add FNN model

model/FNN/FNN.py

@@ -0,0 +1,77 @@
+"""
+paper
+
+Weinan Zhang, Tianming Du, and Jun Wang. Deep learning over multi-field categorical data - - A case study on user response
+
+"""
+
+
+from model.FNN.preprocess import build_features
+import tensorflow as tf
+import numpy as np
+from config import *
+from utils import tf_estimator_model, add_layer_summary
+
+@tf_estimator_model
+def model_fn(features, labels, mode, params):
+    feature_columns= build_features()
+
+    input = tf.feature_column.input_layer(features, feature_columns)
+
+    with tf.variable_scope('init_fm_embedding'):
+        # method1: load from checkpoint directly
+        embeddings = tf.Variable( tf.contrib.framework.load_variable(
+            './checkpoint/FM',
+            'fm_interaction/v'
+        ) )
+        add_layer_summary('fm_init_embeddings', embeddings)
+        dense = tf.matmul(input, embeddings)
+        add_layer_summary('input', dense)
+
+    with tf.variable_scope( 'Dense' ):
+        for i, unit in enumerate( params['hidden_units'] ):
+            dense = tf.layers.dense( dense, units=unit, activation='relu', name='dense{}'.format( i ) )
+            dense = tf.layers.batch_normalization( dense, center=True, scale=True, trainable=True,
+                                                   training=(mode == tf.estimator.ModeKeys.TRAIN) )
+            dense = tf.layers.dropout( dense, rate=params['dropout_rate'],
+                                       training=(mode == tf.estimator.ModeKeys.TRAIN) )
+            add_layer_summary( dense.name, dense )
+
+    with tf.variable_scope('output'):
+        y = tf.layers.dense(dense, units= 1, name = 'output')
+        tf.summary.histogram(y.name, y)
+
+    return y
+
+def build_estimator(model_dir):
+
+    run_config = tf.estimator.RunConfig(
+        save_summary_steps=50,
+        log_step_count_steps=50,
+        keep_checkpoint_max=3,
+        save_checkpoints_steps=50
+    )
+
+    # load pretrained FM embedding layer to initialize the embedding layer in FNN
+    estimator = tf.estimator.Estimator(
+        model_fn = model_fn,
+        config = run_config,
+        params = {
+            'learning_rate' :0.002,
+            'hidden_units':[20,10],
+            'dropout_rate':0.1,
+        },
+        model_dir= model_dir
+    )
+
+    return estimator
+
+
+# check name of all the tensor in the checkpoint
+
+if __name__ == '__main__':
+    print ('checking name of all the tensor in the FNN pretrain checkpoint')
+    from tensorflow.python.tools.inspect_checkpoint import  print_tensors_in_checkpoint_file
+    latest_ckp = tf.train.latest_checkpoint('./checkpoint/FM')
+    print_tensors_in_checkpoint_file( latest_ckp, all_tensors=True )
+    print_tensors_in_checkpoint_file(latest_ckp, all_tensors=False, tensor_name='fm_interaction/v' )

model/FNN/preprocess.py

@@ -0,0 +1,18 @@
+from config import *
+import tensorflow as tf
+
+def build_features():
+    f_sparse = []
+
+    for col, config in EMB_CONFIGS.items():
+        ind = tf.feature_column.categorical_column_with_hash_bucket(col, hash_bucket_size = config['hash_size'])
+        one_hot = tf.feature_column.indicator_column(ind)
+        f_sparse.append(one_hot)
+
+    for col, config in BUCKET_CONFIGS.items():
+        num = tf.feature_column.numeric_column( col )
+        bucket = tf.feature_column.bucketized_column( num, boundaries=config )
+        f_sparse.append( bucket )
+
+    return f_sparse
+