Float division for stepsize computaton in Mira

LxMLS / lxmls-toolkit

Machine Learning applied to Natural Language Processing Toolkit used in the Lisbon Machine Learning Summer School

Other

222 stars 216 forks source link

Float division for stepsize computaton in Mira #149

Closed ChristopherBrix closed 5 years ago

ChristopherBrix commented 5 years ago

This resolves #119

ChristopherBrix commented 5 years ago

This broke the Mira test! There, a training accuracy of 96% was assumed. Actually, we can get 100% now. The test was created in commit a18bafe2c9098059ac0a2f38ee185be4cdd027c5 by @pedrobalage. I guess he assumed that the current code was ok, and used the (wrong) output to configure the test.

I've fixed the test.

ramon-astudillo commented 5 years ago

I've fixed the test.

Nice did not even give me time to comment on this ;-). Lets wait for Travis but this is looking solved