fau-tools 2.1.1

Introduction

This is a small tool that uses the PyTorch framework, providing assistance in completing classification task using CNN.

Features: train model, print training process, save training files, plot figures, etc.

Install

pip install fau-tools

The helpers in fau_tools.torch_tools that read and plot a training record need two extra packages:

pip install "fau-tools[plot]"

Usage

import

The following code is recommended.

import fau_tools

quick start

The tutor will use a simple example to help you get started quickly!

The following example uses Fau-tools to train a model in MNIST hand-written digits dataset.

import torch
import torch.nn as nn
import torch.utils.data as tdata
import torchvision

import fau_tools


# A simple CNN network
class CNN(nn.Module):
  def __init__(self):
    super().__init__()
    self.conv = nn.Sequential(
      nn.Conv2d(1, 16, 3, 1, 1),  # -> (16, 28, 28)
      nn.ReLU(),
      nn.MaxPool2d(2),  # -> (16, 14, 14)

      nn.Conv2d(16, 32, 3, 1, 1),  # -> (32, 14, 14)
      nn.ReLU(),
      nn.MaxPool2d(2)  # -> (32, 7, 7)
    )
    self.output = nn.Linear(32 * 7 * 7, 10)


  def forward(self, x):
    x = self.conv(x)
    x = x.flatten(1)
    return self.output(x)


# Hyper Parameters definition
total_epoch = 10
lr = 1E-2
batch_size = 1024

# Load dataset
train_data      = torchvision.datasets.MNIST('datasets', True, torchvision.transforms.ToTensor(), download=True)
test_data       = torchvision.datasets.MNIST('datasets', False, torchvision.transforms.ToTensor())
train_data.data = train_data.data[:6000]  # mini data
test_data.data  = test_data.data[:2000]  # mini data

# Get data loader
train_loader = tdata.DataLoader(train_data, batch_size, True)
test_loader  = tdata.DataLoader(test_data, batch_size)

# Initialize model, optimizer and loss function
model = CNN()
loss_function = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(), lr)

# Train!
fau_tools.TaskRunner(model, train_loader, test_loader, loss_function, optimizer, total_epoch, exp_path="MNIST").train()

Now, we can run the python script, and the training process will be visualized as the following picture.

Three files named best.pth, scalars.csv and exp_info.txt will be saved.

The first file is the weight of trained model.

The second file records scalar value changes in the training process.

The third file saves information about the experiment.

The above is the primary usage of this tool, but there are also some other snazzy features, which will be introduced later. [TODO]

END

Hope you could like it! And welcome issues and pull requests.