Replace pandas.Series.map() in SAR

[zegerius]

Description

pandas' map() implementation is notoriously slow (see related issue for statistics). By replacing this with a stdlib implementation, we see significant improvements.

With a makeshift benchmark on the test suite (n=10) using pytest -s tests/unit/test_sar_singlenode.py --durations=0 shows an improvement.

new impl    | pandas impl | diff
------------+-------------+-----------------------------------------------------------------
0.93s setup | 0.95s setup | -3%     tests/unit/test_sar_singlenode.py::test_sar_item_similarity[1-cooccurrence-count]
0.54s call  | 0.61s call  | -11%    tests/unit/test_sar_singlenode.py::test_sar_item_similarity[1-lift-lift]
0.48s call  | 0.51s call  | -6%     tests/unit/test_sar_singlenode.py::test_sar_item_similarity[3-jaccard-jac]
0.46s call  | 0.46s call  | 0       tests/unit/test_sar_singlenode.py::test_sar_item_similarity[3-lift-lift]
0.45s call  | 0.45s call  | 0       tests/unit/test_sar_singlenode.py::test_sar_item_similarity[1-jaccard-jac]
0.40s call  | 0.41s call  | -2%     tests/unit/test_sar_singlenode.py::test_sar_item_similarity[3-cooccurrence-count]
0.37s call  | 0.38s call  | -3%     tests/unit/test_sar_singlenode.py::test_sar_item_similarity[1-cooccurrence-count]
0.27s call  | 0.34s call  | -23%    tests/unit/test_sar_singlenode.py::test_recommend_k_items[3-cooccurrence-count]
0.26s call  | 0.27s call  | -4%     tests/unit/test_sar_singlenode.py::test_recommend_k_items[3-jaccard-jac]
0.26s call  | 0.26s call  | 0       tests/unit/test_sar_singlenode.py::test_user_affinity
0.25s call  | 0.25s call  | 0       tests/unit/test_sar_singlenode.py::test_recommend_k_items[3-lift-lift]
0.03s call  | 0.03s call  | 0       tests/unit/test_sar_singlenode.py::test_get_item_based_topk
0.02s call  | 0.02s call  | 0       tests/unit/test_sar_singlenode.py::test_get_normalized_scores
0.01s call  | 0.01s call  | 0       tests/unit/test_sar_singlenode.py::test_fit[lift-True]
0.01s call  | 0.01s call  | 0       tests/unit/test_sar_singlenode.py::test_predict[lift-True]
0.01s call  | 0.01s call  | 0       tests/unit/test_sar_singlenode.py::test_predict[jaccard-False]
0.01s call  | 0.01s call  | 0       tests/unit/test_sar_singlenode.py::test_predict_all_items
0.01s call  | 0.01s call  | 0       tests/unit/test_sar_singlenode.py::test_get_popularity_based_topk
0.01s call  | 0.01s call  | 0       tests/unit/test_sar_singlenode.py::test_fit[jaccard-False]

This PR also solves some trailing spaces (thanks VSC).

Related Issues

#1022

Checklist:

• [ x ] I have followed the contribution guidelines and code style for this project.
• [ x ] I have added tests covering my contributions.
• [ x ] I have updated the documentation accordingly.
• [ x ] This PR is being made to staging and not master.

[yueguoguo]

LGTM - thanks @zegerius for the improvement!

[miguelgfierro]

this is awesome! thanks