WIP: adding a nuget to package the datasets.

[sfilipi]

The samples that appear in our documentation should have code that the user can copy/pate and execute.
For that purpose packaging the datasets in a NuGet whose namespace we can reference from the samples.

Addresses #1137

[sfilipi]

@eerhardt @Ivanidzo4ka where can i find more info about the nupkgproj syntax?
I need to add a post install step, to copy the files to some known project directory, so we can reference them from the samples.

[Ivanidzo4ka]

[](start = 0, length = 1)

SHOCK

[Ivanidzo4ka]

[](start = 0, length = 1)

AND HORROR

[sfilipi]

go on vacation :) #Closed

[Ivanidzo4ka]

in nuspec world I use targets file (to force certain actions) and I put files into specific folder https://docs.microsoft.com/en-us/nuget/reference/nuspec#including-content-files
Don't know anything about nupkgproj

---

In reply to: 427498975 [](ancestors = 427498975)

[eerhardt]

This is out of date.

[eerhardt]

I think we'd probably want to use a different nuget package name than Microsoft.ML.Datasets. That seems like it could be a useful package name in the future that we may not want to take here. Maybe adding Test or Sample or some other such word would help users realize that this NuGet package isn't for typical consumption.

[eerhardt]

in nuspec world I use targets file (to force certain actions) and I put files into specific folder https://docs.microsoft.com/en-us/nuget/reference/nuspec#including-content-files

I would use contentFiles, since they are respected when using PackageReferences:

https://docs.microsoft.com/en-us/nuget/reference/nuspec#using-the-contentfiles-element-for-content-files

[eerhardt]

@mairaw @JRAlexander - do we have other instances of where MSDN/docs.microsoft.com examples require you to install a NuGet package? I'm not sure I've seen this before.

I'm wondering if we can't just have a simple loop that creates the data into a CollectionDataSource instead?

[sfilipi]

The File operations in the .NEt docs indeed create the data in memory than write it/read it.
In the ML packages (R, Python) usually the datasets are packaged and shipped separately.

@GalOshri, to weigh in wrt what to snap to.

---

In reply to: 427859939 [](ancestors = 427859939)

[sfilipi]

We won't package the datasets in a nuget, we will point users to downloading them from github, in the samples.