remove wide&deep from public API

NVIDIA-Merlin · Mar 31, 2022 · 5610f92 · 5610f92
1 parent 094a0f6
commit 5610f92
Show file tree

Hide file tree

Showing 4 changed files with 14 additions and 8 deletions.
diff --git a/merlin/models/tf/__init__.py b/merlin/models/tf/__init__.py
@@ -51,7 +51,6 @@
     AsDenseFeatures,
     AsSparseFeatures,
     CategoricalOneHot,
-    CrossFeatures,
     ExpandDims,
     LabelToOneHot,
     StochasticSwapNoise,
@@ -96,7 +95,7 @@
 )
 from merlin.models.tf.models import benchmark
 from merlin.models.tf.models.base import Model, RetrievalModel
-from merlin.models.tf.models.ranking import DCNModel, DeepFMModel, DLRMModel, WideAndDeepModel
+from merlin.models.tf.models.ranking import DCNModel, DeepFMModel, DLRMModel
 from merlin.models.tf.models.retrieval import (
     MatrixFactorizationModel,
     TwoTowerModel,
@@ -156,7 +155,6 @@
     "AsDenseFeatures",
     "AsSparseFeatures",
     "CategoricalOneHot",
-    "CrossFeatures",
     "ElementwiseSum",
     "ElementwiseSumItemMulti",
     "AsTabular",
@@ -201,7 +199,6 @@
     "DLRMModel",
     "DCNModel",
     "DeepFMModel",
-    "WideAndDeepModel",
     "losses",
     "LossType",
     "sample_batch",

diff --git a/merlin/models/tf/blocks/core/transformations.py b/merlin/models/tf/blocks/core/transformations.py
@@ -559,6 +559,10 @@ class CrossFeatures(TabularBlock):
     """Transformation performing crosses of categorical features
     based on the provided keys.
 
+    There are some open questions in this experimental implementation:
+    #TODO : remove the dependency with tensorflow input api
+    #TODO : Add possibility of loading cross keys from schema
+
     Parameters
     ----------
     keys : List[List[str]]

diff --git a/merlin/models/tf/models/ranking.py b/merlin/models/tf/models/ranking.py
@@ -248,8 +248,6 @@ def WideAndDeepModel(
     **kwargs
 ) -> Model:
     """Experimental implementation of Wide&Deep model [1];
-    The wide and deep components are trained with a single
-    optimizer in this first implementation.
     If deep_block is not provided, the model is equivalent to
     a linear model with cross-product interactions.
 
@@ -261,6 +259,9 @@ def WideAndDeepModel(
         deep_block=ml.MLPBlock([64, 128])
     )
 
+    #TODO : Implementation of a BlocksOptimizer class that accepts different optimizers
+    #  for different subsets of the model’s blocks
+
     References:
     -----------
     [1] Heng-Tze, Cheng et al.

diff --git a/tests/tf/models/test_ranking.py b/tests/tf/models/test_ranking.py
@@ -88,7 +88,9 @@ def test_deep_fm_model_single_task_from_pred_task(ecommerce_data, num_epochs=5,
 
 @pytest.mark.parametrize("run_eagerly", [True, False])
 def test_wide_and_deep_model_single_task_from_pred_task(ecommerce_data, run_eagerly, num_epochs=5):
-    model = ml.WideAndDeepModel(
+    from merlin.models.tf.models.ranking import WideAndDeepModel
+
+    model = WideAndDeepModel(
         ecommerce_data.schema,
         embedding_dim=64,
         keys=[["item_category", "item_intention"]],
@@ -105,8 +107,10 @@ def test_wide_and_deep_model_single_task_from_pred_task(ecommerce_data, run_eage
 
 
 def test_wide_and_deep_model_wrong_keys(ecommerce_data):
+    from merlin.models.tf.models.ranking import WideAndDeepModel
+
     with pytest.raises(ValueError) as exc_info:
-        ml.WideAndDeepModel(ecommerce_data.schema, embedding_dim=64, keys=[["item_1", "item_2"]])
+        WideAndDeepModel(ecommerce_data.schema, embedding_dim=64, keys=[["item_1", "item_2"]])
     assert "Make sure the cross features keys are present in the schema" in str(exc_info.value)