This repository contains the refactored code for ProtoCNN model that predicts the function of protein domains, based on the PFam dataset. The repository also includes utility functions that handle the data, CLI input, tests and Dockerfile that builds local environment. The code before refactoring can be seen in ml-pipeline-refactoring.ipynb.
Build image: docker build --tag pfam-image .
Run container: docker run -d --name pfam -v '<DATA_DIR>':'/src/random_split' pfam-image, where <DATA_DIR> is the location of the PFam dataset on your machine.
Access container's shell: docker exec -it pfam bash
CLI for ProtoCNN model can be accessed using pfam comand in the container's shell. The documentation can be accessed using pfam -h command. The basic usage consists of initializing the model and datasets using pfam init, then training the model using pfam train and evaluating the model on dev dataset using pfam validate. Finally, when you're done working with the model you can use pfam del to delete model and datasets.
IMPORTNANT: pfam can only be accessed from /src directory inside the shell.
Repository includes unit tests that can be launched using pytest command in the /src directory.