced data cleaning, preprocessing, feature transformation, and stratified data splitting.

Core Philosophy and Design Principles

Zero Heavy Dependencies: Built from the ground up using pure NumPy, Pandas, SciPy, Matplotlib, Seaborn, and OpenCV-headless. No Torch, TensorFlow, Scikit-Learn, or imbalanced-learn dependencies required.
Unified Minimalist API: Powerful operations encapsulated in intuitive, one-line functions (load, eda, clean, preprocess, transform, split).
Cross-Platform Compatibility: Tested and verified across 10 environments (Python 3.8 to 3.12 on Linux Ubuntu and Windows).

Installation

Install Chokkhu directly from PyPI via pip:

pip install --upgrade chokkhu

Complete End-to-End Pipeline in One Workflow

import chokkhu as ck

df = ck.load("dataset.csv")

ck.eda.tabular("dataset.csv", target_col="price", save_reports=True)

df_clean = ck.clean(
    df,
    missing="knn",
    outliers="isolation_forest",
    duplicates=True,
    fix_data_types=True
)

df_proc, state = ck.preprocess(
    df_clean,
    target="price",
    scale="standard",
    encode="onehot",
    select_features="correlation"
)

df_trans = ck.transform(
    df_proc,
    target="price",
    resample="smote",
    pca=10
)

X_tr, X_val, X_te, y_tr, y_val, y_te = ck.split(
    df_trans,
    target="price",
    test_size=0.2,
    val_size=0.1,
    stratify=True,
    random_state=42
)

Detailed API Reference

1. Data Loading (`chokkhu.load`)

Auto-detects file format and loads tabular or image datasets seamlessly:

import chokkhu as ck

df = ck.load("data.csv")

images_data = ck.load("images_folder/", data_type="image", target_size=(128, 128))

2. Exploratory Data Analysis (`chokkhu.eda`)

Tabular EDA

Generates univariate, bivariate, and multivariate statistical reports with interactive visualizations:

ck.eda.tabular(
    dataset_path="data.csv",
    target_col="target",
    save_reports=True,
    save_dir="./eda_reports"
)

Image EDA

Analyzes resolution distributions, color spaces, blur scores (Laplacian variance), pure NumPy Shannon entropy, SNR, edge intensity, and perceptual duplicates:

ck.eda.image(
    dataset_path="dataset_images/",
    save_reports=True,
    save_dir="./image_reports"
)

3. Data Cleaning (`chokkhu.clean`)

All-in-one data sanitation function:

df_cleaned = ck.clean(
    data=df,
    missing="knn",
    missing_threshold=0.5,
    outliers="iqr",
    outlier_action="remove",
    duplicates=True,
    fix_data_types=True
)

4. Data Preprocessing (`chokkhu.preprocess`)

Stateful feature scalers, encoders, and feature selection:

df_processed, preprocessor_state = ck.preprocess(
    data=df_cleaned,
    target="target_column",
    scale="standard",
    encode="onehot",
    select_features="mutual_info",
    top_k_features=15
)

5. Data Transformation (`chokkhu.transform`)

Dimensionality reduction, class imbalance resampling, polynomial feature engineering, and image augmentations:

df_transformed = ck.transform(
    data=df_processed,
    target="target_column",
    pca=5,
    lda=2,
    tsne=2,
    resample="smote",
    polynomial=2,
    log_features=["skewed_col"]
)

aug_images_dict = ck.transform(
    data=images_data,
    augment=True,
    augment_techniques=["horizontal_flip", "rotate", "brightness", "contrast", "noise", "crop", "blur", "cutout"],
    augment_factor=3
)

6. Data Splitting (`chokkhu.split`)

Multi-way stratified partitioning and cross-validation generators:

X_train, X_test, y_train, y_test = ck.split(
    df_transformed,
    target="target_column",
    test_size=0.2,
    stratify=True,
    random_state=42
)

X_train, X_val, X_test, y_train, y_val, y_test = ck.split(
    df_transformed,
    target="target_column",
    test_size=0.2,
    val_size=0.1,
    stratify=True,
    random_state=42
)

for fold, (train_df, val_df) in enumerate(ck.split(df, method="kfold", n_splits=5)):
    print(f"Fold {fold}: Train shape={train_df.shape}, Val shape={val_df.shape}")

7. Model Training (`chokkhu.train`)

Chokkhu provides a unified, single-function API chokkhu.train() to train any machine learning or reinforcement learning model. All models are implemented completely from scratch using pure NumPy and SciPy.

Dynamic Hyperparameters: You can pass any hyperparameter dynamically as **kwargs directly into the train() function to override the defaults.

Supervised Learning

Models: linear_regression, ridge, lasso, elastic_net, logistic_regression, knn, naive_bayes, svm, decision_tree, random_forest, gradient_boosting

# Train a Random Forest with default parameters
model = ck.train(model="random_forest", X_train=X_train, y_train=y_train)

# Override hyperparameters dynamically!
model = ck.train(
    model="random_forest",
    X_train=X_train,
    y_train=y_train,
    n_estimators=500,
    max_depth=15,
    min_samples_split=5
)

Unsupervised Learning

Models: kmeans, dbscan, hierarchical

# Train DBSCAN clustering
model = ck.train(
    model="dbscan",
    X_train=X_train,
    eps=1.5,
    min_samples=10
)

Reinforcement Learning (RL)

Models: q_learning (more coming soon!)
Environments: Comes with built-in lightweight environments (like GridWorld), or you can pass your own custom OpenAI Gym environment!

# Using default built-in GridWorld environment
model = ck.train(model="q_learning", episodes=1000)

# Using your own custom environment
model = ck.train(
    model="q_learning",
    env=my_custom_gym_env,
    episodes=5000,
    learning_rate=0.01,
    epsilon_decay=0.99
)

Feature Matrix Summary

Category	Modules & Algorithms Available
I/O	CSV, TSV, JSON, Excel (.xlsx), Parquet, NumPy (.npy, .npz), Image Folders
EDA	Univariate, Bivariate, Multivariate Correlation, Interactive HTML Reports, Image Quality & Blur Metrics, GLCM Texture, Perceptual Duplicates
Cleaning	KNN Imputer, MICE (Iterative), Tukey IQR, Isolation Forest, Z-Score, Winsorization, Auto Dtype Fixer
Preprocessing	Standard, MinMax, Robust, Power, Quantile Scalers; One-Hot, Target, Binary, Frequency Encoders; Variance, Correlation, Mutual Info, RFE Selectors
Transformation	PCA, SVD, LDA, t-SNE, SMOTE, ADASYN, Tomek Links, Image Augmentation (Flip, Rotation, Brightness, Noise, Crop, Blur, Cutout, Mixup), Polynomial Features
ML Models	Linear/Logistic Regression, KNN, Naive Bayes, SVM, Decision Tree, Random Forest, GBM, K-Means, DBSCAN, Hierarchical, Q-Learning (RL)
Splitting	Train/Test, Train/Val/Test (3-way), Stratified Splitting, K-Fold, Stratified K-Fold, TimeSeriesSplit

Contributing

We welcome community contributions! Feel free to submit issues or pull requests to help expand Chokkhu:

Fork the Repository
Create your Feature Branch (git checkout -b feature/AmazingFeature)
Commit your Changes (git commit -m 'Add some AmazingFeature')
Push to the Branch (git push origin feature/AmazingFeature)
Open a Pull Request

License

Distributed under the MIT License. See LICENSE for more details.

Metadata

Release files for chokkhu 0.7.3

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for chokkhu 0.7.3
File	Size	Uploaded
chokkhu-0.7.3.tar.gz	56.6 kB	Aug 21, 2026	Details

Built distribution (wheel)

Table of built distributions (wheels) for chokkhu 0.7.3
File	Interpreter	ABI	Platform	Reset
chokkhu-0.7.3-py3-none-any.whl 75.9 kB Aug 21, 2026	Python 3	none	any	Details

Total release size: 132.5 kB

Release files / chokkhu-0.7.3.tar.gz

Download URL	chokkhu-0.7.3.tar.gz
Size	56.6 kB
Tags	Source
SHA-256 checksum How to use checksums	`1252657a1b882a3d46705b108fef6bd7966a0674fa6bb6d11305db76a9301593`
BLAKE2b-256 checksum How to use checksums	`67296cb538a05ab01b5f8d9da1823743b4d35dc8a668e89384e71618b278982e`
Upload date	Aug 21, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	No
Uploaded via	`twine/7.0.0 CPython/3.13.14`

Release files / chokkhu-0.7.3-py3-none-any.whl

Download URL	chokkhu-0.7.3-py3-none-any.whl
Size	75.9 kB
Tags	Python 3
SHA-256 checksum How to use checksums	`8b743208354872d5ec7782817be809b9d38c05c6771952011c7ad3ffce2bc909`
BLAKE2b-256 checksum How to use checksums	`4cd61c99cb31c463ef6b9166d98d1a63d7bc116278a2d986de17373814f80958`
Upload date	Aug 21, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	No
Uploaded via	`twine/7.0.0 CPython/3.13.14`