How to change a default batch size? the default batch size = 2.

[Umair6977]

I am waiting for a response. Thanks

[CherBass]

If you go into options.py you can change batch_size param to some other even number. You could also start training using this command in the terminal to change the batch_size: python train.py --batch_size 4 Please note that all experiments were done using batch_size=2 and larger batch sizes were not tested. Also, make sure your branch is up-to-date as there is a small bug fix that is necessary.

[Umair6977]

Thanks for your response, (everything rum smoothly with default batch_size= 2) but I already tried options.py and then got an error, (images_a[0:1, ::], mask_a[0:1, ::], images_a1[0:1, ::], images_a2[0:1, ::], images_a3[0:1, ::], images_a4[0:1, ::], images_a5[0:1, ::]), 3) RuntimeError: Sizes of tensors must match except in dimension 1. Got 1 and 32 (The offending index is 1)

is it directly related to batch_size, need your suggestions,

Thanks

[CherBass]

As I've mentioned the fix for this is already in master so you need to pull the latest branch

lines 563-664 in model.py:

            mask_a = (self.mask_a.unsqueeze(1)).detach()
            mask_b = (self.mask_b.unsqueeze(1)).detach()

[Umair6977]

The batch_size has been changed now but there is another value error now occurred

ValueError: Found input variables with inconsistent numbers of samples: [674, 676]

need your suggestions.!

Thanks

[CherBass]

Can you send me the full error message, alongside your param config? (i.e. options.py params)

[Umair6977]

Thanks a lot! it has been resolved, if I found another bug I will let you know.

stay blessed

[Umair6977]

Hi, what's the function of these code lines in (train.py)

example validation

    # try:
    #     _validation(opts, model, healthy_val_dataloader, anomaly_val_dataloader)
    # except Exception as e:
    #     print(f'Encountered error during validation - {e}')
    #     raise e

example test

try:
    _test(opts, model, healthy_test_dataloader, anomaly_test_dataloader)
except Exception as e:
    print(f'Encountered error during validation - {e}')
    raise e

save last model

saver.write_model(ep, total_it, iter_counter, model, model_name='model_last')
saver.write_img(ep, total_it, model)

return

i found this error

Encountered error during validation - Found input variables with inconsistent numbers of samples: [674, 676] Traceback (most recent call last): File "train.py", line 635, in main() File "train.py", line 178, in main raise e File "train.py", line 175, in main _test(opts, model, healthy_test_dataloader, anomaly_test_dataloader) File "train.py", line 419, in _test val_accuracy = accuracy_score(val_pred_temp, val_labels) File "/home/data/Umair/.conda/envs/Umair_pytorch/lib/python3.6/site-packages/sklearn/metrics/_classification.py", line 185, in accuracy_score y_type, y_true, y_pred = _check_targets(y_true, y_pred) File "/home/data/Umair/.conda/envs/Umair_pytorch/lib/python3.6/site-packages/sklearn/metrics/_classification.py", line 80, in _check_targets check_consistent_length(y_true, y_pred) File "/home/data/Umair/.conda/envs/Umair_pytorch/lib/python3.6/site-packages/sklearn/utils/validation.py", line 212, in check_consistent_length " samples: %r" % [int(l) for l in lengths]) ValueError: Found input variables with inconsistent numbers of samples: [674, 676]

Thanks.!

[CherBass]

validation >> validation function to test on the validation dataset - it's meant to compute classification scores and do some example translations (between class 0 to 1, and to create a feature attribution map). This is done at the end of every epoch.

test>> testing function to test on the test dataset - it's meant to compute classification scores and do some example translations (between class 0 to 1, and to create a feature attribution map). This is done once at the end of training.

save last model >> saves a checkpoint of last model + does some plotting

Sorry but I'm not sure how you got the error. I can't replicate it even when batch_size=4. Also, please note you're not meant to change the param val_batch_size as it could lead to errors and thus those functions might not work.

[Umair6977]

yes, I got it. I have run it with all params. but there is no Test accuracy and some other matrices have been shown graphically, but before a JSON file and some graph has been established such as val_accuracy, val_f1, val_pre, val_recall. PLZ, can you guide me on how to show graphically and generate a JSON file to measure the test accuracy for all previously defined matrices for the Test set? Thanks!

[CherBass]

Test metrics should already be saved at the end (after code finished running) in test_results.json.

Since these are just single values, there is no point to plot them graphically. In validation we plot the metrics across epochs to see how the training is going.

[Umair6977]

There is only parameters.json file has been saved, NOT test_results.json.

[CherBass]

Most likely that means that _test() function hasn't finished running successfully so you will have to debug why it didn't run

[Umair6977]

Maybe there should be something missing but I think the test() function also has run successfully. you can see this.

[Umair6977]

As an expert, if you think is there anything missing plz give some valuable suggestions to make it accurate. It's the whole overview of the process running.

Bundle of Thanks!