Suggestions for improvement of 'make download'

[9bow]

Hello, I am translating PyTorch Tutorials in Korean.
(I have been translating drafts of the 0.4 and the 0.3.1 with some contributers.)

After 1.0 has been released recently, I have been taking a 1.0 tutorial.
In the meantime, I noticed some improvements to make download as follows:

1. Using wget and curl are mixed.
    => Suggestion: What if we unify as one? I suggest wget already using the -N (timestamping) option. Some data like UrbanSound8K have a very large data size(over 5GB), and it seems like it will take a very long time to build each time.

2. The download location is not specified properly.
    => Suggestion: How about creating a temporary directory (eg _data) using wget's -P option? Could be using a variable like $DATADIR in Makefile. Also, I think this will make managing .gitignore easier.

3. How about using templates for downloading / unpacking data files?
    => Suggestion: I saw > null, but in fact, everything is created in a file named null because it is not /dev/null. I'm not sure I need sudo or chmod to download the data. Would not it be enough to just (1) download the file and / (2) (if necessary) extract / (3) move or delete downloaded file?

Please review and comment. If others agree, I'll modify the Makefile.

Thanks.

[soumith]

cc: @brianjo

[brianjo]

Junghwan, this sounds like a really good idea and making these consistent has been on my list as well. I would love to see/test your modification for this.

[9bow]

Good. I will make a PR after comparing the before and after downloading the files using the modified version of Makefile. (But please review once again to see if it works fine.)

[9bow]

@brianjo Could you review this PR? (#390)