training.py

module src.training

Created By: Andrew Player
File Name: training.py
Date Created: 01-25-2021
Description: Contains the code for training models

---

function train

train(
    model_name: str,
    dataset_path: str,
    input_shape: int = 1024,
    num_epochs: int = 10,
    num_filters: int = 16,
    batch_size: int = 64,
    learning_rate: float = 0.001,
    dropout: float = 0.2
) → Any

Trains a model.

Parameters:

• model_name : str
  The name for the saved model.
• train_path : str
  The path to the training dataset.
• test_path : str
  The path to the validation dataset.
• input_shape : int, Optional
  The input shape of the model: (1, input_shape, input_shape, 1).
• num_epochs : int, Optional
  The number of epochs. This is the number of times the model trains over the training dataset.
• num_filters : int, Optional
  The base number of filters for the convolutional layers.
• batch_size : int, Optional
  The number of samples to train on before updating the model weights. For the best accuracy, this should be 0; however, higher values will lead to much quicker training.
• learning_rate : float, Optional
  The rate at which the model will update weights in response to estimated error.
• dropout : float, Optional
  The percentage of network nodes that will be randomly dropped out during the training process. This helps to mitigate overfitting.

Returns:

• history : any
  A history object containing the loss at each epoch of training.

---

class DataGenerator

Dataset Generator for sequencially passing files from storange into the model.

method __init__

__init__(file_list, path, tile_size, crop_size)

---

method on_epoch_end

on_epoch_end()

---

This file was automatically generated via andrewplayer3's fork of lazydocs.