https://github.com/Microsoft/CNTK
Revision 59a2c26d7d1c4bf0d577baa6b166fe5c17639ab6 authored by KeDengMS on 31 March 2017, 18:07:11 UTC, committed by KeDengMS on 31 March 2017, 18:07:11 UTC
When minibatch size is big (like 10000), the diagonal matrix in times node for gradient could be big. This change implements a ColumnwiseScaleAndWeightedAdd to reduce the cost in that case.
1 parent c4b9dd5
History
Tip revision: 59a2c26d7d1c4bf0d577baa6b166fe5c17639ab6 authored by KeDengMS on 31 March 2017, 18:07:11 UTC
Make sparse label CE backprop faster
Tip revision: 59a2c26
File Mode Size
install
README.md -rw-r--r-- 2.0 KB
ctf2bin.py -rw-r--r-- 10.5 KB
pytest.ini -rw-r--r-- 29 bytes
txt2ctf.py -rw-r--r-- 6.7 KB
uci2ctf.py -rw-r--r-- 6.2 KB

README.md

back to top