speedup compressed features loading

[bobchennan]

Now it is much faster for larger data and I also tested with TEST_COMPRESSED_FEATS.py:

Before:

0.0351510047913
0.0127110481262
(u'AMI_ES2011a_H00_FEE041_0005553_0005685', 0.0012089051)
(u'AMI_ES2011a_H00_FEE041_0003714_0003915', 0.006009698)
(u'AMI_ES2011a_H00_FEE041_0004643_0004763', 0.0013620704)
(u'AMI_ES2011a_H00_FEE041_0003427_0003714', 0.006772384)
(u'AMI_ES2011a_H00_FEE041_0005133_0005553', 0.0052853823)
(u'AMI_ES2011a_H00_FEE041_0004763_0005020', 0.0033664554)
(u'AMI_ES2011a_H00_FEE041_0005856_0006217', 0.006143585)
(u'AMI_ES2011a_H00_FEE041_0004332_0004439', 0.0022902265)
(u'AMI_ES2011a_H00_FEE041_0005020_0005133', 0.002434183)
(u'AMI_ES2011a_H00_FEE041_0003915_0004332', 0.0066184625)

After:

0.00288486480713
0.00546503067017
(u'AMI_ES2011a_H00_FEE041_0005553_0005685', 0.0016530156)
(u'AMI_ES2011a_H00_FEE041_0003714_0003915', 0.004995942)
(u'AMI_ES2011a_H00_FEE041_0004643_0004763', 0.0010585152)
(u'AMI_ES2011a_H00_FEE041_0003427_0003714', 0.005950749)
(u'AMI_ES2011a_H00_FEE041_0005133_0005553', 0.0082992315)
(u'AMI_ES2011a_H00_FEE041_0004763_0005020', 0.0045164824)
(u'AMI_ES2011a_H00_FEE041_0005856_0006217', 0.0057396963)
(u'AMI_ES2011a_H00_FEE041_0004332_0004439', 0.0020962954)
(u'AMI_ES2011a_H00_FEE041_0005020_0005133', 0.0034917593)
(u'AMI_ES2011a_H00_FEE041_0003915_0004332', 0.0073646307)

Running time is not meaningful since it is too small. It is still not identical but errors are in similar range.

[KarelVesely84]

Hi, it looks good. Aha, I see it is faster because it is not done 'element-wise' but 'matrix-wise'.
Did you check if it runs both with 'python2' and 'python3' ? If yes, I'll merge it...

[KarelVesely84]

And thank you for creating the change, the coffee spirit of the Kaldi community will be grateful :)

[bobchennan]

Yes I tested with python2.7 and python3.5.3.

Running log for python 2.7:

0.0194351673126
0.00395321846008
(u'AMI_ES2011a_H00_FEE041_0005553_0005685', 0.0016530156)
(u'AMI_ES2011a_H00_FEE041_0003714_0003915', 0.004995942)
(u'AMI_ES2011a_H00_FEE041_0004643_0004763', 0.0010585152)
(u'AMI_ES2011a_H00_FEE041_0003427_0003714', 0.005950749)
(u'AMI_ES2011a_H00_FEE041_0005133_0005553', 0.0082992315)
(u'AMI_ES2011a_H00_FEE041_0004763_0005020', 0.0045164824)
(u'AMI_ES2011a_H00_FEE041_0005856_0006217', 0.0057396963)
(u'AMI_ES2011a_H00_FEE041_0004332_0004439', 0.0020962954)
(u'AMI_ES2011a_H00_FEE041_0005020_0005133', 0.0034917593)
(u'AMI_ES2011a_H00_FEE041_0003915_0004332', 0.0073646307)

Running log for python 3.5.3:

0.0035649039782583714
0.006509726983495057
AMI_ES2011a_H00_FEE041_0003427_0003714 0.005950749
AMI_ES2011a_H00_FEE041_0004643_0004763 0.0010585152
AMI_ES2011a_H00_FEE041_0004763_0005020 0.0045164824
AMI_ES2011a_H00_FEE041_0004332_0004439 0.0020962954
AMI_ES2011a_H00_FEE041_0003915_0004332 0.0073646307
AMI_ES2011a_H00_FEE041_0005553_0005685 0.0016530156
AMI_ES2011a_H00_FEE041_0005133_0005553 0.0082992315
AMI_ES2011a_H00_FEE041_0003714_0003915 0.004995942
AMI_ES2011a_H00_FEE041_0005020_0005133 0.0034917593
AMI_ES2011a_H00_FEE041_0005856_0006217 0.0057396963

[KarelVesely84]

okay, merging!