Ray Train / Ray Tune fails with Matplotlib

This simple script finishes training, but then the actor coordinating the training dies at the end, causing the fault tolerance engine in Ray to re-run, creating an endless loop.

It may be caused by the inclusion of matplotlib, but that doesn’t matter – data scientists will often have such code, and Ray Train shouldn’t cause it to explode.

System info

% which python
/Users/will/opt/anaconda3/envs/torch_to_train/bin/python
% which ipython
/Users/will/opt/anaconda3/envs/torch_to_train/bin/ipython
% python --version
Python 3.8.12
% ipython
In [1]: import ray
In [2]: ray.__version__
Out[2]: '1.9.2'

Code:

# -*- coding: utf-8 -*-
import torch
import torchvision
import torchvision.transforms as transforms
import matplotlib.pyplot as plt
import numpy as np
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
 
from ray import train
import ray.train.torch
from ray.train import Trainer
from ray.train.torch import TorchConfig
 
 
########################################################################
# The output of torchvision datasets are PILImage images of range [0, 1].
# We transform them to Tensors of normalized range [-1, 1].
 
 
########################################################################
# Let us show some of the training images, for fun.
 
 
def imshow(img):
   img = img / 2 + 0.5     # unnormalize
   npimg = img.numpy()
   plt.imshow(np.transpose(npimg, (1, 2, 0)))
   plt.show()
 
 
 
class Net(nn.Module):
   def __init__(self):
       super().__init__()
       self.conv1 = nn.Conv2d(3, 6, 5)
       self.pool = nn.MaxPool2d(2, 2)
       self.conv2 = nn.Conv2d(6, 16, 5)
       self.fc1 = nn.Linear(16 * 5 * 5, 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 = torch.flatten(x, 1) # flatten all dimensions except batch
       x = F.relu(self.fc1(x))
       x = F.relu(self.fc2(x))
       x = self.fc3(x)
       return x
 
 
def my_train_func(config):
   # setup data and transofmrers
   transform = transforms.Compose(
       [transforms.ToTensor(),
       transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))]
   )
 
   batch_size = 4
 
   trainset = torchvision.datasets.CIFAR10(root='./data', train=True,
                                           download=True, transform=transform)
   trainloader = torch.utils.data.DataLoader(trainset, batch_size=batch_size,
                                           shuffle=True, num_workers=2)
 
   testset = torchvision.datasets.CIFAR10(root='./data', train=False,
                                       download=True, transform=transform)
   testloader = torch.utils.data.DataLoader(testset, batch_size=batch_size,
                                           shuffle=False, num_workers=2)
 
   classes = ('plane', 'car', 'bird', 'cat',
           'deer', 'dog', 'frog', 'horse', 'ship', 'truck')
  
   # setup model
   net = Net()
   net = train.torch.prepare_model(net)
 
   # setup loss/optimizer
   criterion = nn.CrossEntropyLoss()
   optimizer = optim.SGD(net.parameters(), lr=0.001, momentum=0.9)
 
       # get some random training images
   dataiter = iter(trainloader)
   images, labels = dataiter.next()
   # # show images
   # imshow(torchvision.utils.make_grid(images))
   # # print labels
   # print(' '.join(f'{classes[labels[j]]:5s}' for j in range(batch_size)))
 
   for epoch in range(2):  # loop over the dataset multiple times
 
       running_loss = 0.0
       for i, data in enumerate(trainloader, 0):
           # get the inputs; data is a list of [inputs, labels]
           inputs, labels = data
 
           # zero the parameter gradients
           optimizer.zero_grad()
 
           # forward + backward + optimize
           outputs = net(inputs)
           loss = criterion(outputs, labels)
           loss.backward()
           optimizer.step()
 
           # print statistics
           running_loss += loss.item()
           if i % 2000 == 1999:    # print every 2000 mini-batches
               print(f'[{epoch + 1}, {i + 1:5d}] loss: {running_loss / 2000:.3f}')
               running_loss = 0.0
 
   print('Finished Training')
 
   ########################################################################
   # Let's quickly save our trained model:
 
   PATH = './cifar_net.pth'
   torch.save(net.state_dict(), PATH)
 
   ########################################################################
   # See `here <https://pytorch.org/docs/stable/notes/serialization.html>`_
   # for more details on saving PyTorch models.
   #
   # 5. Test the network on the test data
   # ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   #
   # We have trained the network for 2 passes over the training dataset.
   # But we need to check if the network has learnt anything at all.
   #
   # We will check this by predicting the class label that the neural network
   # outputs, and checking it against the ground-truth. If the prediction is
   # correct, we add the sample to the list of correct predictions.
   #
   # Okay, first step. Let us display an image from the test set to get familiar.
 
   dataiter = iter(testloader)
   images, labels = dataiter.next()
 
   # print images
   imshow(torchvision.utils.make_grid(images))
   print('GroundTruth: ', ' '.join(f'{classes[labels[j]]:5s}' for j in range(4)))
 
   ########################################################################
   # Next, let's load back in our saved model (note: saving and re-loading the model
   # wasn't necessary here, we only did it to illustrate how to do so):
 
   net = Net()
   net.load_state_dict(torch.load(PATH))
 
   ########################################################################
   # Okay, now let us see what the neural network thinks these examples above are:
 
   outputs = net(images)
 
   ########################################################################
   # The outputs are energies for the 10 classes.
   # The higher the energy for a class, the more the network
   # thinks that the image is of the particular class.
   # So, let's get the index of the highest energy:
   _, predicted = torch.max(outputs, 1)
 
   print('Predicted: ', ' '.join(f'{classes[predicted[j]]:5s}'
                               for j in range(4)))
 
   ########################################################################
   # The results seem pretty good.
   #
   # Let us look at how the network performs on the whole dataset.
 
   correct = 0
   total = 0
   # since we're not training, we don't need to calculate the gradients for our outputs
   with torch.no_grad():
       for data in testloader:
           images, labels = data
           # calculate outputs by running images through the network
           outputs = net(images)
           # the class with the highest energy is what we choose as prediction
           _, predicted = torch.max(outputs.data, 1)
           total += labels.size(0)
           correct += (predicted == labels).sum().item()
 
   print(f'Accuracy of the network on the 10000 test images: {100 * correct // total} %')
 
   ########################################################################
   # That looks way better than chance, which is 10% accuracy (randomly picking
   # a class out of 10 classes).
   # Seems like the network learnt something.
   #
   # Hmmm, what are the classes that performed well, and the classes that did
   # not perform well:
 
   # prepare to count predictions for each class
   correct_pred = {classname: 0 for classname in classes}
   total_pred = {classname: 0 for classname in classes}
 
   # again no gradients needed
   with torch.no_grad():
       for data in testloader:
           images, labels = data
           outputs = net(images)
           _, predictions = torch.max(outputs, 1)
           # collect the correct predictions for each class
           for label, prediction in zip(labels, predictions):
               if label == prediction:
                   correct_pred[classes[label]] += 1
               total_pred[classes[label]] += 1
 
 
   # print accuracy for each class
   for classname, correct_count in correct_pred.items():
       accuracy = 100 * float(correct_count) / total_pred[classname]
       print(f'Accuracy for class: {classname:5s} is {accuracy:.1f} %')
 
 
 
if __name__ == '__main__':
   config = {}
   trainer = Trainer(backend="torch", num_workers=4)
  
   trainer.start() # set up resources
   trainer.run(my_train_func, config=config)
   trainer.shutdown()

[matthewdeng]

I took a quick look, here’s a simple repro:

import numpy as np
import matplotlib.pyplot as plt
from ray.train import Trainer

def plot_func():
    arr = np.array([[1]])
    plt.imshow(arr)
    plt.show()

trainer = Trainer(backend="torch", num_workers=1)
trainer.start()
trainer.run(plot_func)

The root cause indeed seems to be because Ray Train executes the training function in a separate thread. This is because the main thread for each process is currently used to allow the Trainer (more specifically the WorkerGroup) to call ray.get to get the workers’ reported/checkpointed data.

Fixing this likely requires us to rearchitect the backend pretty significantly. While I do agree that this isn’t a great user experience, I’m not sure if we want to allocate 20+ hours to fix it at this moment.

As a short term solution, would it be reasonable to add an FAQ to the documentation and explain how to work around this - specifically to just run the plot outside of the Trainer? The user experience isn’t terrible as everything can still run in a single Python script (and by doing this you don’t end up creating num_worker plots).

[matthewdeng]

@worldveil I created a PR to add this to the docs. Please take a look!