def _input_tensor_test(data_file, batch_size=100):
"""test for categorical_column and cross_column input."""
sess = tf.InteractiveSession()
features, labels = _CsvDataset(data_file).input_fn('train',
batch_size=batch_size)
print(features['age'].eval())
print(features['min_term_6m'].eval())
tag_list = tf.feature_column.categorical_column_with_hash_bucket(
'tag_list', 10000)
min_term_6m = tf.feature_column.categorical_column_with_hash_bucket(
'min_term_6m', 30)
cell_provinceXage = tf.feature_column.crossed_column(
['cell_province', 'age'], 600)
for f in [tag_list, min_term_6m, cell_provinceXage]:
# f_dense = tf.feature_column.indicator_column(f)
f_embed = tf.feature_column.embedding_column(f, 5)
input_tensor = tf.feature_column.input_layer(features, [f_embed])
sess.run(tf.global_variables_initializer())
# input_tensor = tf.feature_column.input_layer(features, [f_dense])
print('{} input tensor:\n {}'.format(f, input_tensor.eval()))
dense_tensor = tf.feature_column.input_layer(
features, [min_term_6m, tag_list, cell_provinceXage])
print('total input tensor:\n {}'.format(sess.run(dense_tensor)))
wide_columns, deep_columns = _build_model_columns()
dense_tensor = tf.feature_column.input_layer(features, deep_columns)
sess.run(tf.global_variables_initializer()
) # fix Attempting to use uninitialized value error.
sess.run(tf.tables_initializer()) # fix Table not initialized error.
print(sess.run(dense_tensor))
def main(_):
if FLAGS.model_version <= 0:
print('Please specify a positive value for version number.')
sys.exit(-1)
# tf.estimator.export.build_parsing_serving_input_receiver_fn
# tf.estimator.export.build_raw_serving_input_receiver_fn
# If these utilities do not meet your needs, you are free to write your own serving_input_receiver_fn()
# feature_spec = {
# "some_feature": tf.FixedLenFeature([], dtype=tf.string, default_value=""),
# "some_feature": tf.VarLenFeature(dtype=tf.string),
# }
#
# def _serving_input_receiver_fn():
# serialized_tf_example = tf.placeholder(dtype=tf.string, shape=None,
# name='input_example_tensor')
# # key (e.g. 'examples') should be same with the inputKey when you
# # buid the request for prediction
# receiver_tensors = {'examples': serialized_tf_example}
# features = tf.parse_example(serialized_tf_example, your_feature_spec)
# return tf.estimator.export.ServingInputReceiver(features, receiver_tensors)
# estimator.export_savedmodel(export_dir, _serving_input_receiver_fn)
wide_columns, deep_columns = _build_model_columns()
feature_columns = wide_columns + deep_columns
# for f in feature_columns:
# print(f._parse_example_spec)
# A dict mapping each feature key to a FixedLenFeature or VarLenFeature value.
feature_spec = tf.feature_column.make_parse_example_spec(feature_columns)
serving_input_receiver_fn = \
tf.estimator.export.build_parsing_serving_input_receiver_fn(feature_spec)
model_dir = os.path.join(model_base_dir, FLAGS.model_type)
export_dir = os.path.join(FLAGS.export_dir, FLAGS.model_type)
print("building custom estimator")
model = build_custom_estimator(model_dir, FLAGS.model_type)
print("exporting saved model")
model.export_savedmodel(export_dir,
serving_input_receiver_fn,
as_text=CONF['as_text'],
checkpoint_path=FLAGS.checkpoint_path)