[GitHub] ZiyueHuang commented on a change in pull request #8180: Add wide and deep model into sparse example

[GitBox <gi...@apache.org>]

ZiyueHuang commented on a change in pull request #8180: Add wide and deep model into sparse example
URL: https://github.com/apache/incubator-mxnet/pull/8180#discussion_r151158202
 
 

 ##########
 File path: example/sparse/wide_deep_model.py
 ##########
 @@ -0,0 +1,59 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+
+import mxnet as mx
+from weighted_softmax_ce import *
+
+
+def wide_deep_model(num_linear_features, num_embed_features, num_cont_features, 
+                    input_dims, hidden_units, positive_cls_weight):
+    csr_data = mx.symbol.Variable("csr_data", stype='csr')
+    label = mx.symbol.Variable("softmax_label")
+
+    norm_init = mx.initializer.Normal(sigma=0.01)
+    # weight with row_sparse storage type to enable sparse gradient updates
+    weight = mx.symbol.Variable("linear_weight", shape=(num_linear_features, 2),
+                                init=norm_init, stype='row_sparse')
+    bias = mx.symbol.Variable("linear_bias", shape=(2,))
+    dot = mx.symbol.sparse.dot(csr_data, weight)
+    linear_out = mx.symbol.broadcast_add(dot, bias)
+
+    dns_data = mx.symbol.Variable("dns_data")
+    x = mx.symbol.slice(data=dns_data, begin=(0, 0),
+                        end=(None, num_embed_features))
+    embeds = mx.symbol.split(data=x, num_outputs=num_embed_features, squeeze_axis=1)
+
+    x = mx.symbol.slice(data=dns_data, begin=(0, num_embed_features),
+                        end=(None, num_embed_features + num_cont_features))
+    features = [x]
+
+    for i, embed in enumerate(embeds):
+        embed_weight = mx.symbol.Variable('embed_%d_weight' % i, stype='row_sparse')
+        features.append(mx.symbol.contrib.SparseEmbedding(data=embed, weight=embed_weight,
+                        input_dim=input_dims[i], output_dim=hidden_units[0]))
+
+    hidden = mx.symbol.concat(*features, dim=1)
+    hidden = mx.symbol.BatchNorm(data=hidden)
+    hidden = mx.symbol.FullyConnected(data=hidden, num_hidden=hidden_units[1])
+    hideen = mx.symbol.Activation(data=hidden, act_type='relu')
+    hidden = mx.symbol.FullyConnected(data=hidden, num_hidden=hidden_units[2])
+    hideen = mx.symbol.Activation(data=hidden, act_type='relu')
+    deep_out = mx.symbol.FullyConnected(data=hidden, num_hidden=2)
+
+    out = mx.symbol.Custom(linear_out+deep_out, label, op_type='weighted_softmax_ce_loss',
 
 Review comment:
   Yes, there are 24720 for the first class and 7841 for the second class. TF version doesn't use weighted ce loss.

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services