Simple bagging with features from HT-L3s and 1000 individuals from Zak's Facebook dataset.
To get the data you'll first need to get access to the [email protected]
S3 account. Then you can use s3cmd
to retrieve the precomputed numpy
memmap files:
$ s3cmd get -r s3://fb1000/fb1000_data .
$ cd fb1000_data
$ cat L3_Prime_X_trn_zscored.mm.split?? > L3_Prime_X_trn_zscored.mm
$ sha1sum *.mm
395e45223b0012a7783947e32373920812563427 L3_Prime_X_trn_zscored.mm
9817c3a8bf40ead293392c273689ecddef97ed45 L3_Prime_X_tst_zscored.mm
7ae7f27b72ced3b953cc9d5922b9e8e473027bf7 L3_Prime_Y_trn.mm
829107227ba9aa49a23dce7ea9d72aa19fe1a8eb L3_Prime_Y_tst.mm
$ grep "^[[:upper:]].* = [[:digit:]].*$" fb1k_demo.py
DATA_STRIDE = 5
N_BAGS = 32
BAG_SIZE = 1e4
$ python fb1k_demo.py
(...)
Total time (all bags): 1609.31552291
>>> Testing...
Accuracy=0.81