Issues Accessing Test Set Dataloader in tsai

[Loccret]

Hello everyone,

I am encountering some confusion regarding the test set dataloader in tsai. I have created my dls (data loaders) using the following code:

import tsai.all as ta

C, label, splits = ta.combine_split_data([C_train, C_valid, C_test], 
                                        [label_train, label_valid, label_test])
tfms  = [None, [ta.TSClassification()]]
batch_tfms = [ta.TSStandardize(by_sample=True, by_var=False)]

device = torch.device('cuda:0')
dls = ta.get_ts_dls(C, label, splits=splits, tfms=tfms, test_items=C_test,
                batch_tfms=batch_tfms, bs=10000, shuffle_train=True, device=device)

I need to extract rocket features, and therefore I require access to dls.train, dls.valid, and dls.test separately. However, when attempting to access dls.test, I receive an error message:

AttributeError: test

However, when training other models such as Transformer and InceptionTime, after fitting, learn.get_preds(ds_idx = 2) can access the test data correctly. Therefore, I know that the test dataloader is indeed included in the dls. However, I do not know how to access this test dataloader.

Can someone please help me with this issue?

Thank you in advance!

[vrodriguezf]

Does learn.dls[2]work?

[Loccret]

Hey @vrodriguezf .

Thank you for your reply.

I'm currently working with ROCKET to determine if there are any performance improvements, and I'm following the example provided in the documentation which can be found here.

It appears that no learner has been created.

I would like get something like:

X_test, y_test = create_rocket_features(dls.test, model)

Unfortunately, I'm receiving an error message which reads:

AttributeError: test

Would you please tell me how to do it?

[Loccret]

This is a bug that was reported in #724.

In the latest version (0.3.4), the test set can be accessed correctly by using dls[2].