axelbaher / ecg-classification

Getting started

Implemented models

1. LeNet5
2. AlexNet
3. VGGNet (configuration D)
4. GoogLeNet
5. ResNet-34 (convolutional)

Preparing system:

1. Install Git:
   If you on Linux, you have already installed it.
   Оn Windows:
   https://git-scm.com/download/win
2. Install Python:
   If you on Linux, you have already installed it.
   On Windows:
   https://www.python.org/downloads/
3. Clone repository: git clone https://github.com/axelBaher/ecg-classification.git
4. Setup virtual environment and install packages into it:
   python setup.py
   If script doesn't work for whatever reason, just run this command:
   pip install -r requirements.txt
   In this way, all the packages will be installed in your main (system) Python path.
5. Go to folder with scripts:
   cd main
6. Get necessary db and generate data:
   python prep.py

Train

To start training, you need to run this command, in the figure brackets you need to type model, which will be trained: python train.py --config {model_name}
There are five models to choose (type exactly, as it will be written below):
LeNet5, AlexNet, VGGNetD, GoogLeNet, ResNet34 In the config/training/{model_name} you can find configuration, that will be used in training.

Interence

To start inference, you need to run this commmand:
python inference.py -name {model_name} -epoch {number_of_training_epoch} -b_size {batch_size} -val_split {validation_split} [-loss {loss_function}] [-opt {optimizer}]
You need to input model name and parameters for program to find pretrained weights.

Pipeline

To start pipeline, you need to run this command:
python pipeline.py In the config/pipeline.json you can configure, which models will be trained and tested and with which parameters.

ecg-classification's Issues

Create venv in prep script.

Implement creating venv in prep.py to download there all Python packages.

1D models support.

Make models compatible with 1D arrays.

Add weights saving.

Implement weights saving into file to use it when testing.

Make pipeline.

Make pipeline for training and testing all models with different value sets of parameters. Maybe implement stat table.

Log system (pipeline)

Implement log system, which will be tracking used hyperparameters and results.

Log system (train, test).

Make log system while training and testing process.

Make visualization.

Make visualization for testing results.

Add config system.

Using argparse, implement config system for model choosing via console while training/testing. Also, maybe add config to prep script.

Add clear script.

Add script, that will delete all generated data and db from machine.