/pycaret-streamlit

An End-to-End Machine Learning Web Application for Classification and Regression problem using AutoML framework Pycaret.

Primary LanguageJupyter Notebook

Streamlit-pycaret Web Application

Build and Deploy to GKE Docker Image CI Azure CICD Pipeline

An End-to-End Machine Learning Web Application for Classification and Regression problem. Currently supported for csv and excel files. The application relies on these two excellent libraries for machine learning:

Features

  1. Drap and drop file from local system for training.

  2. Simple Data Exploration.

  3. Many Preprocessing methods:

    • Sample and Split
    • Data Preprocessing (Missing Values Imputation, One Hot Encoding, Handle Unknown Levels, Fix Imbalance for Classification)
    • Scale and Transform (Normalization, Transformation, Target Transformation)
    • Feature Engineering (Feature Interaction, Polynomial Features, Trigonometry Features, Group Features, Bin Numeric Features, Combine Rare Levels)
    • Feature Selection (Feature Importance, Remove Multicollinearity, Principal Components Analysis, Ignore Variances)
    • Unsupervised (Create Clusters, Remove Outliers)

  4. Model Training:

    • Compare all available Machine Learning Algorithm automatically.
    • Train a selected single model
    • Train an ensemble model
    • Hyperparameter tuning for single model

  5. Model Result Visualization:

    • All plots for Regression and Classification
    • SHAP Value

  6. Prediction and Save Model:

    • Online Prediction
    • Batch Prediction
    • Save whole Machine Learning Pipeline as pickle file

Install and Run

  1. Clone the repository to you computer:
git clone https://github.com/redcican/pycaret-eidodata.git
cd pycaret-eidodata
  1. Creata a conda virtual or python virtual environment and then activate it.
conda create -n myvirtual-name python=3.8 -y
conda activate myvirtual-name
  1. Install requirements
pip install -r requirements.txt
  1. Run streamlit locally and start web service:
streamlit run app.py

Run on Docker

  1. Pull the Docker image:
docker pull cican/github-action:latest
  1. Run the Docker image locally:
docker run -p 8501:8501 cican/github-action
  1. Open the localhost:8501