jupyter kernel dies on creating instance of `TSDatasets` class on M1 mac

[bellerofonte]

I am trying to launch sample notebooks on my M1 Max.

Unfortunately, I am constantly getting a dead kernel when it tries to instantiate TSDatasets

f.e. in the Time Series data preparation notebook kernel dies on executing:

# [25]
tfms  = [None, TSRegression()]
dsets = TSDatasets(X, y, tfms=tfms, splits=splits, inplace=True)      #  <-------------------
dsets

in the How to use Transformers with Time Series? notebook kernel dies on executing:

# [4]
bs = 64
n_epochs = 100
tfms  = [None, [Categorize()]]
dsets = TSDatasets(X, y, tfms=tfms, splits=splits, inplace=True)    #  <--------------------
dls   = TSDataLoaders.from_dsets(dsets.train, dsets.valid, bs=bs, batch_tfms=TSStandardize())
dls.show_batch()

I have tried to debug these cells but kernel dies before even entering inside those lines.

The Jupyter's log shows following:

15:23:43.736 [info] Restarted 716c6430-96ac-45d6-933f-e6322005f3ba
15:23:44.599 [info] Handle Execution of Cells 9,10,12,15,16,17,20,21,22,25,27,29,31 for ~/Projects/tsai-test/07_Time_Series_Classification_with_Transformers.ipynb
15:23:44.606 [info] Cell 9 completed in 0.002s (start: 1716297824604, end: 1716297824606)
15:23:46.900 [info] Cell 10 completed in 2.29s (start: 1716297824609, end: 1716297826899)
15:23:47.041 [info] Cell 12 completed in 0.139s (start: 1716297826902, end: 1716297827041)
15:23:47.708 [error] Disposing session as kernel process died ExitCode: undefined, Reason: 
15:23:47.729 [info] Cell 15 completed in -1716297827.053s (start: 1716297827053, end: undefined)

computer_setup() outputs following:

os              : macOS-14.4.1-arm64-arm-64bit
python          : 3.12.3
tsai            : 0.3.9
fastai          : 2.7.15
fastcore        : 1.5.38
torch           : 2.2.2
device          : mps
cpu cores       : 10
threads per cpu : 1
RAM             : 32.0 GB
GPU memory      : N/A

I am running these notebooks under conda env which has been created:

conda create -n tsai-py12 python=3.12
conda activate tsai-py12
conda install numpy pandas matplotlib scipy scikit-learn plotly seaborn ipykernel
python -m pip install tsai

The same issue when using python 3.10 and 3.11

Finally, I have tried to run plain torch example to see if it works on my M1 and it succeeded.

example

```python import torch import torch.nn as nn import torch.optim as optim import torch.nn.functional as F # Define the neural network class Net(nn.Module): def __init__(self): super(Net, self).__init__() self.fc1 = nn.Linear(784, 128) # Input layer (28x28 images flattened to 784 pixels) self.fc2 = nn.Linear(128, 64) # Hidden layer self.fc3 = nn.Linear(64, 10) # Output layer (10 classes for MNIST digits) def forward(self, x): x = x.view(-1, 784) # Flatten the input tensor x = F.relu(self.fc1(x)) x = F.relu(self.fc2(x)) x = self.fc3(x) return x # Load the MNIST dataset from torchvision import datasets, transforms transform = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.5,), (0.5,))]) trainset = datasets.MNIST('data/', download=True, train=True, transform=transform) testset = datasets.MNIST('data/', download=True, train=False, transform=transform) # Create data loaders trainloader = torch.utils.data.DataLoader(trainset, batch_size=64, shuffle=True) testloader = torch.utils.data.DataLoader(testset, batch_size=64, shuffle=False) # Create an instance of the neural network net = Net() # Define the loss function and optimizer criterion = nn.CrossEntropyLoss() optimizer = optim.SGD(net.parameters(), lr=0.01, momentum=0.9) # Train the network epochs = 5 for epoch in range(epochs): running_loss = 0.0 for inputs, labels in trainloader: optimizer.zero_grad() outputs = net(inputs) loss = criterion(outputs, labels) loss.backward() optimizer.step() running_loss += loss.item() print(f'Epoch {epoch+1}, Loss: {running_loss/len(trainloader):.4f}') # Test the network correct = 0 total = 0 with torch.no_grad(): for inputs, labels in testloader: outputs = net(inputs) _, predicted = torch.max(outputs.data, 1) total += labels.size(0) correct += (predicted == labels).sum().item() print(f'Accuracy on test set: {100 * correct / total:.2f}%') ```

Any ideas what's going wrong here?

[bellerofonte]

I have managed to fix that issue buy recreating python environment in a different way:

conda create -n tsai-py12 python=3.12
conda activate tsai-py12
conda install ipykernel ipywidgets
python -m pip install tsai
python -m pip uninstall scikit-learn
python -m pip install scikit-learn==1.4.0

The rollback to scikit-learn==1.4.0 was necessary because otherwise I got ImportError: cannot import name '_get_column_indices' from 'sklearn.utils' on calling from tsai.all import *