Could not convert string to float: 'tensor(0.3494,'

vision
1

Hi,

I’ve saved my model train and validation accuracy using with open( )... as f , f.close() in .txt format. The output is like this:

tensor(0.3494, device='cuda:0', dtype=torch.float64)
tensor(0.4721, device='cuda:0', dtype=torch.float64)
tensor(0.5966, device='cuda:0', dtype=torch.float64)
tensor(0.7131, device='cuda:0', dtype=torch.float64)
tensor(0.8192, device='cuda:0', dtype=torch.float64)
tensor(0.9051, device='cuda:0', dtype=torch.float64)
tensor(0.9604, device='cuda:0', dtype=torch.float64)
.
.
.

Now I want to plot the results:

accs_trn = np.loadtxt('Test/accs_trn.txt', dtype=float)
accs_val = np.loadtxt('Test/accs_val.txt', dtype=float)
plt.figure(figsize=(12, 4))
plt.plot(accs_trn[0][0])
plt.plot(accs_val[0][0])
plt.xlabel('Iteration')
plt.ylabel('Accuracy')
plt.title('model')
plt.show()

But I get this error:
ValueError: could not convert string to float: 'tensor(0.3494,'

I know what this error means but I don’t know how to handle it.
Can you please help me with how to use these .txt files to plot?

2

Since you’ve stored the data as strings you could use regex to match the actual floating point value and transform it into a float literal.

1 Like
3

Dear @ptrblck,
Thanks for your reply.

Can you help me with how I should save the accuracy of the model so I can be able to reload it and plot it easily?

4

You could save the binary data directly in numpy or PyTorch.
Here is a simple example:

accuracy = torch.tensor([0.876])
print(accuracy)
# tensor([0.8760])

torch.save(accuracy, "acc.pt")

loaded = torch.load("acc.pt")
print(loaded)
# tensor([0.8760])

# same for numpy
acc_np = accuracy.numpy()
with open("acc_np.npy", "wb") as f:
    np.save(f, acc_np)

np_loaded = np.load("acc_np.npy")
print(np_loaded)
# [0.876]
1 Like