This is definitely a rookie question and I’ve tried finding the answer, but been unsuccessful, hence the post, so thanks in advance for any help!
I am on lesson 1 trying out my own data set and I have imagenet style labeled data. I have training data and test data, but no validation data, so I used valid_pct in the ImageDataBunch.from_folder() method.
As it happens, my test data is also in labeled folders. I have a decent model that I would now like to try on the test data to see how it does. But I cannot find the API calls to make this happen.
My data is organized in test and train folders (and in respective label folders under there). I used ImageDataBunch.from_folder() to load the data with the parameter test=“test” set so that it would load the test data.
I see that the ImageDataBunch object does have a test_ds property (which I presume is the loaded test data) but this object only has the images loaded (x), the labels (y) are empty. I need these labels to compare accuracy of the model.
When I try learn.get_preds(ds_type=“test”), I thought I would get back predictions for test data, but based on the number of results returned it is returning predictions for training data.
Once again, assistance would be most appreciated