automl-rdatasets

🔎🧙 Contextual search and autoML on R datasets using spacy and tpot. Served as a Streamlit web app. Find inspiration (and starter code) for your next ML project. Search for a dataset in your domain of interest from R's extensive collection of data. Run AutoML to find and generate Python code for an optimised ML pipeline!

🎬 Live Demo

You can find the live demo here.

Using the app:

1. Find relevant R datasets using the search bar
2. Select a R dataset
3. Select an outcome variable and supervised ML task (regression or classification) to perform
4. Press the "Run AutoML" button to perform AutoML and generate Python code for the best ML pipeline

Note: The running time for pipeline optimization and evaluation time per iteration is limited to to 10 minutes max. In practice, AutoML with TPOT should be run with multiple instances in parallel for much longer (hours or days). You can modify this limit by modifying the "Maximum running time" Streamlit slider in app.py.

🚀 Install

automl-rdatasets has been tested on Python 3.8 and depends on the following packages:

• sklearn
• spacy
• streamlit
• streamlit-pandas-profiling
• missingno
• numpy
• numba
• pandas
• tpot

To use run the demo locally, you must first clone this repo:

git clone git@github.com:topher-lo/automl-rdatasets.git
cd [..path/to/repo]

Then install its dependencies using either pip:

python3 -m venv env
source env/bin/activate
pip3 install -r requirements.txt

Or run it as a containerised application using docker and docker-compose. First create a file .env with the line:

PORT=8501

to specify which port to publish the data app to. Then run the following commands on your CLI:

docker-compose build
docker-compose up -d

✨ How to run this demo

In your virtual environment, run the following command from the automl-rdatasets dir:

streamlit run app.py

The web app will be available at http://localhost:8501/ Otherwise, if you are using docker, the web app will be available at whichever port you specified in the .env file.

🦾 Tech stack

• Uses spaCy's pre-trained Word2Vec word embeddings and cosine similarity to perform contextual search
• Generates the data profiling report using pandas-profiling
• Generates missing value plots using missingno
• Performs AutoML using TPOT
• UI inspired by Traingenerator

Getting in touch

If you are having a problem with this demo, please raise a GitHub issue. For anything else, you can reach me at: lochristopherhy@gmail.com