demo

Author: formath

README.md

@@ -21,27 +21,35 @@ cmake ..
 make
 ```
 
-# Demo
+# Simple Demo
+This demo used `c=a+b` to show how to save the model and load it using C++ for prediction. [tensorflow_c++_api_prediction_basic](http://mathmach.com/2017/10/09/tensorflow_c++_api_prediction_basic/)
+```bash
+cd demo/simple_model
+# train
+sh train.sh
+# predict
+sh predict.sh
+```
 
-## Transform text file into TFRecord
+# Deep CTR Model Demo
+This demo show a real-wrold deep model usage in click through rate prediction. [tensorflow_c++_api_prediction_advance](http://mathmach.com/2017/10/11/tensorflow_c++_api_prediction_advance/)
+
+## Transform LibFM data into TFRecord
+* LibFM format: `label fieldId:featureId:value ...`
 ```bash
-cd demo
+cd demo/deep_model
 sh trans_data_to_tfrecord.sh
 cd ..
 ```
 
 ## Train model
 ```bash
-cd demo
 sh train.sh
-cd ..
 ```
 
 ## Predict using C++
 ```bash
-cd demo
-sh test.sh
-cd ..
+sh predict.sh
 ```
 
 # Reference

demo/libfm.data demo/deep_model/data/libfm.data

@@ -11,6 +11,6 @@
 #    --output_node_names=cross_entropy
 #cd -
 
-python ../python/freeze_graph.py \
-    --model_dir=./model \
+python ../../python/freeze_graph.py \
+    --model_dir=./saved_model \
     --output_node_names=Softmax

demo/freeze_graph.sh demo/deep_model/freeze_graph.sh

@@ -0,0 +1,3 @@
+#!/usr/bin/env bash
+
+# TODO

demo/deep_model/predict.sh

@@ -0,0 +1,9 @@
+#!/usr/bin/env bash
+
+python ../../python/train.py \
+    --dict "./data/dict.data" \
+    --continuous_fields "" \
+    --sparse_fields "9,6,116" \
+    --linear_fields "152,179" \
+    --train_file "./data/libfm.tfrecord" \
+    --valid_file "./data/libfm.tfrecord"

demo/deep_model/train.sh

@@ -0,0 +1,18 @@
+#!/usr/bin/env bash
+
+# generate field dict
+python ../../python/dict.py \
+	'0' \
+	'9,6,116' \
+	'152,179' \
+	./data/libfm.data \
+	./data/dict.data
+
+# transform libfm data into tfrecord
+python ../../python/data.py \
+	./data/dict.data \
+	'0' \
+	'9,6,116' \
+	'152,179' \
+	./data/libfm.data \
+	./data/libfm.tfrecord

demo/deep_model/trans_data_to_tfrecord.sh

@@ -0,0 +1,3 @@
+#!/usr/bin/env bash
+
+../../bin/simple_model.bin "./saved_model/graph.pb"

demo/simple_model/predict.sh

@@ -0,0 +1,3 @@
+#!/usr/bin/env bash
+
+python ../../python/simple_model.py

demo/simple_model/train.sh

@@ -47,7 +47,10 @@ def forward(self, sparse_id, sparse_val, linear_id, linear_val, continuous_val):
         self.hiddenW = []
         self.hiddenB = []
 
+        # sparse field embedding
         net = self.concat(self.sparse_field, sparse_id, sparse_val)
+
+        # concat sparse field embedding and continuous field
         if len(self.continuous_field) > 0:
             net = tf.concat([net, continuous_val], 1, name='concat_sparse_continuous')
 
@@ -63,7 +66,7 @@ def forward(self, sparse_id, sparse_val, linear_id, linear_val, continuous_val):
             #net = tf.nn.dropout(net, self.drop_out, name='dropout_'+str(i))
             #tf.summary.histogram('hidden_w' + str(i), weight)
 
-        # merge linear sparse
+        # merge linear sparse field embedding
         if len(self.linear_field) > 0:
             linear_embedding = self.concat(self.linear_field, linear_id, linear_val)
             net = tf.concat([net, linear_embedding], 1, name='concat_linear')

demo/train.sh

@@ -14,4 +14,4 @@
     print(b.eval()) # 6.0
     print(c.eval()) # 30.0
 
-    tf.train.write_graph(sess.graph_def, 'simple_model/', 'graph.pb', as_text=False)
+    tf.train.write_graph(sess.graph_def, 'saved_model/', 'graph.pb', as_text=False)

demo/trans_data_to_tfrecord.sh

@@ -4,7 +4,7 @@
 import os
 import datetime
 import tensorflow as tf
-from model import Model
+from deep_model import Model
 from data import Data
 
 # config
@@ -16,9 +16,8 @@
 flags.DEFINE_integer("valid_batch_size", 100, "validate set batch size")
 flags.DEFINE_integer("thread_num", 1, "number of thread to read data")
 flags.DEFINE_integer("min_after_dequeue", 100, "min_after_dequeue for shuffle queue")
-flags.DEFINE_string("model_dir", "./model/", "model dirctory")
+flags.DEFINE_string("model_dir", "./saved_model/", "model dirctory")
 flags.DEFINE_string("tensorboard_dir", "./tensorboard/", "summary data saved for tensorboard")
-flags.DEFINE_string("model_type", "wide_and_deep", "model type, option: wide, deep, wide_and_deep")
 flags.DEFINE_string("optimizer", "adagrad", "optimization algorithm")
 flags.DEFINE_integer('steps_to_validate', 1, 'steps to validate and print')
 flags.DEFINE_bool("train_from_checkpoint", False, "reload model from checkpoint and go on training")

python/model.py python/deep_model.py

@@ -19,7 +19,8 @@ int main(int argc, char* argv[]) {
   // when using `bazel run` since the cwd isn't where you call
   // `bazel run` but from inside a temp folder.)
   GraphDef graph_def;
-  status = ReadBinaryProto(Env::Default(), "../demo/simple_model/graph.pb", &graph_def);
+  std::string model_path = argv[1];
+  status = ReadBinaryProto(Env::Default(), model_path, &graph_def);
   if (!status.ok()) {
     std::cout << status.ToString() << std::endl;
     return 1;