.. DO NOT EDIT.
.. THIS FILE WAS AUTOMATICALLY GENERATED BY SPHINX-GALLERY.
.. TO MAKE CHANGES, EDIT THE SOURCE PYTHON FILE:
.. "beginner/introyt/trainingyt.py"
.. LINE NUMBERS ARE GIVEN BELOW.

.. only:: html

    .. note::
        :class: sphx-glr-download-link-note

        Click :ref:`here <sphx_glr_download_beginner_introyt_trainingyt.py>`
        to download the full example code

.. rst-class:: sphx-glr-example-title

.. _sphx_glr_beginner_introyt_trainingyt.py:


`Introduction <introyt1_tutorial.html>`_ ||
`Tensors <tensors_deeper_tutorial.html>`_ ||
`Autograd <autogradyt_tutorial.html>`_ ||
`Building Models <modelsyt_tutorial.html>`_ ||
`TensorBoard Support <tensorboardyt_tutorial.html>`_ ||
**Training Models** ||
`Model Understanding <captumyt.html>`_

Training with PyTorch
=====================

Follow along with the video below or on `youtube <https://www.youtube.com/watch?v=jF43_wj_DCQ>`__.

.. raw:: html

   <div style="margin-top:10px; margin-bottom:10px;">
     <iframe width="560" height="315" src="https://www.youtube.com/embed/jF43_wj_DCQ" frameborder="0" allow="accelerometer; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe>
   </div>

Introduction
------------

In past videos, we’ve discussed and demonstrated:

- Building models with the neural network layers and functions of the torch.nn module
- The mechanics of automated gradient computation, which is central to
  gradient-based model training 
- Using TensorBoard to visualize training progress and other activities

In this video, we’ll be adding some new tools to your inventory:

- We’ll get familiar with the dataset and dataloader abstractions, and how
  they ease the process of feeding data to your model during a training loop 
- We’ll discuss specific loss functions and when to use them
- We’ll look at PyTorch optimizers, which implement algorithms to adjust
  model weights based on the outcome of a loss function

Finally, we’ll pull all of these together and see a full PyTorch
training loop in action.


Dataset and DataLoader
----------------------
 
The ``Dataset`` and ``DataLoader`` classes encapsulate the process of
pulling your data from storage and exposing it to your training loop in
batches.

The ``Dataset`` is responsible for accessing and processing single
instances of data.
 
The ``DataLoader`` pulls instances of data from the ``Dataset`` (either
automatically or with a sampler that you define), collects them in
batches, and returns them for consumption by your training loop. The
``DataLoader`` works with all kinds of datasets, regardless of the type
of data they contain.
 
For this tutorial, we’ll be using the Fashion-MNIST dataset provided by
TorchVision. We use ``torchvision.transforms.Normalize()`` to
zero-center and normalize the distribution of the image tile content,
and download both training and validation data splits.

.. GENERATED FROM PYTHON SOURCE LINES 65-96

.. code-block:: default


    import torch
    import torchvision
    import torchvision.transforms as transforms

    # PyTorch TensorBoard support
    from torch.utils.tensorboard import SummaryWriter
    from datetime import datetime


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

    # Create datasets for training & validation, download if necessary
    training_set = torchvision.datasets.FashionMNIST('./data', train=True, transform=transform, download=True)
    validation_set = torchvision.datasets.FashionMNIST('./data', train=False, transform=transform, download=True)

    # Create data loaders for our datasets; shuffle for training, not for validation
    training_loader = torch.utils.data.DataLoader(training_set, batch_size=4, shuffle=True)
    validation_loader = torch.utils.data.DataLoader(validation_set, batch_size=4, shuffle=False)

    # Class labels
    classes = ('T-shirt/top', 'Trouser', 'Pullover', 'Dress', 'Coat',
            'Sandal', 'Shirt', 'Sneaker', 'Bag', 'Ankle Boot')

    # Report split sizes
    print('Training set has {} instances'.format(len(training_set)))
    print('Validation set has {} instances'.format(len(validation_set)))






.. rst-class:: sphx-glr-script-out

 .. code-block:: none


      0%|          | 0.00/26.4M [00:00<?, ?B/s]
      0%|          | 32.8k/26.4M [00:00<01:53, 232kB/s]
      0%|          | 65.5k/26.4M [00:00<01:54, 230kB/s]
      0%|          | 98.3k/26.4M [00:00<01:54, 229kB/s]
      1%|          | 197k/26.4M [00:00<01:04, 409kB/s] 
      1%|1         | 360k/26.4M [00:00<00:38, 674kB/s]
      3%|2         | 721k/26.4M [00:00<00:19, 1.30MB/s]
      5%|5         | 1.44M/26.4M [00:01<00:09, 2.51MB/s]
     11%|#         | 2.85M/26.4M [00:01<00:04, 4.84MB/s]
     19%|#9        | 5.14M/26.4M [00:01<00:02, 8.32MB/s]
     31%|###1      | 8.29M/26.4M [00:01<00:01, 12.5MB/s]
     43%|####3     | 11.4M/26.4M [00:01<00:00, 15.3MB/s]
     55%|#####5    | 14.5M/26.4M [00:01<00:00, 17.3MB/s]
     66%|######6   | 17.5M/26.4M [00:01<00:00, 18.3MB/s]
     78%|#######8  | 20.6M/26.4M [00:02<00:00, 19.3MB/s]
     89%|########9 | 23.6M/26.4M [00:02<00:00, 19.7MB/s]
    100%|##########| 26.4M/26.4M [00:02<00:00, 11.9MB/s]

      0%|          | 0.00/29.5k [00:00<?, ?B/s]
    100%|##########| 29.5k/29.5k [00:00<00:00, 186kB/s]
    100%|##########| 29.5k/29.5k [00:00<00:00, 186kB/s]

      0%|          | 0.00/4.42M [00:00<?, ?B/s]
      1%|          | 32.8k/4.42M [00:00<00:23, 185kB/s]
      1%|1         | 65.5k/4.42M [00:00<00:23, 185kB/s]
      2%|2         | 98.3k/4.42M [00:00<00:23, 185kB/s]
      5%|5         | 229k/4.42M [00:00<00:10, 403kB/s] 
     10%|9         | 426k/4.42M [00:00<00:05, 670kB/s]
     20%|##        | 885k/4.42M [00:01<00:02, 1.30MB/s]
     39%|###9      | 1.74M/4.42M [00:01<00:01, 2.49MB/s]
     79%|#######8  | 3.47M/4.42M [00:01<00:00, 4.83MB/s]
    100%|##########| 4.42M/4.42M [00:01<00:00, 3.11MB/s]

      0%|          | 0.00/5.15k [00:00<?, ?B/s]
    100%|##########| 5.15k/5.15k [00:00<00:00, 26.1MB/s]
    Training set has 60000 instances
    Validation set has 10000 instances




.. GENERATED FROM PYTHON SOURCE LINES 97-99

As always, let’s visualize the data as a sanity check:


.. GENERATED FROM PYTHON SOURCE LINES 99-123

.. code-block:: default


    import matplotlib.pyplot as plt
    import numpy as np

    # Helper function for inline image display
    def matplotlib_imshow(img, one_channel=False):
        if one_channel:
            img = img.mean(dim=0)
        img = img / 2 + 0.5     # unnormalize
        npimg = img.numpy()
        if one_channel:
            plt.imshow(npimg, cmap="Greys")
        else:
            plt.imshow(np.transpose(npimg, (1, 2, 0)))

    dataiter = iter(training_loader)
    images, labels = next(dataiter)

    # Create a grid from the images and show them
    img_grid = torchvision.utils.make_grid(images)
    matplotlib_imshow(img_grid, one_channel=True)
    print('  '.join(classes[labels[j]] for j in range(4)))





.. image-sg:: /beginner/introyt/images/sphx_glr_trainingyt_001.png
   :alt: trainingyt
   :srcset: /beginner/introyt/images/sphx_glr_trainingyt_001.png
   :class: sphx-glr-single-img


.. rst-class:: sphx-glr-script-out

 .. code-block:: none

    Coat  Sneaker  Sandal  Dress




.. GENERATED FROM PYTHON SOURCE LINES 124-130

The Model
---------

The model we’ll use in this example is a variant of LeNet-5 - it should
be familiar if you’ve watched the previous videos in this series.


.. GENERATED FROM PYTHON SOURCE LINES 130-158

.. code-block:: default


    import torch.nn as nn
    import torch.nn.functional as F

    # PyTorch models inherit from torch.nn.Module
    class GarmentClassifier(nn.Module):
        def __init__(self):
            super(GarmentClassifier, self).__init__()
            self.conv1 = nn.Conv2d(1, 6, 5)
            self.pool = nn.MaxPool2d(2, 2)
            self.conv2 = nn.Conv2d(6, 16, 5)
            self.fc1 = nn.Linear(16 * 4 * 4, 120)
            self.fc2 = nn.Linear(120, 84)
            self.fc3 = nn.Linear(84, 10)

        def forward(self, x):
            x = self.pool(F.relu(self.conv1(x)))
            x = self.pool(F.relu(self.conv2(x)))
            x = x.view(-1, 16 * 4 * 4)
            x = F.relu(self.fc1(x))
            x = F.relu(self.fc2(x))
            x = self.fc3(x)
            return x
    

    model = GarmentClassifier()









.. GENERATED FROM PYTHON SOURCE LINES 159-166

Loss Function
-------------

For this example, we’ll be using a cross-entropy loss. For demonstration
purposes, we’ll create batches of dummy output and label values, run
them through the loss function, and examine the result.


.. GENERATED FROM PYTHON SOURCE LINES 166-182

.. code-block:: default


    loss_fn = torch.nn.CrossEntropyLoss()

    # NB: Loss functions expect data in batches, so we're creating batches of 4
    # Represents the model's confidence in each of the 10 classes for a given input
    dummy_outputs = torch.rand(4, 10)
    # Represents the correct class among the 10 being tested
    dummy_labels = torch.tensor([1, 5, 3, 7])
    
    print(dummy_outputs)
    print(dummy_labels)

    loss = loss_fn(dummy_outputs, dummy_labels)
    print('Total loss for this batch: {}'.format(loss.item()))






.. rst-class:: sphx-glr-script-out

 .. code-block:: none

    tensor([[0.1361, 0.2725, 0.0907, 0.0082, 0.4228, 0.7388, 0.4125, 0.5239, 0.7478,
             0.2745],
            [0.9673, 0.0197, 0.7789, 0.9434, 0.9163, 0.0952, 0.0651, 0.4475, 0.3295,
             0.4615],
            [0.0884, 0.4782, 0.7056, 0.8047, 0.7331, 0.6193, 0.0116, 0.1595, 0.9667,
             0.9834],
            [0.9127, 0.2350, 0.8619, 0.5442, 0.4277, 0.9194, 0.4962, 0.0656, 0.8193,
             0.5803]])
    tensor([1, 5, 3, 7])
    Total loss for this batch: 2.5408973693847656




.. GENERATED FROM PYTHON SOURCE LINES 183-200

Optimizer
---------

For this example, we’ll be using simple `stochastic gradient
descent <https://pytorch.org/docs/stable/optim.html>`__ with momentum.

It can be instructive to try some variations on this optimization
scheme:

- Learning rate determines the size of the steps the optimizer
  takes. What does a different learning rate do to the your training
  results, in terms of accuracy and convergence time?
- Momentum nudges the optimizer in the direction of strongest gradient over
  multiple steps. What does changing this value do to your results? 
- Try some different optimization algorithms, such as averaged SGD, Adagrad, or
  Adam. How do your results differ?


.. GENERATED FROM PYTHON SOURCE LINES 200-205

.. code-block:: default


    # Optimizers specified in the torch.optim package
    optimizer = torch.optim.SGD(model.parameters(), lr=0.001, momentum=0.9)









.. GENERATED FROM PYTHON SOURCE LINES 206-225

The Training Loop
-----------------

Below, we have a function that performs one training epoch. It
enumerates data from the DataLoader, and on each pass of the loop does
the following:

- Gets a batch of training data from the DataLoader
- Zeros the optimizer’s gradients 
- Performs an inference - that is, gets predictions from the model for an input batch
- Calculates the loss for that set of predictions vs. the labels on the dataset
- Calculates the backward gradients over the learning weights
- Tells the optimizer to perform one learning step - that is, adjust the model’s
  learning weights based on the observed gradients for this batch, according to the
  optimization algorithm we chose
- It reports on the loss for every 1000 batches.
- Finally, it reports the average per-batch loss for the last
  1000 batches, for comparison with a validation run


.. GENERATED FROM PYTHON SOURCE LINES 225-262

.. code-block:: default


    def train_one_epoch(epoch_index, tb_writer):
        running_loss = 0.
        last_loss = 0.
    
        # Here, we use enumerate(training_loader) instead of
        # iter(training_loader) so that we can track the batch
        # index and do some intra-epoch reporting
        for i, data in enumerate(training_loader):
            # Every data instance is an input + label pair
            inputs, labels = data
        
            # Zero your gradients for every batch!
            optimizer.zero_grad()
        
            # Make predictions for this batch
            outputs = model(inputs)
        
            # Compute the loss and its gradients
            loss = loss_fn(outputs, labels)
            loss.backward()
        
            # Adjust learning weights
            optimizer.step()
        
            # Gather data and report
            running_loss += loss.item()
            if i % 1000 == 999:
                last_loss = running_loss / 1000 # loss per batch
                print('  batch {} loss: {}'.format(i + 1, last_loss))
                tb_x = epoch_index * len(training_loader) + i + 1
                tb_writer.add_scalar('Loss/train', last_loss, tb_x)
                running_loss = 0.
            
        return last_loss









.. GENERATED FROM PYTHON SOURCE LINES 263-276

Per-Epoch Activity
~~~~~~~~~~~~~~~~~~

There are a couple of things we’ll want to do once per epoch: 

- Perform validation by checking our relative loss on a set of data that was not
  used for training, and report this 
- Save a copy of the model

Here, we’ll do our reporting in TensorBoard. This will require going to
the command line to start TensorBoard, and opening it in another browser
tab.


.. GENERATED FROM PYTHON SOURCE LINES 276-326

.. code-block:: default


    # Initializing in a separate cell so we can easily add more epochs to the same run
    timestamp = datetime.now().strftime('%Y%m%d_%H%M%S')
    writer = SummaryWriter('runs/fashion_trainer_{}'.format(timestamp))
    epoch_number = 0

    EPOCHS = 5

    best_vloss = 1_000_000.

    for epoch in range(EPOCHS):
        print('EPOCH {}:'.format(epoch_number + 1))
    
        # Make sure gradient tracking is on, and do a pass over the data
        model.train(True)
        avg_loss = train_one_epoch(epoch_number, writer)
    

        running_vloss = 0.0
        # Set the model to evaluation mode, disabling dropout and using population 
        # statistics for batch normalization.
        model.eval()

        # Disable gradient computation and reduce memory consumption.
        with torch.no_grad():
            for i, vdata in enumerate(validation_loader):
                vinputs, vlabels = vdata
                voutputs = model(vinputs)
                vloss = loss_fn(voutputs, vlabels)
                running_vloss += vloss
    
        avg_vloss = running_vloss / (i + 1)
        print('LOSS train {} valid {}'.format(avg_loss, avg_vloss))
    
        # Log the running loss averaged per batch
        # for both training and validation
        writer.add_scalars('Training vs. Validation Loss',
                        { 'Training' : avg_loss, 'Validation' : avg_vloss },
                        epoch_number + 1)
        writer.flush()
    
        # Track best performance, and save the model's state
        if avg_vloss < best_vloss:
            best_vloss = avg_vloss
            model_path = 'model_{}_{}'.format(timestamp, epoch_number)
            torch.save(model.state_dict(), model_path)
    
        epoch_number += 1






.. rst-class:: sphx-glr-script-out

 .. code-block:: none

    EPOCH 1:
      batch 1000 loss: 1.859723868340254
      batch 2000 loss: 0.7959158919476904
      batch 3000 loss: 0.6875963613968342
      batch 4000 loss: 0.5840691136796958
      batch 5000 loss: 0.601861707421951
      batch 6000 loss: 0.5326420218963176
      batch 7000 loss: 0.5101273439126089
      batch 8000 loss: 0.4920893395068124
      batch 9000 loss: 0.45734640963794665
      batch 10000 loss: 0.4638966364953085
      batch 11000 loss: 0.4171614876713138
      batch 12000 loss: 0.4280265563330031
      batch 13000 loss: 0.4263071584069985
      batch 14000 loss: 0.4181580100507708
      batch 15000 loss: 0.42254892712844594
    LOSS train 0.42254892712844594 valid 0.4007205665111542
    EPOCH 2:
      batch 1000 loss: 0.3819776755711646
      batch 2000 loss: 0.3976921703386761
      batch 3000 loss: 0.37814622786745894
      batch 4000 loss: 0.34922556551004524
      batch 5000 loss: 0.34570265819173074
      batch 6000 loss: 0.3456719932517153
      batch 7000 loss: 0.34924780977396586
      batch 8000 loss: 0.35666742020787207
      batch 9000 loss: 0.33993772263024585
      batch 10000 loss: 0.3685673026727891
      batch 11000 loss: 0.3419286552860576
      batch 12000 loss: 0.33543308569528746
      batch 13000 loss: 0.33815630553881054
      batch 14000 loss: 0.3307116681025509
      batch 15000 loss: 0.3606875884762267
    LOSS train 0.3606875884762267 valid 0.3472764194011688
    EPOCH 3:
      batch 1000 loss: 0.30639552796969655
      batch 2000 loss: 0.3228924395108479
      batch 3000 loss: 0.31235727290814974
      batch 4000 loss: 0.3145854706429818
      batch 5000 loss: 0.29222741585151746
      batch 6000 loss: 0.31345307654022325
      batch 7000 loss: 0.3163325209467439
      batch 8000 loss: 0.31647060124657583
      batch 9000 loss: 0.3141732101450179
      batch 10000 loss: 0.299438458041157
      batch 11000 loss: 0.30074762070312866
      batch 12000 loss: 0.29664640293602135
      batch 13000 loss: 0.3008534389541455
      batch 14000 loss: 0.3202067443535416
      batch 15000 loss: 0.3115060192462843
    LOSS train 0.3115060192462843 valid 0.33710092306137085
    EPOCH 4:
      batch 1000 loss: 0.27951306609585846
      batch 2000 loss: 0.29738874051375025
      batch 3000 loss: 0.28281411081092667
      batch 4000 loss: 0.28956373607418934
      batch 5000 loss: 0.28590615158868604
      batch 6000 loss: 0.2835932457768722
      batch 7000 loss: 0.27401943222882985
      batch 8000 loss: 0.2870593838140994
      batch 9000 loss: 0.27542450427323456
      batch 10000 loss: 0.28151207812947177
      batch 11000 loss: 0.27685102666457534
      batch 12000 loss: 0.2891393224728527
      batch 13000 loss: 0.2877112180689146
      batch 14000 loss: 0.281157614742333
      batch 15000 loss: 0.2708676661506033
    LOSS train 0.2708676661506033 valid 0.30610325932502747
    EPOCH 5:
      batch 1000 loss: 0.2748342432942154
      batch 2000 loss: 0.2569611764227302
      batch 3000 loss: 0.24103483629236688
      batch 4000 loss: 0.2569761824092675
      batch 5000 loss: 0.2605063391393851
      batch 6000 loss: 0.2804330715298693
      batch 7000 loss: 0.25689922451737585
      batch 8000 loss: 0.2736012614666906
      batch 9000 loss: 0.2698981114967537
      batch 10000 loss: 0.26709053535146543
      batch 11000 loss: 0.2632426246944069
      batch 12000 loss: 0.27027270019240224
      batch 13000 loss: 0.2651329760397275
      batch 14000 loss: 0.280449686764925
      batch 15000 loss: 0.2526728889870938
    LOSS train 0.2526728889870938 valid 0.28390395641326904




.. GENERATED FROM PYTHON SOURCE LINES 327-369

To load a saved version of the model:

.. code:: python

    saved_model = GarmentClassifier()
    saved_model.load_state_dict(torch.load(PATH))

Once you’ve loaded the model, it’s ready for whatever you need it for -
more training, inference, or analysis.

Note that if your model has constructor parameters that affect model
structure, you’ll need to provide them and configure the model
identically to the state in which it was saved.

Other Resources
---------------

-  Docs on the `data
   utilities <https://pytorch.org/docs/stable/data.html>`__, including
   Dataset and DataLoader, at pytorch.org
-  A `note on the use of pinned
   memory <https://pytorch.org/docs/stable/notes/cuda.html#cuda-memory-pinning>`__
   for GPU training
-  Documentation on the datasets available in
   `TorchVision <https://pytorch.org/vision/stable/datasets.html>`__,
   `TorchText <https://pytorch.org/text/stable/datasets.html>`__, and
   `TorchAudio <https://pytorch.org/audio/stable/datasets.html>`__
-  Documentation on the `loss
   functions <https://pytorch.org/docs/stable/nn.html#loss-functions>`__
   available in PyTorch
-  Documentation on the `torch.optim
   package <https://pytorch.org/docs/stable/optim.html>`__, which
   includes optimizers and related tools, such as learning rate
   scheduling
-  A detailed `tutorial on saving and loading
   models <https://pytorch.org/tutorials/beginner/saving_loading_models.html>`__
-  The `Tutorials section of
   pytorch.org <https://pytorch.org/tutorials/>`__ contains tutorials on
   a broad variety of training tasks, including classification in
   different domains, generative adversarial networks, reinforcement
   learning, and more 



.. rst-class:: sphx-glr-timing

   **Total running time of the script:** ( 27 minutes  16.034 seconds)


.. _sphx_glr_download_beginner_introyt_trainingyt.py:

.. only:: html

  .. container:: sphx-glr-footer sphx-glr-footer-example


    .. container:: sphx-glr-download sphx-glr-download-python

      :download:`Download Python source code: trainingyt.py <trainingyt.py>`

    .. container:: sphx-glr-download sphx-glr-download-jupyter

      :download:`Download Jupyter notebook: trainingyt.ipynb <trainingyt.ipynb>`


.. only:: html

 .. rst-class:: sphx-glr-signature

    `Gallery generated by Sphinx-Gallery <https://sphinx-gallery.github.io>`_