sasrec tests #193
sasrec tests #193
Conversation
rectools/models/sasrec.py
Outdated
| raise NotImplementedError() | ||
|
|
||
|
|
||
| class IdEmbeddingsItemNet(ItemNetBase): | ||
| """ | ||
| Base class for item embeddings. To use more complicated logic then just id embeddings inherit | ||
| from this class and pass your custom ItemNet to your model params | ||
| Network for item embeddings. |
There was a problem hiding this comment.
| Network for item embeddings. | |
| Network for item embeddings based only on item ids. |
rectools/models/sasrec.py
Outdated
| output = self.ff_relu(self.ff_dropout1(self.ff_linear1(seqs))) | ||
| fin = self.ff_dropout2(self.ff_linear2(output)) | ||
| return fin | ||
|
|
||
|
|
||
| class SASRecTransformerLayers(TransformerLayersBase): | ||
| """Exactly SASRec authors architecture but with torch MHA realisation""" | ||
| """ | ||
| Exactly SASRec author's transformer blocks architecture but with torch MHA realisation. |
There was a problem hiding this comment.
| Exactly SASRec author's transformer blocks architecture but with torch MHA realisation. | |
| Exactly SASRec author's transformer blocks architecture but with pytorch Multi-Head Attention realisation. |
rectools/models/sasrec.py
Outdated
| Parameters | ||
| ---------- | ||
| n_blocks: int | ||
| Number of self-attention blocks. |
There was a problem hiding this comment.
| Number of self-attention blocks. | |
| Number of transformer blocks. |
rectools/models/sasrec.py
Outdated
| Parameters | ||
| ---------- | ||
| n_blocks: int | ||
| Number of self-attention blocks. |
There was a problem hiding this comment.
| Number of self-attention blocks. | |
| Number of transformer blocks. |
rectools/models/sasrec.py
Outdated
| use_causal_attn: bool, default True | ||
| If ``True``, causal mask is used in multi-head self-attention. | ||
| transformer_layers_type: Type(TransformerLayersBase), default `SasRecTransformerLayers` | ||
| Type of transformer layers used for training. |
There was a problem hiding this comment.
| Type of transformer layers used for training. | |
| Type of transformer layers architecture. |
rectools/models/sasrec.py
Outdated
| Parameters | ||
| ---------- | ||
| sessions: torch.Tensor | ||
| User sessions consisting of items. |
There was a problem hiding this comment.
| User sessions consisting of items. | |
| User sessions in the form of sequences of items ids. |
rectools/models/sasrec.py
Outdated
| Parameters | ||
| ---------- | ||
| sessions: torch.Tensor | ||
| User sessions consisting of items. |
There was a problem hiding this comment.
| User sessions consisting of items. | |
| User sessions in the form of sequences of items ids. |
rectools/models/sasrec.py
Outdated
| Returns | ||
| ------- | ||
| torch.Tensor | ||
| User sessions with positional encoding if use_pos_emb is ``True``. |
There was a problem hiding this comment.
| User sessions with positional encoding if use_pos_emb is ``True``. | |
| Encoded user sessions with added positional encoding if `use_pos_emb` is ``True``. |
rectools/models/sasrec.py
Outdated
| Parameters | ||
| ---------- | ||
| sessions: torch.Tensor | ||
| User sessions consisting of items. |
There was a problem hiding this comment.
| User sessions consisting of items. | |
| User sessions in the form of sequences of items ids. |
| Parameters | ||
| ---------- | ||
| sessions: List[List[int]] | ||
| User interaction sequences. |
There was a problem hiding this comment.
| User interaction sequences. | |
| User sessions in the form of sequences of items ids. |
tests/models/test_sasrec.py
Outdated
| model.fit(dataset=dataset) | ||
| users = np.array([10, 30, 40]) | ||
| actual = model.recommend(users=users, dataset=dataset, k=3, filter_viewed=filter_viewed) | ||
| actual[Columns.Item] = actual[Columns.Item].apply(int) |
There was a problem hiding this comment.
This is bad. If items in the dataset were int we need to receive int ids from recommend method. If this doesn't work, we need to find where exactly we are breaking them and fix it. Is it in IdMap when creating new id map with string ["PAD"] and then adding other ids as int?
There was a problem hiding this comment.
Let's not to apply(int) or astype(int) because they can modify float values. Pass check_dtype=False to assert_frames_equal or assert_series_equal instead
There was a problem hiding this comment.
So we still expect incorrect id types in reco? Could you explain the reason?
There was a problem hiding this comment.
We receive int item ids but pd.Series has dtype object. So the values are correct and they are not float. But dtype is not equal to int.
There was a problem hiding this comment.
Let's convert them back to int in the code (I mean in the main code, not test)
If I'm a user and I give ints, I expect ints in the reco. Besides, object is much slower for future iterations. And also there may be problems with metric calculation since pandas will fail trying to merge int and object.
| from tests.testing_utils import assert_id_map_equal, assert_interactions_set_equal | ||
|
|
||
|
|
||
| @pytest.mark.filterwarnings("ignore::pytorch_lightning.utilities.warnings.PossibleUserWarning") |
There was a problem hiding this comment.
Could you please add comments for this and the next line? (What the warnings are and why do we ignore them)
tests/models/test_sasrec.py
Outdated
| model.fit(dataset=dataset) | ||
| users = np.array([10, 30, 40]) | ||
| actual = model.recommend(users=users, dataset=dataset, k=3, filter_viewed=filter_viewed) | ||
| actual[Columns.Item] = actual[Columns.Item].apply(int) |
There was a problem hiding this comment.
So we still expect incorrect id types in reco? Could you explain the reason?
Merge with updated branch
added sasrec test