Mat1 and mat2 shapes cannot be multiplied (64x3201 and 64x64)

vision
1

I need some help. I’m having some problems with my code

class NLPModel(nn.Module):
    def __init__(self, num_users, num_books):
        super(NLPModel, self).__init__()
        self.user_embedding = nn.Embedding(num_users+1, 32)
        self.book_embedding = nn.Embedding(num_books+1, 32)
        self.concatenation_layer = nn.Linear(64, 32*2)
        self.review_embedding = nn.Sequential(
            nn.Embedding(5000, 32),
            nn.Flatten()
        )
        #self.dot_product = nn.Sequential(
            #nn.Flatten(),
            #nn.Linear(32*2, 1)
        #)
        self.dot_product = nn.Sequential(
            nn.Linear(32*2, 32*2),
            nn.ReLU(),
            nn.Dropout(0.5),
            nn.Linear(32*2, 1)
        )
        self.concat = nn.Sequential(
            nn.Linear(32+32, 64),
            nn.ReLU(),
            nn.Dropout(0.5),
            nn.Linear(64, 32),
            nn.ReLU(),
            nn.Dropout(0.5),
            nn.Linear(32, 1)
        )
    
    def forward(self, user_input, book_input, review_input):
       user_embedded = self.user_embedding(user_input)
       book_embedded = self.book_embedding(book_input)
       concatenated = torch.cat([user_embedded, book_embedded], dim=1)
       concatenated = self.concatenation_layer(concatenated)  # add this line
       dot_product = self.dot_product(concatenated)
       review_embedded = self.review_embedding(review_input)
       concat = torch.cat([review_embedded, dot_product], dim=1)
       return self.concat(concat)

class NLPDataset(Dataset):
    def __init__(self, data, tokenizer, vocab, max_length):
        self.users = torch.tensor(data['customer_id'].values)
        self.books = torch.tensor(data['product_id'].values)
        self.ratings = torch.tensor(data['star_rating'].values)
        
        self.review_body_padded = []
        for review in data['review_body']:
            tokens = [vocab[token] if token in vocab else vocab['<unk>'] for token in tokenizer(review)]
            indices = torch.tensor(tokens, dtype=torch.long)
            if indices.shape[0] > max_length:
                indices = indices[:max_length]
            padded_indices = torch.cat([indices, torch.zeros(max_length - indices.shape[0], dtype=torch.long)])
            self.review_body_padded.append(padded_indices)
        
        self.review_body_padded = torch.stack(self.review_body_padded).long()
        self.review_body_padded[self.review_body_padded == 0] = 1  
        self.review_body_padded = torch.log(self.review_body_padded + 1)

        
        self.vocab_size = len(vocab)

    def __len__(self):
        return len(self.ratings)
    
    def __getitem__(self, idx):
        return self.users[idx], self.books[idx], self.ratings[idx], self.review_body_padded[idx]

tokenizer = get_tokenizer('basic_english')
texts = [text for text in df_nlp['review_body']]
vocab = build_vocab_from_iterator(map(tokenizer, texts), min_freq=1, specials=["<pad>", "<unk>"])
vocab.set_default_index(vocab["<unk>"])
vocab_size = len(vocab) 
max_length = 100

# set up data loaders
train_data = df_nlp.sample(frac=0.8)
test_data = df_nlp.drop(train_data.index)
train_dataset = NLPDataset(train_data, tokenizer, vocab, max_length)
test_dataset = NLPDataset(test_data, tokenizer, vocab, max_length)
train_loader = DataLoader(train_dataset, batch_size=64, shuffle=True)
test_loader = DataLoader(test_dataset, batch_size=64, shuffle=False)

# set up model and optimizer
num_users = df_nlp['customer_id'].nunique()
num_books = df_nlp['product_id'].nunique()
model_nlp = NLPModel(num_users, num_books)
optimizer = optim.Adam(model_nlp.parameters(), lr=0.01)
criterion = nn.MSELoss()

num_epochs = 10

for epoch in range(num_epochs):
    train_loss = 0.0
    for user, book, rating, review in train_loader:
      optimizer.zero_grad()
      output = model_nlp(user.long(), book.long(), review.long().reshape(-1, 100))
      loss = criterion(output.squeeze(), rating.float())
      loss.backward()
      optimizer.step()
      train_loss += loss.item() * user.size(0)

    train_loss /= len(train_loader.dataset)
    print('Epoch: {}, Training Loss: {:.4f}'.format(epoch+1, train_loss))

and I keep getting this error

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
<ipython-input-110-3bc9b0fc6272> in <module>
      5     for user, book, rating, review in train_loader:
      6       optimizer.zero_grad()
----> 7       output = model_nlp(user.long(), book.long(), review.long().reshape(-1, 100))
      8       loss = criterion(output.squeeze(), rating.float())
      9       loss.backward()

5 frames
/usr/local/lib/python3.9/dist-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
   1192         if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
   1193                 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1194             return forward_call(*input, **kwargs)
   1195         # Do not call functions when jit is used
   1196         full_backward_hooks, non_full_backward_hooks = [], []

<ipython-input-105-ab06f5dda1a4> in forward(self, user_input, book_input, review_input)
     37        review_embedded = self.review_embedding(review_input)
     38        concat = torch.cat([review_embedded, dot_product], dim=1)
---> 39        return self.concat(concat)

/usr/local/lib/python3.9/dist-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
   1192         if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
   1193                 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1194             return forward_call(*input, **kwargs)
   1195         # Do not call functions when jit is used
   1196         full_backward_hooks, non_full_backward_hooks = [], []

/usr/local/lib/python3.9/dist-packages/torch/nn/modules/container.py in forward(self, input)
    202     def forward(self, input):
    203         for module in self:
--> 204             input = module(input)
    205         return input
    206 

/usr/local/lib/python3.9/dist-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
   1192         if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
   1193                 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1194             return forward_call(*input, **kwargs)
   1195         # Do not call functions when jit is used
   1196         full_backward_hooks, non_full_backward_hooks = [], []

/usr/local/lib/python3.9/dist-packages/torch/nn/modules/linear.py in forward(self, input)
    112 
    113     def forward(self, input: Tensor) -> Tensor:
--> 114         return F.linear(input, self.weight, self.bias)
    115 
    116     def extra_repr(self) -> str:

RuntimeError: mat1 and mat2 shapes cannot be multiplied (64x3201 and 64x64)

I have also used this code

# print the shapes of the tensors
print("user shape:", user.shape)
print("book shape:", book.shape)
print("review shape:", review.shape)
    
# print the first sample in each tensor
print("user:", user[0])
print("book:", book[0])
print("review:", review[0])

and got these results

user shape: torch.Size([64])
book shape: torch.Size([64])
review shape: torch.Size([64, 100])
user: tensor(15314, dtype=torch.int16)
book: tensor(11322, dtype=torch.int16)
review: tensor([ 9.5303,  3.8286,  5.6312,  2.7081,  5.3936,  3.7612,  1.3863, 11.3508,
         4.5951,  4.4188,  1.6094,  3.2581,  1.3863,  9.2828,  2.3979,  2.9444,
         3.1355,  5.4072,  2.0794,  3.4012,  4.5747,  5.0370,  1.0986,  9.5303,
         3.8286,  5.6312,  2.1972,  4.4188,  5.6240,  1.9459,  8.7178,  1.3863,
         8.5804,  1.9459,  2.8332,  6.9354,  1.6094,  3.2581,  8.0583,  3.8286,
         7.4782,  8.1654,  3.0445,  2.1972,  4.0775,  6.2285,  1.6094,  1.7918,
         7.5893,  5.6525,  2.0794,  4.7875,  2.6391,  6.4441,  2.0794,  1.3863,
         6.8987,  1.0986,  1.7918,  4.7958,  2.0794,  4.7875,  2.7081,  4.5747,
         5.0370,  1.6094,  8.1654,  7.5191,  7.0229,  5.7301,  1.3863,  4.6347,
         3.7377,  7.8675,  5.1930,  1.9459,  1.3863,  5.7838,  1.6094,  1.7918,
         8.5860,  2.8332,  8.0968,  2.3026,  5.5947,  5.3566,  3.8067,  4.7791,
         9.6541,  1.0986,  1.3863,  2.7726,  2.3026,  6.1181,  1.6094,  6.8448,
         1.6094,  7.1420,  1.7918,  4.2485])
2

Based on the error message the first linear layer in self.dot_product or self.concat raises the error as 64 input features are expected while the input activation contains 3201 features.
You could either change the in_features value of the corresponding linear layer to 3201 or make sure the activation contains only 64 features.

3

Now i have this error mat1 and mat2 shapes cannot be multiplied (64x64 and 32x1)
I have updated the code like this

class NLPModel(nn.Module):
    def __init__(self, num_users, num_books):
        super(NLPModel, self).__init__()
        self.user_embedding = nn.Embedding(num_users+1, 32)
        self.book_embedding = nn.Embedding(num_books+1, 32)
        self.concatenation_layer = nn.Linear(64, 32*2)
        self.review_embedding = nn.Sequential(
            nn.Embedding(5000, 32),
            nn.Flatten()
        )
 
        self.dot_product = nn.Sequential(
            nn.Linear(64, 32*2),
            nn.ReLU(),
            nn.Dropout(0.5),
            nn.Linear(32, 1)
        )
        self.concat = nn.Sequential(
            nn.Linear(32+32, 64),
            nn.ReLU(),
            nn.Dropout(0.5),
            nn.Linear(64, 32),
            nn.ReLU(),
            nn.Dropout(0.5),
            nn.Linear(32, 1)
        )
    
    def forward(self, user_input, book_input, review_input):
        user_embedded = self.user_embedding(user_input)
        book_embedded = self.book_embedding(book_input)
        concatenated = torch.cat([user_embedded, book_embedded], dim=1)
        concatenated = self.concatenation_layer(concatenated)  
        dot_product = self.dot_product(concatenated)
        review_embedded = self.review_embedding(review_input)
        concat = torch.cat([review_embedded, dot_product], dim=1)
        return self.concat(concat)

class NLPDataset(Dataset):
    def __init__(self, data, tokenizer, vocab, max_length):
        self.users = torch.tensor(data['customer_id'].values)
        self.books = torch.tensor(data['product_id'].values)
        self.ratings = torch.tensor(data['star_rating'].values)
        
        self.review_body_padded = []
        for review in data['review_body']:
            tokens = [vocab[token] if token in vocab else vocab['<unk>'] for token in tokenizer(review)]
            indices = torch.tensor(tokens, dtype=torch.long)
            if indices.shape[0] > max_length:
                indices = indices[:max_length]
            padded_indices = torch.cat([indices, torch.zeros(max_length - indices.shape[0], dtype=torch.long)], dim=0)
            self.review_body_padded.append(padded_indices)
        
        self.review_body_padded = torch.stack(self.review_body_padded).long()
        self.review_body_padded = torch.log(self.review_body_padded + 1)
        
        self.vocab_size = len(vocab)

    def __len__(self):
        return len(self.ratings)
    
    def __getitem__(self, idx):
        return self.users[idx], self.books[idx], self.ratings[idx], self.review_body_padded[idx]

tokenizer = get_tokenizer('basic_english')
texts = [text for text in df_nlp['review_body']]
vocab = build_vocab_from_iterator(map(tokenizer, texts), min_freq=1, specials=["<pad>", "<unk>"])
vocab.set_default_index(vocab["<unk>"])
vocab_size = len(vocab) 
max_length = 100

# set up data loaders
train_data = df_nlp.sample(frac=0.8)
test_data = df_nlp.drop(train_data.index)
train_dataset = NLPDataset(train_data, tokenizer, vocab, max_length)
test_dataset = NLPDataset(test_data, tokenizer, vocab, max_length)
train_loader = DataLoader(train_dataset, batch_size=64, shuffle=True)
test_loader = DataLoader(test_dataset, batch_size=64, shuffle=False)

# set up model and optimizer
num_users = df_nlp['customer_id'].nunique()
num_books = df_nlp['product_id'].nunique()
model_nlp = NLPModel(num_users, num_books)
optimizer = optim.Adam(model_nlp.parameters(), lr=0.01)
criterion = nn.MSELoss()

num_epochs = 10

for epoch in range(num_epochs):
    train_loss = 0.0
    for user, book, rating, review in train_loader:
      optimizer.zero_grad()
      output = model_nlp(user.long(), book.long(), review.long())
      loss = criterion(output.squeeze(), rating.float())
      loss.backward()
      optimizer.step()
      train_loss += loss.item() * user.size(0)

    train_loss /= len(train_loader.dataset)
    print('Epoch: {}, Training Loss: {:.4f}'.format(epoch+1, train_loss))

and got this error

RuntimeError                              Traceback (most recent call last)
<ipython-input-191-5581c569afb2> in <module>
      5     for user, book, rating, review in train_loader:
      6       optimizer.zero_grad()
----> 7       output = model_nlp(user.long(), book.long(), review.long())
      8       loss = criterion(output.squeeze(), rating.float())
      9       loss.backward()

5 frames
/usr/local/lib/python3.9/dist-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
   1192         if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
   1193                 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1194             return forward_call(*input, **kwargs)
   1195         # Do not call functions when jit is used
   1196         full_backward_hooks, non_full_backward_hooks = [], []

<ipython-input-185-2f5cb88e83fc> in forward(self, user_input, book_input, review_input)
     31         concatenated = torch.cat([user_embedded, book_embedded], dim=1)
     32         concatenated = self.concatenation_layer(concatenated)
---> 33         dot_product = self.dot_product(concatenated)
     34         review_embedded = self.review_embedding(review_input)
     35         concat = torch.cat([review_embedded, dot_product], dim=1)

/usr/local/lib/python3.9/dist-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
   1192         if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
   1193                 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1194             return forward_call(*input, **kwargs)
   1195         # Do not call functions when jit is used
   1196         full_backward_hooks, non_full_backward_hooks = [], []

/usr/local/lib/python3.9/dist-packages/torch/nn/modules/container.py in forward(self, input)
    202     def forward(self, input):
    203         for module in self:
--> 204             input = module(input)
    205         return input
    206 

/usr/local/lib/python3.9/dist-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
   1192         if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
   1193                 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1194             return forward_call(*input, **kwargs)
   1195         # Do not call functions when jit is used
   1196         full_backward_hooks, non_full_backward_hooks = [], []

/usr/local/lib/python3.9/dist-packages/torch/nn/modules/linear.py in forward(self, input)
    112 
    113     def forward(self, input: Tensor) -> Tensor:
--> 114         return F.linear(input, self.weight, self.bias)
    115 
    116     def extra_repr(self) -> str:

RuntimeError: mat1 and mat2 shapes cannot be multiplied (64x64 and 32x1)

Note this is my first NLP with PyTorch

4

Kindly check the dimensions of the matrices - to multiply a m x n matrix with a x b matrix, n must be equal to a (n==a)

5

this is now the new error (ValueError: Using a target size (torch.Size([64])) that is different to the input size (torch.Size([64, 3264])) is deprecated. Please ensure they have the same size.)

class NLPModel(nn.Module):
    def __init__(self, num_users, num_books):
        super(NLPModel, self).__init__()
        self.user_embedding = nn.Embedding(num_users+1, 32)
        self.book_embedding = nn.Embedding(num_books+1, 32)
        self.concatenation_layer = nn.Linear(64, 64)
        self.review_embedding = nn.Sequential(
            nn.Embedding(5000, 32),
            nn.Flatten()
        )

        self.dot_product = nn.Sequential(
            nn.Linear(64, 64),
            nn.ReLU(),
            nn.Dropout(0.5),
            nn.Linear(64, 64)
        )
        self.concat = nn.Sequential(
            nn.Linear(64, 64),
            nn.ReLU(),
            nn.Dropout(0.5),
            nn.Linear(64, 64),
            nn.ReLU(),
            nn.Dropout(0.5),
            nn.Linear(64, 64)
        )
    
    def forward(self, user_input, book_input, review_input):
        user_embedded = self.user_embedding(user_input)
        book_embedded = self.book_embedding(book_input)
        concatenated = torch.cat([user_embedded, book_embedded], dim=1)
        concatenated = self.concatenation_layer(concatenated)  
        dot_product = self.dot_product(concatenated)
        review_embedded = self.review_embedding(review_input)
        concat = torch.cat([review_embedded, dot_product], dim=1)
        return concat

class NLPDataset(Dataset):
    def __init__(self, data, tokenizer, vocab, max_length):
        self.users = torch.tensor(data['customer_id'].values)
        self.books = torch.tensor(data['product_id'].values)
        self.ratings = torch.tensor(data['star_rating'].values)
        
        self.review_body_padded = []
        for review in data['review_body']:
            tokens = [vocab[token] if token in vocab else vocab['<unk>'] for token in tokenizer(review)]
            indices = torch.tensor(tokens, dtype=torch.long)
            if indices.shape[0] > max_length:
                indices = indices[:max_length]
            padded_indices = torch.cat([indices, torch.zeros(max_length - indices.shape[0], dtype=torch.long)], dim=0)
            self.review_body_padded.append(padded_indices)
        
        self.review_body_padded = torch.stack(self.review_body_padded).long()
        self.review_body_padded = torch.log(self.review_body_padded + 1)
        
        self.vocab_size = len(vocab)

    def __len__(self):
        return len(self.ratings)
    
    def __getitem__(self, idx):
        return self.users[idx], self.books[idx], self.ratings[idx], self.review_body_padded[idx]

tokenizer = get_tokenizer('basic_english')
texts = [text for text in df_nlp['review_body']]
vocab = build_vocab_from_iterator(map(tokenizer, texts), min_freq=1, specials=["<pad>", "<unk>"])
vocab.set_default_index(vocab["<unk>"])
vocab_size = len(vocab) 
max_length = 100

# set up data loaders
train_data = df_nlp.sample(frac=0.8)
test_data = df_nlp.drop(train_data.index)
train_dataset = NLPDataset(train_data, tokenizer, vocab, max_length)
test_dataset = NLPDataset(test_data, tokenizer, vocab, max_length)
train_loader = DataLoader(train_dataset, batch_size=64, shuffle=True)
test_loader = DataLoader(test_dataset, batch_size=64, shuffle=False)

# set up model and optimizer
num_users = df_nlp['customer_id'].nunique()
num_books = df_nlp['product_id'].nunique()
model_nlp = NLPModel(num_users, num_books)
optimizer = optim.Adam(model_nlp.parameters(), lr=0.01)
criterion = nn.BCELoss()

num_epochs = 10

for epoch in range(num_epochs):
    train_loss = 0.0
    for user, book, rating, review in train_loader:
      optimizer.zero_grad()
      output = model_nlp(user.long(), book.long(), review.long())
      loss = criterion(output, rating.long())
      loss.backward()
      optimizer.step()
      train_loss += loss.item() * user.size(0)

    train_loss /= len(train_loader.dataset)
    print('Epoch: {}, Training Loss: {:.4f}'.format(epoch+1, train_loss))

and this is the new error

ValueError                                Traceback (most recent call last)
<ipython-input-137-f80539a6f64a> in <module>
      6       optimizer.zero_grad()
      7       output = model_nlp(user.long(), book.long(), review.long())
----> 8       loss = criterion(output, rating.long())
      9       loss.backward()
     10       optimizer.step()

2 frames
/usr/local/lib/python3.9/dist-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
   1192         if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
   1193                 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1194             return forward_call(*input, **kwargs)
   1195         # Do not call functions when jit is used
   1196         full_backward_hooks, non_full_backward_hooks = [], []

/usr/local/lib/python3.9/dist-packages/torch/nn/modules/loss.py in forward(self, input, target)
    617 
    618     def forward(self, input: Tensor, target: Tensor) -> Tensor:
--> 619         return F.binary_cross_entropy(input, target, weight=self.weight, reduction=self.reduction)
    620 
    621 

/usr/local/lib/python3.9/dist-packages/torch/nn/functional.py in binary_cross_entropy(input, target, weight, size_average, reduce, reduction)
   3084         reduction_enum = _Reduction.get_enum(reduction)
   3085     if target.size() != input.size():
-> 3086         raise ValueError(
   3087             "Using a target size ({}) that is different to the input size ({}) is deprecated. "
   3088             "Please ensure they have the same size.".format(target.size(), input.size())

ValueError: Using a target size (torch.Size([64])) that is different to the input size (torch.Size([64, 3264])) is deprecated. Please ensure they have the same size.

I think need to add a flatten layer but I don’t know where to add it

6

In line # 619 return F.binary_cross_entropy(input, target, weight=self.weight, reduction=self.reduction) the dimensions of the input and target arguments passed to the function forward are not matching. You cannot compute the cross_entropy between, let’s say, 10 values and 20 values. That is what the error is indicating.

7

ok, how to fix this problem
if you have time you can check the code in colab
https://colab.research.google.com/drive/1BH-vsOyBMrQPWslQThuKmcHOi2ffJ6rs?usp=sharing