CODEWITHSUNDEEP
python-3.xmachine-learningpytorch
James Wong
James Wong

Reputation: 355

RuntimeError: mat1 and mat2 shapes cannot be multiplied (1x7 and 1x1)

I faced this issue when I tried to train the model. This model is meant to predict Celsius to Fahrenheit.

Do help me to review my code below:

# this file is a simple example of how to use pytorch to train a model
# to predict fahrenheit from celsius

import numpy as np
import torch
import torch.nn as nn

celsius_arr = [-40, -10, 0, 8, 15, 22, 38]
fahrenheit_arr = [-40, 14, 32, 46, 59, 72, 100]

# convert arrays to numpy arrays
celsius_arr = np.array(celsius_arr, dtype=np.float32)
fahrenheit_arr = np.array(fahrenheit_arr, dtype=np.float32)


# build a model to predict fahrenheit from celsius
# 1. define the model
class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.linear = nn.Linear(1, 1)

    def forward(self, x):
        x = self.linear(x)
        return x


# 2. construct the loss and optimizer
model = Net()

criterion = nn.MSELoss()
optimizer = torch.optim.SGD(model.parameters(), lr=0.01)

# 3. training loop
for epoch in range(1000):
    inputs = torch.from_numpy(celsius_arr)
    labels = torch.from_numpy(fahrenheit_arr)

    # forward pass
    outputs = model(inputs)
    loss = criterion(outputs, labels)

    # backward pass
    optimizer.zero_grad()
    loss.backward()
    optimizer.step()

    if epoch % 100 == 0:
        print(f'epoch {epoch}, loss = {loss.item():.4f}')

# 4. inference
print(model(torch.tensor([100.0])))

# 5. save the model
torch.save(model.state_dict(), 'celsius_fahrenheit_model.pth')

Upvotes: 0

Views: 232

Answers (1)

ndrwnaguib
ndrwnaguib

Reputation: 6135

It seems like the following two arrays are expected to represent the data samples, and not features per sample.

# convert arrays to numpy arrays
celsius_arr = np.array(celsius_arr, dtype=np.float32)
fahrenheit_arr = np.array(fahrenheit_arr, dtype=np.float32)

You need to unsqueeze them (a.k.a, expand_dims) so that they would have the expected sample size, which is 7 in your case.

celsius_arr = np.expand_dims(celsius_arr, axis=1)
fahrenheit_arr = np.expand_dims(fahrenheit_arr, axis=1)

You can also use torch.unsqueeze

celsius_arr = torch.from_numpy(celsius_arr).unsqueeze(1)
fahrenheit_arr = torch.from_numpy(fahrenheit_arr).unsqueeze(1)

Upvotes: 1

Related Questions