[joss] feature request: accessible utility to import a dataset

[hbaniecki]

openjournals/joss-reviews#3934

This package has good documentation. Going through the examples I came up with a feature request, which would greatly benefit introducing newcomers and prototyping code.

I like the first example in README to be straightforward and copy-paste ready, which is not the case here (looking at missing code ...).

How about implementing some import_dataset(url) / download(url) functionality in utils or Dataset that would, for example, download the dataset .zip file and unpack it (sample code) so that one can load the data into exemplary code:

from pyss3 import SS3

Dataset.import_dataset("https://github.com/sergioburdisso/pyss3/blob/master/examples/datasets/movie_review.zip")
x_train, y_train = Dataset.load_from_files("movie_review/train")
x_test, y_test = Dataset.load_from_files("movie_review/test")

clf = SS3()
clf.fit(x_train, y_train)
y_pred = clf.predict(x_test)

Implementation details and naming may vary, but it would be nice to easily run code from README.

[sergioburdisso]

Hi again @hbaniecki!

Wow, this is an amazing idea 👏👏👏

What do you think if adding a method called load_from_url() to the Dataset class, which would do the same thing as the current Dataset.load_from_files() but instead of loading the dataset from disk, it will do it from an URL, as you suggested.

Perhaps load_from_url() should take two arguments, load_from_url(url, folder=None), first the url from which to download the zipped dataset and secondly, an optional argument called something like folder that let the user to specify a particular folder to use from inside the zipped dataset. The example would end up being something like:

from pyss3 import SS3

x_train, y_train = Dataset.load_from_url("https://url/to/movie_review.zip", "train")
x_test, y_test = Dataset.load_from_url("https://url/to/movie_review.zip", "test")

clf = SS3()
clf.fit(x_train, y_train)
y_pred = clf.predict(x_test)

What do you think?
(Again, thank you for this suggestion, I think it is an awesome idea 💪😎👍)

[hbaniecki]

Thanks! If the data won't be saved on disk and only loaded into x_train, y_train etc. then load_from_url(url, folder=None) makes perfect sense.

[sergioburdisso]

Hi @hbaniecki! sorry for the delay, I just had to wait for the weekend to get down on this. I've added the suggested methods and also updated the README.md. Just check it out and let me know if everthing is OK 💪 🤓 👍

Below I'm pasting the commit message that marked this issue as closed:

Now datasets can be directly loaded via a given url, not only from disk.
To achieve this, two methods have been added to Dataset class:

• Dataset.load_from_url(...)
• Dataset.load_from_url_multilabel(...)

These methods download and extract the zip file (given by the url)
into the system's temporary folder and then call
Dataset.load_from_files() to load it
(or Dataset.load_from_files_multilabel(), respectively).

Note: If the same url is used consecutively, the already downloaded
files will be used as a cache (to avoid downloading and extracting
them again).

[hbaniecki]

No worries. Thanks! Works great.