Add deep recommender example. #23
Conversation
examples/deep_recommender.py
Outdated
| usually not be immediately usable in a model. | ||
|
|
||
| For example: | ||
| - User and item ids may be strings (titles, usernames) or large, noncontiguous |
There was a problem hiding this comment.
NITs:
ids --> IDs (in other places as well)
noncontiguous --> non-contiguous
There was a problem hiding this comment.
Also did ids -> IDs in other examples.
examples/deep_recommender.py
Outdated
|
|
||
| Of course, complex models also have their disadvantages. The first is | ||
| computational cost, as larger models require both more memory and more | ||
| computation to fit and serve. The second is the requirement for more data. In |
There was a problem hiding this comment.
Does fit here mean .fit()? Maybe, we should use train?
examples/deep_recommender.py
Outdated
|
|
||
| Nevertheless, effort put into building and fine-tuning larger models often pays | ||
| off. In this tutorial, we will illustrate how to build a deep retrieval model | ||
| using Keras Recommenders. We'll do this by building progressively more complex |
There was a problem hiding this comment.
At some point, we need to decide the name of the library and uniformly change it everywhere :P
There was a problem hiding this comment.
Yes. Well I removed it here for now, it doesn't really add anything.
examples/deep_recommender.py
Outdated
|
|
||
| ### Normalizing continuous features | ||
|
|
||
| Continuous features may need normlization so that they fall within an acceptable |
There was a problem hiding this comment.
normlization -> normalization
examples/deep_recommender.py
Outdated
| ### Normalizing continuous features | ||
|
|
||
| Continuous features may need normlization so that they fall within an acceptable | ||
| range for the model. We will give two examples of such normalization |
examples/deep_recommender.py
Outdated
|
|
||
| """ | ||
| This looks correct, the layer is tokenizing titles into individual words. | ||
|
|
There was a problem hiding this comment.
Don't need newline here and below?
examples/deep_recommender.py
Outdated
| """ | ||
|
|
||
| AGE_BINS_COUNT = 10 | ||
| use_age_feature = keras.utils.FeatureSpace.float_discretized( |
There was a problem hiding this comment.
use_age_feature --> user_age_feature?
examples/deep_recommender.py
Outdated
| ### Extract raw candidate features | ||
|
|
||
| First, we gather all the raw features from the dataset in lists. That is the | ||
| titles of the movies and the genres. Note that one ore more genres are |
examples/deep_recommender.py
Outdated
|
|
||
| """ | ||
| Now we need to pad genres with an Out Of Vocabulary value to be able to | ||
| represent genres with as a fixed size vector. We'll pad with zeros for |
There was a problem hiding this comment.
Remove with before as?
| keras.layers.GlobalAveragePooling1D(), | ||
| ] | ||
| ) | ||
| self.movie_genres_embedding = keras.Sequential( |
There was a problem hiding this comment.
Should we add a comment here, saying that for every movie, we embed every token, and then take the mean of all token embeddings?
There was a problem hiding this comment.
Added a comment for the title token embeddings and the genres embeddings.
In this example, we add features for both movies and user, and compare networks of different depths for both towers.
hertschuh
force-pushed
the
deep_rec_example
branch
from
24ad8cb to
a0e1768
Compare
In this example, we add features for both movies and user, and compare networks of different depths for both towers.