Skip to content

Latest commit

 

History

History
87 lines (59 loc) · 3.56 KB

README.md

File metadata and controls

87 lines (59 loc) · 3.56 KB

PyTorch Training Framework README

This Folder contains a PyTorch training framework for a specific machine learning task. The code is organized into several modules and provides functionality for training a neural network model, saving and loading model checkpoints, and reporting training progress.

Table of Contents

Introduction

This PyTorch training framework is designed for a specific machine learning task. It includes the following key components:

  • AverageMeter: A utility class for computing and storing the average and current values.
  • save_model: A function to save the model checkpoint, including the model's state dictionary and optimizer state if provided.
  • load_model: A function to load a saved model checkpoint, including the model's state dictionary and optimizer state if available.
  • normal_accuracy: A function to calculate accuracy between predictions and labels.
  • teacher_forcing_decay: A function to calculate the teacher forcing ratio for a given epoch during training.
  • train: The main training function that performs model training and reports training progress.

Requirements

Before using this code, make sure you have the following dependencies installed:

  • Python 3.x
  • PyTorch
  • pandas
  • tqdm

You can install the required packages using pip:

pip install torch pandas tqdm

Usage

To use this training framework, you can follow these steps:

  1. Define your neural network model in PyTorch.
  2. Prepare your training data as a PyTorch tensor.
  3. Set up the training hyperparameters and configurations.
  4. Call the train function with the necessary parameters to start training.

Here is an example of how to use this framework:

# Import the required modules and classes
import os
import torch
import time
import Loss
import torch.nn as nn
import pandas as pd
from torch.optim import lr_scheduler
from tqdm import tqdm

# Define your neural network model here

# Prepare your training data as a PyTorch tensor

# Set up training hyperparameters and configurations

# Call the train function
train(data_tensor, prediction_input_size, prediction_horizon, _divition_factr, model, model_name, epochs, load_saved_model, ckpt_save_freq, ckpt_save_path, ckpt_path, report_path, criterion, optimizer, optimizer_koopman, lr_scheduler, sleep_time, Validation_save_threshold, device, if_validation)

Training

The train function is the core of this framework. It takes various parameters to control the training process, including the model, training data, optimizer, and more. The training loop is performed for the specified number of epochs, and the progress is reported using a Pandas DataFrame.

Model Checkpoints

You can save and load model checkpoints using the save_model and load_model functions. Model checkpoints include the model's state dictionary and optimizer state if provided. This allows you to resume training from a saved checkpoint or use a trained model for inference.

Reports

Training progress is reported in a Pandas DataFrame. The report includes information such as the model name, training mode, epoch, learning rate, batch size, batch index, loss, and average loss. This report is saved as a CSV file for later analysis.

Feel free to customize and adapt this framework to your specific machine learning task. Happy training!


Please make sure to customize the usage section with your actual code, model, data, and training parameters.