60 minute blitz not wokring - trainimgs and valimgs not defined

[KronosTheLate]

From the 60-minute blitz:

using Statistics
using Flux, Flux.Optimise
using Images: channelview
using Metalhead
using Metalhead: trainimgs, valimgs
using Images.ImageCore
using Flux: onehotbatch, onecold, flatten
using Base.Iterators: partition

However, I get the following error:

So the example is not functional.

[DhairyaLGandhi]

I think this is broken with the current Metalhead release. Could you try with [email protected]

[KronosTheLate]

I just tried with the correct enviroment activated and instantiated. So the example works. But it should still be updated, right? To lower the barrier of entry - I am not sure if the people finding their way to the introduction all are comfortable with cloning Git-repos and juggling new enviroments in Julia.

[DhairyaLGandhi]

Absolutely. I think this should be something we get back into metalhead, since it's good for user convenience, and the api looks right as well.

[KronosTheLate]

Cool. Would it be too much to make an issue over there, asking for it to be added again, with a reference to this conversation?

[DhairyaLGandhi]

Can do that. Ideally we wouldn't want to add more deps, but have an simple implementation for data preprocessing etc like we used to.

[darsnack]

These functions already exist in MLDatasets.jl. Metalhead.jl was intentionally stripped down to just the models to reduce feature duplication across packages. Instead, the tutorial should be updated to use MLDatasets.jl.

[casasgomezuribarri]

using MLDatasets: trainimgs and using MLDatasets: valimgs do not work either

Also, running the line Metalhead.download(Metalhead.CIFAR10) gives an error too:

(Running Metalhead.download(CIFAR10) or Metalhead.download(MLDatasets.CIFAR10) returns an error

ERROR: LoadError: MethodError: no method matching download(::Module)

How can I go through the tutorial?

[DhairyaLGandhi]

Use Metalhead v0.5

[casasgomezuribarri]

Doing

> ]add Metalhead@0.5
> using Metalhead: trainimgs 

returns the same error (Pkg.status says I'm using Metalhead v0.5.3)

Doing

> ]add Metalhead@0.5.0

returns an incompatibility error with Flux

ERROR: Unsatisfiable requirements detected for package Flux [587475ba]:
 Flux [587475ba] log:
 ├─possible versions are: 0.4.1-0.12.8 or uninstalled
 ├─restricted to versions * by an explicit requirement, leaving only versions 0.4.1-0.12.8
 ├─restricted by compatibility requirements with Metalhead [dbeba491] to versions: 0.10.0-0.10.4
 │ └─Metalhead [dbeba491] log:
 │   ├─possible versions are: 0.3.0-0.6.1 or uninstalled
 │   └─restricted to versions 0.5.0 by an explicit requirement, leaving only versions 0.5.0
 └─restricted by compatibility requirements with CuArrays [3a865a2d] to versions: [0.4.1-0.8.3, 0.11.0-0.12.8] or uninstalled — no versions left
   └─CuArrays [3a865a2d] log:
     ├─possible versions are: 0.2.1-2.2.2 or uninstalled
     └─restricted by julia compatibility requirements to versions: uninstalled

]rm Flux does not solve the problem

[ToucheSir]

Simply changing the package name will not work. Please see the updated tutorial at https://fluxml.ai/tutorials/2020/09/15/deep-learning-flux.html.

[darsnack]

using MLDatasets: trainimgs and using MLDatasets: valimgs do not work either

This should be using Metalhead.CIFAR10: traintensor, testtensor (or traindata/testdata to get the images + labels. The updated tutorial that @ToucheSir linked reflects this.