Fix the inconsistency of prediction results when the model is loaded with trained weights

Author: 浅梦

deepctr/__init__.py

@@ -1,4 +1,4 @@
 from .utils import check_version
 
-__version__ = '0.7.2'
+__version__ = '0.7.3'
 check_version(__version__)

deepctr/layers/utils.py

@@ -69,7 +69,7 @@ def __init__(self, l2_reg=0.0, mode=0, use_bias=False, **kwargs):
 
         self.l2_reg = l2_reg
         # self.l2_reg = tf.contrib.layers.l2_regularizer(float(l2_reg_linear))
-        if mode not in [0,1,2]:
+        if mode not in [0, 1, 2]:
             raise ValueError("mode must be 0,1 or 2")
         self.mode = mode
         self.use_bias = use_bias
@@ -81,24 +81,35 @@ def build(self, input_shape):
                                         shape=(1,),
                                         initializer=tf.keras.initializers.Zeros(),
                                         trainable=True)
-        if self.mode != 0 :
-            self.dense = tf.keras.layers.Dense(units=1, activation=None, use_bias=False,
-                                           kernel_regularizer=tf.keras.regularizers.l2(self.l2_reg))
+        if self.mode == 1:
+            self.kernel = self.add_weight(
+                'linear_kernel',
+                shape=[int(input_shape[-1]), 1],
+                initializer=tf.keras.initializers.glorot_normal(),
+                regularizer=tf.keras.regularizers.l2(self.l2_reg),
+                trainable=True)
+        elif self.mode == 2 :
+            self.kernel = self.add_weight(
+                'linear_kernel',
+                shape=[int(input_shape[1][-1]), 1],
+                initializer=tf.keras.initializers.glorot_normal(),
+                regularizer=tf.keras.regularizers.l2(self.l2_reg),
+                trainable=True)
 
         super(Linear, self).build(input_shape)  # Be sure to call this somewhere!
 
-    def call(self, inputs , **kwargs):
-
+    def call(self, inputs, **kwargs):
         if self.mode == 0:
             sparse_input = inputs
             linear_logit = reduce_sum(sparse_input, axis=-1, keep_dims=True)
         elif self.mode == 1:
             dense_input = inputs
-            linear_logit = self.dense(dense_input)
+            fc = tf.tensordot(dense_input, self.kernel, axes=(-1, 0))
+            linear_logit = fc
         else:
             sparse_input, dense_input = inputs
-
-            linear_logit = reduce_sum(sparse_input, axis=-1, keep_dims=False) + self.dense(dense_input)
+            fc = tf.tensordot(dense_input, self.kernel, axes=(-1, 0))
+            linear_logit = reduce_sum(sparse_input, axis=-1, keep_dims=False) + fc
         if self.use_bias:
             linear_logit += self.bias
 

docs/source/History.md

@@ -1,4 +1,5 @@
 # History
+- 03/04/2020 : [v0.7.3](https://github.com/shenweichen/DeepCTR/releases/tag/v0.7.3) released.Fix the inconsistency of prediction results when the model is loaded with trained weights.
 - 02/08/2020 : [v0.7.2](https://github.com/shenweichen/DeepCTR/releases/tag/v0.7.2) released.Fix some bugs.
 - 01/28/2020 : [v0.7.1](https://github.com/shenweichen/DeepCTR/releases/tag/v0.7.1) released.Simplify [VarLenSparseFeat](./Features.html#varlensparsefeat),support setting weight_normalization.Fix problem of embedding size of `SparseFeat` in `linear_feature_columns`.
 - 11/24/2019 : [v0.7.0](https://github.com/shenweichen/DeepCTR/releases/tag/v0.7.0) released.Refactor [feature columns](./Features.html#feature-columns).Different features can use different `embedding_dim` and group-wise interaction is available by setting `group_name`.

docs/source/conf.py

@@ -26,7 +26,7 @@
 # The short X.Y version
 version = ''
 # The full version, including alpha/beta/rc tags
-release = '0.7.2'
+release = '0.7.3'
 
 
 # -- General configuration ---------------------------------------------------

docs/source/index.rst

@@ -34,12 +34,12 @@ You can read the latest code at https://github.com/shenweichen/DeepCTR
 
 News
 -----
+03/04/2020 : Fix the inconsistency of prediction results when the model is loaded with trained weights. `Changelog <https://github.com/shenweichen/DeepCTR/releases/tag/v0.7.3>`_
+
 02/08/2020 : Fix some bugs. `Changelog <https://github.com/shenweichen/DeepCTR/releases/tag/v0.7.2>`_
 
 01/28/2020 : Simplify `VarLenSparseFeat <./Features.html#varlensparsefeat>`_ ,support setting weight_normalization.Fix problem of embedding size of ``SparseFeat`` in ``linear_feature_columns``. `Changelog <https://github.com/shenweichen/DeepCTR/releases/tag/v0.7.1>`_
 
-11/24/2019 : Refactor `feature columns <./Features.html#feature-columns>`_ . Different features can use different ``embedding_dim`` and  group-wise interaction is available by setting ``group_name``. `Changelog <https://github.com/shenweichen/DeepCTR/releases/tag/v0.7.0>`_
-
 DisscussionGroup
 -----------------------
 

examples/run_dien.py

@@ -8,17 +8,17 @@
 def get_xy_fd(use_neg=False, hash_flag=False):
     feature_columns = [SparseFeat('user', 3, embedding_dim=10, use_hash=hash_flag),
                        SparseFeat('gender', 2, embedding_dim=4, use_hash=hash_flag),
-                       SparseFeat('item', 3 + 1, embedding_dim=8, use_hash=hash_flag),
-                       SparseFeat('item_gender', 2 + 1, embedding_dim=4, use_hash=hash_flag),
-                       DenseFeat('score', 1)]
+                       SparseFeat('item_id', 3 + 1, embedding_dim=8, use_hash=hash_flag),
+                       SparseFeat('cate_id', 2 + 1, embedding_dim=4, use_hash=hash_flag),
+                       DenseFeat('pay_score', 1)]
 
     feature_columns += [
-        VarLenSparseFeat(SparseFeat('hist_item', vocabulary_size=3 + 1, embedding_dim=8, embedding_name='item'),
+        VarLenSparseFeat(SparseFeat('hist_item_id', vocabulary_size=3 + 1, embedding_dim=8, embedding_name='item_id'),
                          maxlen=4, length_name="seq_length"),
-        VarLenSparseFeat(SparseFeat('hist_item_gender', 2 + 1, embedding_dim=4, embedding_name='item_gender'), maxlen=4,
+        VarLenSparseFeat(SparseFeat('hist_cate_id', 2 + 1, embedding_dim=4, embedding_name='cate_id'), maxlen=4,
                          length_name="seq_length")]
 
-    behavior_feature_list = ["item", "item_gender"]
+    behavior_feature_list = ["item_id", "cate_id"]
     uid = np.array([0, 1, 2])
     ugender = np.array([0, 1, 0])
     iid = np.array([1, 2, 3])  # 0 is mask value
@@ -30,17 +30,17 @@ def get_xy_fd(use_neg=False, hash_flag=False):
 
     behavior_length = np.array([3, 3, 2])
 
-    feature_dict = {'user': uid, 'gender': ugender, 'item': iid, 'item_gender': igender,
-                    'hist_item': hist_iid, 'hist_item_gender': hist_igender,
-                    'score': score, "seq_length": behavior_length}
+    feature_dict = {'user': uid, 'gender': ugender, 'item_id': iid, 'cate_id': igender,
+                    'hist_item_id': hist_iid, 'hist_cate_id': hist_igender,
+                    'pay_score': score, "seq_length": behavior_length}
 
     if use_neg:
-        feature_dict['neg_hist_item'] = np.array([[1, 2, 3, 0], [1, 2, 3, 0], [1, 2, 0, 0]])
-        feature_dict['neg_hist_item_gender'] = np.array([[1, 1, 2, 0], [2, 1, 1, 0], [2, 1, 0, 0]])
+        feature_dict['neg_hist_item_id'] = np.array([[1, 2, 3, 0], [1, 2, 3, 0], [1, 2, 0, 0]])
+        feature_dict['neg_hist_cate_id'] = np.array([[1, 1, 2, 0], [2, 1, 1, 0], [2, 1, 0, 0]])
         feature_columns += [
-            VarLenSparseFeat(SparseFeat('neg_hist_item', vocabulary_size=3 + 1, embedding_dim=8, embedding_name='item'),
+            VarLenSparseFeat(SparseFeat('neg_hist_item_id', vocabulary_size=3 + 1, embedding_dim=8, embedding_name='item_id'),
                              maxlen=4, length_name="seq_length"),
-            VarLenSparseFeat(SparseFeat('neg_hist_item_gender', 2 + 1, embedding_dim=4, embedding_name='item_gender'),
+            VarLenSparseFeat(SparseFeat('neg_hist_cate_id', 2 + 1, embedding_dim=4, embedding_name='cate_id'),
                              maxlen=4, length_name="seq_length")]
 
     x = {name: feature_dict[name] for name in get_feature_names(feature_columns)}

examples/run_din.py

@@ -7,22 +7,22 @@
 def get_xy_fd():
 
     feature_columns = [SparseFeat('user',3,embedding_dim=10),SparseFeat(
-        'gender', 2,embedding_dim=4), SparseFeat('item', 3 + 1,embedding_dim=8), SparseFeat('item_gender', 2 + 1,embedding_dim=4),DenseFeat('score', 1)]
-    feature_columns += [VarLenSparseFeat(SparseFeat('hist_item', vocabulary_size=3 + 1,embedding_dim=8,embedding_name='item'), maxlen=4),
-                        VarLenSparseFeat(SparseFeat('hist_item_gender', 2 + 1,embedding_dim=4, embedding_name='item_gender'), maxlen=4)]
+        'gender', 2,embedding_dim=4), SparseFeat('item_id', 3 + 1,embedding_dim=8), SparseFeat('cate_id', 2 + 1,embedding_dim=4),DenseFeat('pay_score', 1)]
+    feature_columns += [VarLenSparseFeat(SparseFeat('hist_item_id', vocabulary_size=3 + 1,embedding_dim=8,embedding_name='item_id'), maxlen=4),
+                        VarLenSparseFeat(SparseFeat('hist_cate_id', 2 + 1,embedding_dim=4, embedding_name='cate_id'), maxlen=4)]
 
-    behavior_feature_list = ["item", "item_gender"]
+    behavior_feature_list = ["item", "cate_id"]
     uid = np.array([0, 1, 2])
     ugender = np.array([0, 1, 0])
     iid = np.array([1, 2, 3])  # 0 is mask value
-    igender = np.array([1, 2, 1])  # 0 is mask value
-    score = np.array([0.1, 0.2, 0.3])
+    cate_id = np.array([1, 2, 1])  # 0 is mask value
+    pay_score = np.array([0.1, 0.2, 0.3])
 
     hist_iid = np.array([[1, 2, 3, 0], [1, 2, 3, 0], [1, 2, 0, 0]])
-    hist_igender = np.array([[1, 1, 2, 0], [2, 1, 1, 0], [2, 1, 0, 0]])
+    hist_cate_id = np.array([[1, 1, 2, 0], [2, 1, 1, 0], [2, 1, 0, 0]])
 
-    feature_dict = {'user': uid, 'gender': ugender, 'item': iid, 'item_gender': igender,
-                    'hist_item': hist_iid, 'hist_item_gender': hist_igender, 'score': score}
+    feature_dict = {'user': uid, 'gender': ugender, 'item_id': iid, 'cate_id': cate_id,
+                    'hist_item_id': hist_iid, 'hist_cate_id': hist_cate_id, 'pay_score': pay_score}
     x = {name:feature_dict[name] for name in get_feature_names(feature_columns)}
     y = [1, 0, 1]
     return x, y, feature_columns, behavior_feature_list

examples/run_dsin.py

@@ -9,37 +9,37 @@ def get_xy_fd(hash_flag=False):
     feature_columns = [SparseFeat('user', 3, embedding_dim=10, use_hash=hash_flag),
                        SparseFeat('gender', 2, embedding_dim=4, use_hash=hash_flag),
                        SparseFeat('item', 3 + 1, embedding_dim=4, use_hash=hash_flag),
-                       SparseFeat('item_gender', 2 + 1, embedding_dim=4, use_hash=hash_flag),
-                       DenseFeat('score', 1)]
+                       SparseFeat('cate_id', 2 + 1, embedding_dim=4, use_hash=hash_flag),
+                       DenseFeat('pay_score', 1)]
     feature_columns += [
         VarLenSparseFeat(SparseFeat('sess_0_item', 3 + 1, embedding_dim=4, use_hash=hash_flag, embedding_name='item'),
                          maxlen=4), VarLenSparseFeat(
-            SparseFeat('sess_0_item_gender', 2 + 1, embedding_dim=4, use_hash=hash_flag, embedding_name='item_gender'),
+            SparseFeat('sess_0_cate_id', 2 + 1, embedding_dim=4, use_hash=hash_flag, embedding_name='cate_id'),
             maxlen=4)]
     feature_columns += [
         VarLenSparseFeat(SparseFeat('sess_1_item', 3 + 1, embedding_dim=4, use_hash=hash_flag, embedding_name='item'),
                          maxlen=4), VarLenSparseFeat(
-            SparseFeat('sess_1_item_gender', 2 + 1, embedding_dim=4, use_hash=hash_flag, embedding_name='item_gender'),
+            SparseFeat('sess_1_cate_id', 2 + 1, embedding_dim=4, use_hash=hash_flag, embedding_name='cate_id'),
             maxlen=4)]
 
-    behavior_feature_list = ["item", "item_gender"]
+    behavior_feature_list = ["item", "cate_id"]
     uid = np.array([0, 1, 2])
     ugender = np.array([0, 1, 0])
     iid = np.array([1, 2, 3])  # 0 is mask value
-    igender = np.array([1, 2, 1])  # 0 is mask value
+    cateid = np.array([1, 2, 1])  # 0 is mask value
     score = np.array([0.1, 0.2, 0.3])
 
     sess1_iid = np.array([[1, 2, 3, 0], [1, 2, 3, 0], [0, 0, 0, 0]])
-    sess1_igender = np.array([[1, 1, 2, 0], [2, 1, 1, 0], [0, 0, 0, 0]])
+    sess1_cate_id = np.array([[1, 1, 2, 0], [2, 1, 1, 0], [0, 0, 0, 0]])
 
     sess2_iid = np.array([[1, 2, 3, 0], [0, 0, 0, 0], [0, 0, 0, 0]])
-    sess2_igender = np.array([[1, 1, 2, 0], [0, 0, 0, 0], [0, 0, 0, 0]])
+    sess2_cate_id = np.array([[1, 1, 2, 0], [0, 0, 0, 0], [0, 0, 0, 0]])
 
     sess_number = np.array([2, 1, 0])
 
-    feature_dict = {'user': uid, 'gender': ugender, 'item': iid, 'item_gender': igender,
-                    'sess_0_item': sess1_iid, 'sess_0_item_gender': sess1_igender, 'score': score,
-                    'sess_1_item': sess2_iid, 'sess_1_item_gender': sess2_igender, }
+    feature_dict = {'user': uid, 'gender': ugender, 'item': iid, 'cate_id': cateid,
+                    'sess_0_item': sess1_iid, 'sess_0_cate_id': sess1_cate_id, 'pay_score': score,
+                    'sess_1_item': sess2_iid, 'sess_1_cate_id': sess2_cate_id, }
 
     x = {name: feature_dict[name] for name in get_feature_names(feature_columns)}
     x["sess_length"] = sess_number

setup.py

@@ -9,7 +9,7 @@
 
 setuptools.setup(
     name="deepctr",
-    version="0.7.2",
+    version="0.7.3",
     author="Weichen Shen",
     author_email="[email protected]",
     description="Easy-to-use,Modular and Extendible package of deep learning based CTR(Click Through Rate) prediction models with tensorflow 1.x and 2.x .",