Ran this test:
pytest tests/unit/test_notebooks_python.py -k test_surprise_deep_dive_runs
See this error output:
E papermill.exceptions.PapermillExecutionError:
E ---------------------------------------------------------------------------
E Exception encountered at "In [10]":
E ---------------------------------------------------------------------------
E TypeError Traceback (most recent call last)
E
E ----> 1 eval_rmse = rmse(test, predictions)
E 2 eval_mae = mae(test, predictions)
E 3 eval_rsquared = rsquared(test, predictions)
E 4 eval_exp_var = exp_var(test, predictions)
E 5
E
E ~\notebooks\Recommenders\reco_utils\evaluation\python_evaluation.py in check_column_dtypes_wrapper(rating_true, rating_pred, col_user, col_item, col_rating, col_prediction, args, *kwargs)
E 57 raise TypeError(
E 58 "data types of column {} are different in true and prediction".format(
E ---> 59 col_user
E 60 )
E 61 )
E
E TypeError: data types of column userID are different in true and prediction
Tested on a Windows CPU DSVM
I created a Windows DSVM (Standard DS3 v2 - 4 vcpus, 14 GB memory). Don't know if this issue is specific to Windows (probably not).
Windows setup steps were (for now):
The test should run successfully
Hmmm it runs well on my environment. The issue was fixed in #553.
@roalexan Step 4 should happen after Step 7 as after git clone the current dir is not the cloned repository. Not sure if this is the cause but have you tried git branch and git status to make sure it is indeed staging branch?
I reran and hit the same error. Did you test on a Windows DSVM? My git status:
(reco_base) C:\Users\rba1\Recommenders>git status
On branch staging
You are right that the steps were not exactly right. Think these are the exact steps:
there is a line in the notebook that converts the prediction datatypes to int, apparently on some windows machines that is interpreted as int32, vs int64. specifying int64 resolves the issue. I'll push a quick change to address this, thanks for testing this @roalexan !
addressed by #620
Most helpful comment
there is a line in the notebook that converts the prediction datatypes to int, apparently on some windows machines that is interpreted as int32, vs int64. specifying int64 resolves the issue. I'll push a quick change to address this, thanks for testing this @roalexan !