It is a Web App for training Machine Learning pipelines with MLJAR AutoML. It works with tabular data. All models are zipped to archive and can be reused to compute predictions in batch mode.
This repo consists of three notebooks:
- notebook for training AutoML with simple UI,
- advanced notebook for training AutoML with more advanced UI (you can select feature engineering methods, algorithms, validation strategy, and evaluation metric),
- notebook for computing predictions.
The Web App is using mljar-supervised for constructing the ML pipeline with AutoML. It automatically handles the following:
- data preprocessing,
- features engineering,
- algorithm selection & tuning,
- ML models explanations,
- automatic documentation.
The Web App is created directly from Jupyter Notebooks with Mercury framework.
The Web App is available online at automl.runmercury.com. Input data upload is limited to 1MB.
Please run the below commands to run Web App locally. It requires Python >= 3.8.
pip install -r requirements.txt
mercury runIf you would like to increase the input file limit, please change the cell:
data_file = mr.File(label="Upload CSV with training data", max_file_size="1MB")and set your max_file_size.
Please change the following cell to increase training time:
time_limit = mr.Select(label="Time limit (seconds)", value="60", choices=["60", "120", "240", "300"])Times are in seconds. Please just increase the values.
Please upload a CSV file with training data, select input features & target, and click Start training.
All models created during the training are available for download as a zip file:
Please use advanced mode if you would like to tweak AutoML parameters:
