Adding Movielens 100k tabular dataset to basicdatasets

[siddvenk]

Description

Adding the Movielens 100k dataset to the tabular datasets. This dataset can be used to train regression models, collaborative filtering models, Graph Neural Networks, and a host of other recommendation system models.

https://grouplens.org/datasets/movielens/100k/

Note

The tests right now are using a local instace of the repo since the metadata hasn't been published to S3. I'll update the tests accordingly once the metadata.json has been pushed to S3 and I can verify using the remote bucket works.

[siddvenk]

Ubuntu/Windows builds are failing because the associated unit test is failing. This is expected with the way I hard coded the path for the repository used in the test. Seems like there are 2 ways I can resolve this:

1. Use a relative path in the repo - test will still use a local repo but relative path should at least make finding the files work
2. Use the default repo that can only work once the metatdata has been added to the djl repo.

I see examples of both being used for the basicdataset tests - do we have a preference?

[lanking520]

Not an expertise in this part, but I think you can try to follow: #1667 this PR to see how they do the test. @zachgk any thoughts?

[siddvenk]

@lanking520 I was thinking of doing something similar to that. I wonder whether it makes sense to standardize across the basicdataset tests? Tests for Cifar10, Mnist, ImageNet, Wiki2Text are using a local repo in the tests, while others are using djl repo from what I can tell (they're not providing a repo when building, so it will use the default djl repo).

Using the djl repo in the tests seems better to me to validate the data is stored correctly, configuration in the code is correct, and the data can be pulled to be built as a dataset. With local repo we can avoid some calls to the repo to fetch the metadata and files, but since the files are cached and don't change often I'm not sure if there's much benefit using a local repo in the tests.

[codecov-commenter]

Codecov Report

Merging #1718 (e01cc90) into master (bb5073f) will decrease coverage by 1.54%.
The diff coverage is 70.71%.

@@             Coverage Diff              @@
##             master    #1718      +/-   ##
============================================
- Coverage     72.08%   70.54%   -1.55%     
- Complexity     5126     5538     +412     
============================================
  Files           473      525      +52     
  Lines         21970    24617    +2647     
  Branches       2351     2662     +311     
============================================
+ Hits          15838    17367    +1529     
- Misses         4925     5928    +1003     
- Partials       1207     1322     +115     

Impacted Files	Coverage Δ
api/src/main/java/ai/djl/modality/cv/Image.java	69.23% <ø> (-4.11%)	⬇️
...rc/main/java/ai/djl/modality/cv/output/Joints.java	71.42% <ø> (ø)
.../main/java/ai/djl/modality/cv/output/Landmark.java	100.00% <ø> (ø)
...main/java/ai/djl/modality/cv/output/Rectangle.java	72.41% <0.00%> (ø)
...i/djl/modality/cv/translator/BigGANTranslator.java	21.42% <ø> (-5.24%)	⬇️
...odality/cv/translator/BigGANTranslatorFactory.java	33.33% <0.00%> (+8.33%)	⬆️
...nslator/InstanceSegmentationTranslatorFactory.java	14.28% <0.00%> (-3.90%)	⬇️
.../ai/djl/modality/cv/translator/YoloTranslator.java	8.33% <0.00%> (-0.50%)	⬇️
.../modality/cv/translator/YoloTranslatorFactory.java	8.33% <0.00%> (-1.67%)	⬇️
...i/djl/modality/cv/translator/YoloV5Translator.java	5.38% <0.00%> (-0.31%)	⬇️
... and 287 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 95ffc30...e01cc90. Read the comment docs.

[zachgk]

I modified the metadata to add a name to the license. That way, if we show the license name somewhere it would at least have some reasonable information conveyed.

The metadata is uploaded so you can modify the test to pull from DJL central rather local now