My problem is actually how to go from training a model ( which i now understand ) to predicting new values ( single to multiple rows) from fresh data… That seems to be difficult.
So, the learn.predict() function takes some input from your model and generates a prediction. In the example code you see, we take one row from our DataFrame and run a prediction on it. The first result is the actual labeled category, the second is the tensor class, the third is the raw probabilities.
So just call learn.predict() to predict whatever you want We have a few tools to make it a little faster or more dynamic such as get_preds as we see in the Rossmann notebook as well
I think Rossmann notebook is difficult to follow because there are no explanation/notes in the notebook just raw code… I understand though that some of what he’s doing requires extensive knowledge in pandas and numpy.
i wish there was some explanation at least similar to @jeremy notebooks …