refactor&update docs

Use a more space efficient implementation in CrossNet
Add custom_objects in utils which will be used in load_model function
Add FAQ in docs

Author: shenweichen

deepctr/layers.py

@@ -122,8 +122,8 @@ def call(self, inputs,**kwargs):
             for j in range(i + 1, num_inputs):
                 row.append(i)
                 col.append(j)
-        p = concatenate([embeds_vec_list[idx] for idx in row],axis=1)# batch num_pairs k
-        q = concatenate([embeds_vec_list[idx] for idx in col],axis=1)  # Reshape([num_pairs, self.embedding_size])
+        p = tf.concat([embeds_vec_list[idx] for idx in row],axis=1)
+        q = tf.concat([embeds_vec_list[idx] for idx in col],axis=1)
         inner_product = p * q
 
         bi_interaction = inner_product
@@ -233,16 +233,13 @@ def call(self, inputs,**kwargs):
         if K.ndim(inputs) !=2 :
             raise ValueError("Unexpected inputs dimensions %d, expect to be 2 dimensions"% (K.ndim(inputs)))
 
-        x = inputs
-        dim = x.get_shape()[-1]
-        x_0 = K.reshape(x,[-1,dim, 1])
+        x_0 = tf.expand_dims(inputs,axis=2)
         x_l = x_0
         for i in range(self.layer_num):
-            dot_ = tf.matmul(x_0, tf.transpose(x_l, [0, 2, 1]))  # K.dot(x_0,K.transpose(x_l))
-            dot_ = K.dot(dot_, self.kernels[i])
-            #x_l = K.bias_add(dot_+ x_l,self.bias[i]) # K.bias_add(dot_, self.bias)
-            x_l = dot_ + x_l + self.bias[i]#K.reshape(self.bias[i],[1,dim,1])
-        x_l = K.reshape(x_l, [-1, dim])
+            xl_w = tf.tensordot(tf.transpose(x_l,[0,2,1]),self.kernels[i],axes=(-1,0))
+            dot_ = tf.matmul(x_0,xl_w)
+            x_l = dot_ + x_l + self.bias[i]
+        x_l = tf.squeeze(x_l,axis=2)
         return x_l
 
     def get_config(self,):
@@ -259,7 +256,7 @@ class MLP(Layer):
         - nD tensor with shape: ``(batch_size, ..., input_dim)``. The most common situation would be a 2D input with shape ``(batch_size, input_dim)``.
 
       Output shape
-        - nD tensor with shape: ``(batch_size, ..., hidden_size[-1])``. For instance, for a 2D input with shape `(batch_size, input_dim)`, the output would have shape ``(batch_size, hidden_size[-1])``.
+        - nD tensor with shape: ``(batch_size, ..., hidden_size[-1])``. For instance, for a 2D input with shape ``(batch_size, input_dim)``, the output would have shape ``(batch_size, hidden_size[-1])``.
 
       Arguments
         - **hidden_size**:list of positive integer, the layer number and units in each layer.

deepctr/sequence.py

@@ -15,7 +15,7 @@ class SequencePoolingLayer(Layer):
         - seq_len is a 2D tensor with shape : ``(batch_size, 1)``,indicate valid length of each sequence.
 
       Output shape
-        - 3D tensor with shape: `(batch_size, 1, embedding_size)`.
+        - 3D tensor with shape: ``(batch_size, 1, embedding_size)``.
 
       Arguments
         - **seq_len_max**:Positive integer indicates that the max length of all the sequence feature,usually same as T.
@@ -78,7 +78,7 @@ class AttentionSequencePoolingLayer(Layer):
         - keys_length is a 2D tensor with shape: ``(batch_size, 1)``
 
       Output shape
-        -3D tensor with shape: ``(batch_size, 1, embedding_size)``.
+        - 3D tensor with shape: ``(batch_size, 1, embedding_size)``.
 
       Arguments
         - **hidden_size**:list of positive integer, the attention net layer number and units in each layer.

deepctr/utils.py

@@ -1,7 +1,7 @@
-
 from tensorflow.python.keras.layers import Input
-
-
+from .activations import *
+from .layers import *
+from .sequence import *
 
 
 def get_input(feature_dim_dict, bias_feature_dim_dict=None):
@@ -18,3 +18,16 @@ def get_input(feature_dim_dict, bias_feature_dim_dict=None):
                             enumerate(bias_feature_dim_dict["dense"])]
         return sparse_input, dense_input, bias_sparse_input, bias_dense_input
 
+
+custom_objects = {'InnerProductLayer': InnerProductLayer,
+                  'OutterProductLayer': OutterProductLayer,
+                  'MLP': MLP,
+                  'PredictionLayer': PredictionLayer,
+                  'FM': FM,
+                  'AFMLayer': AFMLayer,
+                  'CrossNet': CrossNet,
+                  'BiInteractionPooling': BiInteractionPooling,
+                  'LocalActivationUnit': LocalActivationUnit,
+                  'Dice': Dice,
+                  'SequencePoolingLayer': SequencePoolingLayer,
+                  'AttentionSequencePoolingLayer': AttentionSequencePoolingLayer, }

docs/source/FAQ.rst

@@ -0,0 +1,28 @@
+FAQ
+==========
+1. How to save or load weights/models?
+
+To save/load weights,you can write codes just like any other keras models.
+
+.. code-block:: python
+
+    model = DeepFM()
+    model.save_weights('DeepFM_w.h5')
+    model.load_weights('DeepFM_w.h5')
+
+
+To save/load models,just a little different.
+
+.. code-block:: python
+
+    from tensorflow.python.keras.models import  save_model,load_model
+    model = DeepFM()
+    save_model(model, 'DeepFM.h5')# save_model, same as before
+
+    from deepctr.utils import custom_objects
+    model = load_model('DeepFM.h5',custom_objects)# load_model,just add a parameter
+
+2. Does the models support multi-value input?
+
+Now only the `DIN <Features.html#din-deep-interest-network>`_ model support multi-value input,you can use layers in `sequence <deepctr.sequence.html>`_ to build your own models!
+And I will add the feature soon~

docs/source/conf.py

@@ -26,7 +26,7 @@
 # The short X.Y version
 version = ''
 # The full version, including alpha/beta/rc tags
-release = '0.1.3'
+release = '0.1.4'
 
 
 # -- General configuration ---------------------------------------------------

docs/source/index.rst

@@ -19,6 +19,7 @@ You can read the source code at https://github.com/shenweichen/DeepCTR
    Quick-Start
    Features
    Demo
+   FAQ
 
 .. toctree::
    :maxdepth: 3

setup.py

@@ -5,7 +5,7 @@
 
 setuptools.setup(
     name="deepctr",
-    version="0.1.3",
+    version="0.1.4",
     author="Weichen Shen",
     author_email="[email protected]",
     description="DeepCTR is a Easy-to-use,Modular and Extendible package of deep-learning based CTR models ,including serval DNN-based CTR models and lots of core components layer of the models which can be used to build your own custom model.",