train mode now correctly ignores the --pdate argument rather
than falling over.
Arrays of NaN in feature generation were being dropped
rather than added as a column of sentinels.
Multi feature generation was
failing due to feature name count not matching feature count, improved the
asserts around this and fixed the feature names.
Disabled scipy signal-to-noise ratio as this seems to be long-since deprecated. (my understanding is that the scipy version currently in use would fail for this feature).
Please see attached patch that fixes a bunch of bugs around feature generation in mflow.
It has the following...
mflow_bugfixes.txt