fastAPI-based Machine Learning model training api

Overview

this API, developed using fastAPI, enables the automated training of machine learning models, specifically Support Vector Classification (SVC) and xgboost. By providing a structured interface for dataset ingestion, hyperparameter configuration, and performance evaluation, it facilitates efficient experimentation with different model architectures and optimization strategies.

Features

• dynamic dataset ingestion via api endpoints
• implementation of SVC and xgboost classifiers
• optional hyperparameter tuning using Optuna
• configurable parameters, including test size and random state
• robust logging for monitoring and debugging

Installation

Ensure Python (>=3.8) is installed. Clone the repository and install dependencies:

pip install -r requirements.txt

Running the API

To deploy the API locally, execute:

uvicorn main:app --host 0.0.0.0 --port 8000

API Endpoints

Train support vector classifier (SVC)

POST /train/svc

Request Parameters (multipart/form-data):

• data_file (file, required): CSV file containing feature vectors and labels
• label_column (string, required): Name of the target variable column
• test_column (string, optional): Name of a predefined test set column
• test_size (float, default: 0.2): proportion of the dataset allocated for validation
• random_state (int, default: 42): Seed value for reproducibility
• use_optuna (bool, default: False): Enable hyperparameter optimization
• hyperparams (string, optional): JSON-formatted string specifying hyperparameters
• n_trials (int, default: 10): Number of trials for optuna tuning

Sample Response:

{
  "accuracy": 0.92,
  "f1_score": 0.88,
  "confusion_matrix": [[50, 5], [4, 41]]
}

Train xgboost classifier

POST /train/xgboost

Request Parameters (identical to SVC endpoint)

Sample Response:

{
  "accuracy": 0.94,
  "f1_score": 0.90,
  "confusion_matrix": [[52, 3], [2, 43]]
}

Logging & gpu Utilization

• The api includes structured logging to track significant events and errors.
• gpu availability can be checked, but this functionality is currently disabled.

Future Enhancements

• standardize dataset storage, retrieval, and deletion
• implement early stopping based on performance metrics
• refine evaluation metric selection for improved generalization

License

this project is distributed under the BSD License.