Minor error in the 60 minute blitz tutorial

[rossinerbe]

Hey, just an FYI that the "Deep Learning with Flux - A 60 Minute Blitz" tutorial has a minor error.

When the test data is being prepped for the convolutional net the code used is this:

test_x, test_y = CIFAR10.testdata(Float32)
test_labels = onehotbatch(test_y, 0:9)

test = gpu.([(test_x[:,:,:,i], labels[:,i]) for i in partition(1:10000, 1000)])

in the third line the labels should be test_labels otherwise the labels are from the training rather than the test data, which naturally gives pretty random accuracy results later on.

[ToucheSir]

Good catch, mind filing a PR?

[rossinerbe]

Ah, actually it looks like the error is just in the html of the tutorial at https://fluxml.ai/tutorials/2020/09/15/deep-learning-flux.html, the .jl file in the repository is fine.

[ToucheSir]

Yes, that's because the file in this repo is out of date and hasn't been updated to match FluxML/fluxml.github.io#118. So it'd be a 2-step process: typo fix there, and full update here.

[rossinerbe]

#328 Gotcha, submitted the PR.