5
69%
master: 69%

Ran 06 Feb 2019 10:16AM UTC

Files 26

Run time 1s

Badge

Embed ▾

Committed 05 Feb 2019 04:51PM UTC coverage: 13.812% (-0.1%) from 13.912%

Job # 65.5

Build Type

push

travis-ci

Committed by

st-bender

Commit Message

regress: MSLL and more statistics

Adds more statistics to the model evaluation at the end. We use the
trained model to predict the test data, which means we don't `compute()`
again and cannot use the regular `likelihood()` function but have to
make up our own predictive probability density.
From that we calculate the "mean standardized log loss" as described in
R&W, 2006, section 2.5 (2.34).

Also adds "generalized R^2" scores by comparing the model prediction
with a prediction using the training data's mean and variance. Includes
several options for the test variance for now, using the provided test
errors (squared), using the variance from `predict()`, and using the
predictive variance plus the provided errors (noisy targets).

To avoid calculating the statistics twice, we only report the statistics
on the test data set, if the train or test fraction is actually smaller
than one.

Run Details

400 of 2896 relevant lines covered (13.81%)

0.14 hits per line

st-bender / sciapy / 65 / 5
69%
master: 69%

README BADGES
x

Markdown

Textile

RDoc

HTML

Rst

Source Files on job 65.5

st-bender / sciapy / 65 / 5 69% master: 69%

README BADGES x

Markdown

Textile

RDoc

HTML

Rst

Source Files on job 65.5

st-bender / sciapy / 65 / 5
69%
master: 69%

README BADGES
x