Training engine of EchoApp.

Four classes mainly used for training process:

• DatasetHandler: This class is implemented in the dataset directory and creates data generators from a given dataset by using the create_data_generators method.
• PreProcessor: This class is implemented in the model directory and performs required processes before using the data generators in training and finally returns processed data in a new data generator. The following image shows these processes for a dataset:

• Model: This class is implemented in the model directory and define the structure of the model that wants to be trained and metrics that wants to be used for evaluating the model during the training process.
• Trainer: This class is implemented in the training directory and Train the given model based on the inputs and config file moreover store tensorboard and mlflow logs by using train method and determine the best weights for the model, and store the config file by using export method.

The model could be trained by using the following command:

python train.py --experiment_dir /path/to/experiment

by using the following command the program will determine to load the corresponding dataset, model, and preprocess's by using the config file in the experiment directory and then train the model based on the loaded files.

This command could be used to create a conda environment of the project:

conda env create -f environment.yml

This command could be used to install the requirements of this project on google colab:

!pip install -r requirements_colab.txt

• After cloning this project from git, the following commands could be used to create the conda environment based on the environment.yml file: echotrain.py -c or echotrain.py --create_env

• Any time that you need to build the package for model directory the following commands could be used: echotrain.py -b or echotrain.py --build_model. This will generate a .whl file inside dist sub-directory which will be used to install the model package using pip install file_name.whl on any target python environment.

• Also, you could install the requirements needed in the colab by using the following commands: echotrain.py -cr or echotrain.py --colab_requirements

• Finally, the conda environment could be created in the colab by using the following commands: echotrain.py -ce or echotrain.py --colab_conda_env