plot_pred_vs_actual not working for multitask models

[rwilfong]

When running the plot_pred_vs_actual function from perf_plots on a multitask model, it errors out because it does not have '_actual' columns in the pred_df created in plot_pred_vs_actual_from_file.

Reproducibility code:

from atomsci.ddm.pipeline import perf_plots as pp
model_path = 'path_to_multitask_model.tar.gz'
pp.plot_pred_vs_actual(model_path)

[mcloughlin2]

I just pushed a fix for this issue, which was tied to a more general problem with predicting from multitask regression models. Please pull the latest from the 1.7.0 branch and try making the plots and see if they work. Also, check the R2 scores shown in the plot titles and see if they agree with the original scores from when you trained the model. They won't be exactly the same because the uncertainty computation adds randomness, but they should be close.

[mauvais2]

Fixed in aff80dd