Sprinkle M1 with comments on what the evaluation means

[ArturoAmorQ]

Partially addresses #530.

This PR adds comments on the score method in the M1:

• the first time the method is used
• the first time we use a test set for scoring
• in the pipeline video, where care has to be taken on interpreting the result naively

[ArturoAmorQ]

Thanks for the comments @ogrisel !

[lesteve]

I think I know what you are trying to say (that we are measuring the accuracy on the training data and that it is kind of cheating) but I find the wording super confusing ...

In particular: "can a model predict something that it already saw?" I would answer "yes why not, sorry is this a trick question?"

I think you probably mean "can this really be called prediction when we are learning and predicting from the same data" but I can't find a good wording that convinces me.

I kind of think the next section in train-test data split explains this kind of thing already so I would stay short maybe something like this:

Note that here we used the same data to learn and evaluate our model, so can this evaluation be trusted, or is it too good to be true?

[lesteve]

I tried hard to refrain from tweaking the wording but I did not manage to ...

Thanks, merging this one!