Python catalyst data type conflict

Question

I have previously used the catalyst library to fit the classification model for images and had no issues, but now I would like to use it for "table" data, however, It seems that I am doing something wrong. Here is the snippet of the dataset:

      agent_1_feat_0  agent_1_feat_1  ...       E.T  target
0               58.8            85.1  ...  0.837398    True
1               44.8            71.1  ...  0.789474    True
2               46.3            70.8  ...  0.891566   False
3               50.2            77.5  ...  0.505263   False
4               44.9            75.0  ...  0.943396    True
...              ...             ...  ...       ...     ...
2275            55.1            82.8  ...  0.582090   False
2276            49.0            78.0  ...  0.943396    True
2277            46.4            76.5  ...  0.735714    True
2278            57.9            85.2  ...  0.837398    True
2279            43.9            70.0  ...  0.850877    True

[2280 rows x 98 columns]

Then I convert it to the DataLoader:

from torch.utils.data import DataLoader, TensorDataset
from sklearn.model_selection import train_test_split

train, val = train_test_split(a1, test_size=0.67)

tt = torch.tensor(train['target'].values.astype('float32'))
train = torch.tensor(train.drop(columns= 'target').values.astype('float32'))  
train_a1 = DataLoader(dataset = TensorDataset(train, tt), batch_size = 10, shuffle = True)

vt = torch.tensor(val['target'].values.astype('float32'))
val = torch.tensor(val.drop(columns= 'target').values.astype('float32'))  
val_a1 = DataLoader(dataset = TensorDataset(val, vt), batch_size = 10, shuffle = True)

len(train_a1), len(val_a1) # (76, 153)

Here is the model:

import torch.nn as nn
torch.manual_seed(69)

class Net(nn.Module):
    def __init__(self):
        super().__init__()

        self.lin = nn.Sequential(
            nn.Linear(in_features= train.shape[1], out_features= train.shape[1]),
            nn.Linear(in_features= train.shape[1], out_features= 50),
            nn.BatchNorm1d(50),
            nn.Linear(in_features= 50, out_features= 30),
            nn.Linear(in_features= 30, out_features= 30),
            nn.Linear(in_features= 30, out_features= 30),
            nn.Linear(in_features= 30, out_features= 15),
            nn.Linear(in_features= 15, out_features= 15),
            nn.Linear(in_features= 15, out_features= 5),
            nn.BatchNorm1d(5),
            nn.Dropout(p= .2),
            nn.ReLU(inplace=True), 
        )

        self.out = nn.Sequential(
            nn.Linear(in_features= 5, out_features= 2),
            nn.ReLU(inplace=True),
        )

    def forward(self, x):
        # x = x.view(x.size(0), -1)
        print('new', x.shape) # torch.Size([10, 97])
        x = self.lin(x)
        print('lin', x.shape) # torch.Size([10, 5])
        x = self.out(x)
        print('out', x.shape) # torch.Size([10, 2])
        
        return x

model = Net()

Now, what I create the trainer:

import torch.optim as optim

criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.parameters(), lr=0.001)

from catalyst import dl

runner = dl.SupervisedRunner(input_key="features", output_key="logits", target_key="targets", loss_key="loss")
runner.train(
    model= model,
    criterion= criterion,
    optimizer= optimizer,
    loaders= {"train": train_a1, "valid": val_a1},
    num_epochs= 1,
    callbacks=[
        dl.AccuracyCallback(input_key="logits", target_key="targets", topk_args=(1, 2)),
        dl.PrecisionRecallF1SupportCallback(
            input_key="logits", target_key="targets", num_classes=2
        ),
    ],
    logdir="./logs",
    valid_loader="valid",
    valid_metric="loss",
    minimize_valid_metric=True,
    verbose=False,
    load_best_on_end=True,
    seed= 69,
)

So, up to this point, everything works fine, but when I try to execute the last piece of code, I get the error: RuntimeError: Expected object of scalar type Long but got scalar type Float for argument #2 'target' in call to _thnn_nll_loss_forward. I have tried to fix it by setting dtype= torch.long when using torch.tensor but It then complains about the data type being Long :). How to solve this? I suspect the CrossEntropyLoss because it seems that it is not the best choice for 1d data, however, I have no Idea how tofix that.

Python catalyst data type conflict

0 Answers