IndexError target is out of bounds - neural-network

Question

I'm working on a custom dataset of images and using a Neural Net to classify them.
The data set is about 6000 images of 58 classes. But on training I keep getting a "target is out of bounds" error.
I've double checked the number of classes and image size but still get the same error.
#hyperprams
learning_rate = 5e-4
#3 for RGB values
in_channel = 3
#classes from data set
num_classes = 58
# arbitray choice
batch_size = 32
#total number of epochs used to train the model
epochs = 3
traffic_dataset = TrafficSigns(csv_file='annotations.csv',
root_directory='/Users/*****/Desktop/images/',
transform = transforms.ToTensor())
train_size = int(0.8 * len(traffic_dataset))
test_size = len(traffic_dataset) - train_size
train, test = torch.utils.data.random_split(traffic_dataset,
[train_size, test_size])
train_loader = torch.utils.data.DataLoader(train,
batch_size= batch_size,
shuffle= True,
num_workers= 4)
test_loader = torch.utils.data.DataLoader(test,
batch_size = batch_size,
shuffle= True,
num_workers= 4)
#Create a fully connected nn
class Net(nn.Module):
#use the constructor w/ arguments size of data and number of classes
def __init__(self,
input_size,
num_classes):
super(Net, self).__init__()
self.fc1 = nn.Linear(input_size, 60)
self.fc2 = nn.Linear(60, num_classes)
#define your forward step function with relu as the non-linear function of the weights
#x will be the datapassed to the model
def forward(self, x):
x=f.relu(self.fc1(x))
x = self.fc2(x)
return x
#sanity check
test = Net(2028, num_classes)
x = torch.randn(24, 2028)
print(test(x).shape)
#instantiate the class object of NN
net = Net(2028, num_classes)
criterion = nn.CrossEntropyLoss()
nn_optimizer = optim.Adam(net.parameters(),
lr = learning_rate)
#train on multiple epochs using the criterion and gradient decent algorthim estabilished above
for epoch in range(1):
for i, (data, target) in enumerate(tqdm.tqdm(train_loader)):
data = data.reshape(data.shape[0], -1)
#forward
outputs = net(data)
loss = criterion(outputs, target)
#backward propigation
nn_optimizer.zero_grad()
loss.backward()
#gradiant decent choosen
nn_optimizer.step()
Im also using a custom dataset class to import the images and labels.
My first thought was that the class is not iterating over the CSV and images correctly but I can't seem to find where they might be not matching up.
class TrafficSigns(Dataset):
#constructure will need csv file of labels images and the transform function defined above
def __init__(self,
csv_file,
root_directory,
transform = None):
self.labels = pd.read_csv(csv_file)
self.root_directory = root_directory
self.transform = transform
#returns the length
def __len__(self):
return len(self.labels)
#get data index by indes
def __getitem__(self, i):
image_path = os.path.join(self.root_directory, self.labels.iloc[i,0])
image = io.imread(image_path)
y_label = torch.tensor(int(self.labels.iloc[i, 1]))
#if statement needed since transform can be set to None
if self.transform:
image = self.transform(image)
return (image, y_label)
Any help would be awesome, thank you.
Here is the full stacktrace error that's getting thrown.
IndexError Traceback (most recent call last)
/var/folders/t_/rcfcs8g56jn7trwnsvmdyh_r0000gn/T/ipykernel_34551/1839343274.py in <module>
11 #forward
12 outputs = net(data)
---> 13 loss = criterion(outputs, target)
14 #backward propigation
15 nn_optimizer.zero_grad()
~/Library/Python/3.8/lib/python/site-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
1100 if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
1101 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1102 return forward_call(*input, **kwargs)
1103 # Do not call functions when jit is used
1104 full_backward_hooks, non_full_backward_hooks = [], []
~/Library/Python/3.8/lib/python/site-packages/torch/nn/modules/loss.py in forward(self, input, target)
1148
1149 def forward(self, input: Tensor, target: Tensor) -> Tensor:
-> 1150 return F.cross_entropy(input, target, weight=self.weight,
1151 ignore_index=self.ignore_index, reduction=self.reduction,
1152 label_smoothing=self.label_smoothing)
~/Library/Python/3.8/lib/python/site-packages/torch/nn/functional.py in cross_entropy(input, target, weight, size_average, ignore_index, reduce, reduction, label_smoothing)
2844 if size_average is not None or reduce is not None:
2845 reduction = _Reduction.legacy_get_string(size_average, reduce)
-> 2846 return torch._C._nn.cross_entropy_loss(input, target, weight, _Reduction.get_enum(reduction), ignore_index, label_smoothing)
2847
2848
IndexError: Target 125 is out of bounds.

Answer 1

I came across same issue where I used sequential model (LSTM) for next sequence prediction. I check data loader where labels contained -1 because of which cross entropy loss throwing exception. here is my sequence chunks where model found -1 sequence as label in data loader:
Solved please check your null rows and remove those or set accordingly.