Edit

Share via

Train models with PyTorch in Microsoft Fabric

  • Article

This article describes how to train and track the iterations of a PyTorch model. The PyTorch machine learning framework is based on the Torch library. PyTorch is often used for computer vision and natural language processing applications.

Prerequisites

Install PyTorch and torchvision within your notebook. You can install or upgrade the version of these libraries on your environment by using the following command:

pip install torch torchvision

Set up the machine learning experiment

You can create a machine learning experiment by using the MLFLow API. The MLflow set_experiment() function creates a new machine learning experiment named sample-pytorch, if it doesn't already exist.

Run the following code in your notebook and create the experiment:

import mlflow

mlflow.set_experiment("sample-pytorch")

Train and evaluate a Pytorch model

After you set up the experiment, you load the Modified National Institute of Standards and Technology (MNIST) dataset. You generate the test and training datasets, and then create a training function.

Run the following code in your notebook and train the Pytorch model:

import os
import torch
import torch.nn as nn
from torch.autograd import Variable
import torchvision.datasets as dset
import torchvision.transforms as transforms
import torch.nn.functional as F
import torch.optim as optim

# Load the MNIST dataset
root = "/tmp/mnist"
if not os.path.exists(root):
    os.mkdir(root)

trans = transforms.Compose(
    [transforms.ToTensor(), transforms.Normalize((0.5,), (1.0,))]
)

# If the data doesn't exist, download the MNIST dataset
train_set = dset.MNIST(root=root, train=True, transform=trans, download=True)
test_set = dset.MNIST(root=root, train=False, transform=trans, download=True)

batch_size = 100

train_loader = torch.utils.data.DataLoader(
    dataset=train_set, batch_size=batch_size, shuffle=True
)
test_loader = torch.utils.data.DataLoader(
    dataset=test_set, batch_size=batch_size, shuffle=False
) 

print("==>>> total trainning batch number: {}".format(len(train_loader)))
print("==>>> total testing batch number: {}".format(len(test_loader)))

# Define the network
class LeNet(nn.Module):
    def __init__(self):
        super(LeNet, self).__init__()
        self.conv1 = nn.Conv2d(1, 20, 5, 1)
        self.conv2 = nn.Conv2d(20, 50, 5, 1)
        self.fc1 = nn.Linear(4 * 4 * 50, 500)
        self.fc2 = nn.Linear(500, 10)

    def forward(self, x): 
        x = F.relu(self.conv1(x))
        x = F.max_pool2d(x, 2, 2)
        x = F.relu(self.conv2(x))
        x = F.max_pool2d(x, 2, 2)
        x = x.view(-1, 4 * 4 * 50)
        x = F.relu(self.fc1(x))
        x = self.fc2(x)
        return x

    def name(self):
        return "LeNet"

# Train the model
model = LeNet()

optimizer = optim.SGD(model.parameters(), lr=0.01, momentum=0.9)

criterion = nn.CrossEntropyLoss()

for epoch in range(1):
    # Model training
    ave_loss = 0
    for batch_idx, (x, target) in enumerate(train_loader):
        optimizer.zero_grad()
        x, target = Variable(x), Variable(target)
        out = model(x)
        loss = criterion(out, target)
        ave_loss = (ave_loss * batch_idx + loss.item()) / (batch_idx + 1)
        loss.backward()
        optimizer.step()
        if (batch_idx + 1) % 100 == 0 or (batch_idx + 1) == len(train_loader):
            print(
                "==>>> epoch: {}, batch index: {}, train loss: {:.6f}".format(
                    epoch, batch_idx + 1, ave_loss
                )
            )
    # Model testing
    correct_cnt, total_cnt, ave_loss = 0, 0, 0
    for batch_idx, (x, target) in enumerate(test_loader):
        x, target = Variable(x, volatile=True), Variable(target, volatile=True)
        out = model(x)
        loss = criterion(out, target)
        _, pred_label = torch.max(out.data, 1)
        total_cnt += x.data.size()[0]
        correct_cnt += (pred_label == target.data).sum()
        ave_loss = (ave_loss * batch_idx + loss.item()) / (batch_idx + 1)

        if (batch_idx + 1) % 100 == 0 or (batch_idx + 1) == len(test_loader):
            print(
                "==>>> epoch: {}, batch index: {}, test loss: {:.6f}, acc: {:.3f}".format(
                    epoch, batch_idx + 1, ave_loss, correct_cnt * 1.0 / total_cnt
                )
            )

torch.save(model.state_dict(), model.name())

Log model with MLflow

The next task starts an MLflow run and tracks the results within the machine learning experiment. The sample code creates a new model named sample-pytorch. It creates a run with the specified parameters, and logs the run within the sample-pytorch experiment.

Run the following code in your notebook and log the model:

with mlflow.start_run() as run:
    print("log pytorch model:")
    mlflow.pytorch.log_model(
        model, "pytorch-model", registered_model_name="sample-pytorch"
    )

    model_uri = "runs:/{}/pytorch-model".format(run.info.run_id)
    print("Model saved in run %s" % run.info.run_id)
    print(f"Model URI: {model_uri}")

Load and evaluate the model

After you save the model, you can load it for inferencing.

Run the following code in your notebook and load the model for inferencing:

# Inference with loading the logged model
loaded_model = mlflow.pytorch.load_model(model_uri)
print(type(loaded_model))

correct_cnt, total_cnt, ave_loss = 0, 0, 0
for batch_idx, (x, target) in enumerate(test_loader):
    x, target = Variable(x, volatile=True), Variable(target, volatile=True)
    out = loaded_model(x)
    loss = criterion(out, target)
    _, pred_label = torch.max(out.data, 1)
    total_cnt += x.data.size()[0]
    correct_cnt += (pred_label == target.data).sum()
    ave_loss = (ave_loss * batch_idx + loss.item()) / (batch_idx + 1)

    if (batch_idx + 1) % 100 == 0 or (batch_idx + 1) == len(test_loader):
        print(
            "==>>> epoch: {}, batch index: {}, test loss: {:.6f}, acc: {:.3f}".format(
                epoch, batch_idx + 1, ave_loss, correct_cnt * 1.0 / total_cnt
            )
        )

Related content