Changed RandomForestModel to use ColSpec so that it matches LogisticModel
Added notebook comparing Dalhousie and GFW results.
This last confirmed @redhog 's finding that the Dal results for Purse Seiner were superior to the model we are currently using in the pipeline. However, the combination of using multi-windowing, is_daylight and measure_speed, gives a model that is significantly better. See Compare Purse Seiners With Dal.ipynb for details.
@redhog, please take a look and merge if this looks good.
I updated this to compare results for all three models now. It appears that using is_daylight and measure_speed isn't particularly helpful for the models other than PS.
fromtimestampin utils.pyThis last confirmed @redhog 's finding that the Dal results for Purse Seiner were superior to the model we are currently using in the pipeline. However, the combination of using multi-windowing,
is_daylightandmeasure_speed, gives a model that is significantly better. SeeCompare Purse Seiners With Dal.ipynbfor details.@redhog, please take a look and merge if this looks good.
I updated this to compare results for all three models now. It appears that using
is_daylightandmeasure_speedisn't particularly helpful for the models other than PS.