Add GRU4Rec example #20
Conversation
examples/sequential_retrieval.py
Outdated
| import os | ||
|
|
||
| os.environ["KERAS_BACKEND"] = "jax" |
There was a problem hiding this comment.
Any reason to do that? At some point I was thinking we can write tests to run these automatically with each backend. But I never got around to doing that. But for it to work, the backend has to not be set within the file.
Did you test with TF and Torch too?
There was a problem hiding this comment.
Yep, working for all backends!
examples/sequential_retrieval.py
Outdated
|
|
||
| import keras_rs | ||
|
|
||
| keras.utils.set_random_seed(42) |
There was a problem hiding this comment.
I set the seed when I was testing training with different hyperparameters, etc. Can remove/keep it, doesn't matter which way
| [basic retrieval](https://github.com/keras-team/keras-rs/blob/main/examples/basic_retrieval.py) | ||
| example, we are going to use the MovieLens dataset. | ||
|
|
||
| The dataset preparation step is fairly involved. The original ratings dataset |
There was a problem hiding this comment.
Thanks for porting all the dataset massaging stuff here. That was most of the work I imagine.
There was a problem hiding this comment.
Yeah, pretty much. I was trying to not do the conversion to Pandas DF, and somehow use tf.data.group_by_window. At some point, I gave up xD
examples/sequential_retrieval.py
Outdated
| import keras | ||
| import pandas as pd | ||
| import tensorflow as tf # Needed only for the dataset | ||
| from keras import ops |
There was a problem hiding this comment.
Further down, you're sometimes using keras.ops.x and sometimes ops.x. Make it consistent. I don't think this import is necessary.
There was a problem hiding this comment.
Oh, yeah. Sorry about that!
Note: The loss in the TFRS example does not converge. So, changed a few things.