tensorflow feed list feature (multi-hot) to tf.estimator

Question

Some of the feature columns' data type is list . And their length can be different. I want to encode this column as a multi-hot categorical feature and feed it to tf.estimator . I tried the following but the error Unable to get element as bytes showed. I think this is a common practice in deep learning especially recommendation systems eg Deep & Wide model. I found a relevant question here but it doesn't show how to feed to the estimator.

import pandas as pd
import tensorflow as tf

OUTDIR = "./data"

data = {"x": [["a", "c"], ["a", "b"], ["b", "c"]], "y": ["x", "y", "z"]}
df = pd.DataFrame(data)

Y = df["y"]
X = df.drop("y", axis=1)

indicator_features = [
    tf.feature_column.indicator_column(
        categorical_column=tf.feature_column.categorical_column_with_vocabulary_list(
            key="x", vocabulary_list=["a", "b", "c"]
        )
    )
]

model = tf.estimator.LinearClassifier(
    feature_columns=indicator_features, model_dir=OUTDIR
)

training_input_fn = tf.estimator.inputs.pandas_input_fn(
    x=X, y=Y, batch_size=64, shuffle=True, num_epochs=None
)

model.train(input_fn=training_input_fn)

The following error:

INFO:tensorflow:Using default config. INFO:tensorflow:Using config: {'_model_dir': 'testalg', '_tf_random_seed': None, '_save_summary_steps': 100, '_save_checkpoints_steps': None, '_save_checkpoints_secs': 600, '_session_config': None, '_keep_checkpoint_max': 5, '_keep_checkpoint_every_n_hours': 10000, '_log_step_count_steps': 100, '_train_distribute': None, '_device_fn': None, '_service': None, '_cluster_spec': , '_task_type': 'worker', '_task_id': 0, '_global_id_in_cluster': 0, '_master': '', '_evaluation_master': '', '_is_chief': True, '_num_ps_replicas': 0, '_num_worker_replicas': 1} INFO:tensorflow:Calling model_fn. INFO:tensorflow:Done calling model_fn. INFO:tensorflow:Create CheckpointSaverHook. INFO:tensorflow:Graph was finalized. INFO:tensorflow:Running local_init_op. INFO:tensorflow:Done running local_init_op. INFO:tensorflow:Error reported to Coordinator: , Unable to get element as bytes. INFO:tensorflow:Saving checkpoints for 0 into testalg/model.ckpt. ------------------------------------------------------- InternalError Traceback (most recent call last) /home/yinan.li1/anaconda3/lib/python3.6/site-packages/tensorflow/python/client/session.py in _do_call(self, fn, *args) 1321 try: -> 1322 return fn(*args) 1323 except errors.OpError as e:

/home/yinan.li1/anaconda3/lib/python3.6/site-packages/tensorflow/python/client/session.py in _run_fn(feed_dict, fetch_list, target_list, options, run_metadata) 1306 return self._call_tf_sessionrun( -> 1307 options, feed_dict, fetch_list, target_list, run_metadata) 1308

/home/yinan.li1/anaconda3/lib/python3.6/site-packages/tensorflow/python/client/session.py in _call_tf_sessionrun(self, options, feed_dict, fetch_list, target_list, run_metadata) 1408 self._session, options, feed_dict, fetch_list, target_list, -> 1409 run_metadata) 1410 else:

InternalError: Unable to get element as bytes.

During handling of the above exception, another exception occurred:

InternalError Traceback (most recent call last) in () 44 45 ---> 46 model.train(input_fn=training_input_fn)

/home/yinan.li1/anaconda3/lib/python3.6/site-packages/tensorflow/python/estimator/estimator.py in train(self, input_fn, hooks, steps, max_steps, saving_listeners) 364 365 saving_listeners = _check_listeners_type(saving_listeners) --> 366 loss = self._train_model(input_fn, hooks, saving_listeners) 367 logging.info('Loss for final step: %s.', loss) 368 return self

/home/yinan.li1/anaconda3/lib/python3.6/site-packages/tensorflow/python/estimator/estimator.py in _train_model(self, input_fn, hooks, saving_listeners) 1117
return self._train_model_distributed(input_fn, hooks, saving_listeners) 1118 else: -> 1119 return self._train_model_default(input_fn, hooks, saving_listeners) 1120 1121 def _train_model_default(self, input_fn, hooks, saving_listeners):

/home/yinan.li1/anaconda3/lib/python3.6/site-packages/tensorflow/python/estimator/estimator.py in _train_model_default(self, input_fn, hooks, saving_listeners)
1133 return self._train_with_estimator_spec(estimator_spec, worker_hooks, 1134
hooks, global_step_tensor, -> 1135 saving_listeners) 1136 1137 def _train_model_distributed(self, input_fn, hooks, saving_listeners):

/home/yinan.li1/anaconda3/lib/python3.6/site-packages/tensorflow/python/estimator/estimator.py in _train_with_estimator_spec(self, estimator_spec, worker_hooks, hooks, global_step_tensor, saving_listeners) 1334 loss = None 1335 while not mon_sess.should_stop(): -> 1336 _, loss = mon_sess.run([estimator_spec.train_op, estimator_spec.loss]) 1337 return loss 1338

/home/yinan.li1/anaconda3/lib/python3.6/site-packages/tensorflow/python/training/monitored_session.py in exit (self, exception_type, exception_value, traceback) 687 if exception_type in [errors.OutOfRangeError, StopIteration]: 688 exception_type = None --> 689 self._close_internal(exception_type) 690 # exit should return True to suppress an exception. 691 return exception_type is None

/home/yinan.li1/anaconda3/lib/python3.6/site-packages/tensorflow/python/training/monitored_session.py in _close_internal(self, exception_type) 724 if self._sess is None: 725 raise RuntimeError('Session is already closed.') --> 726 self._sess.close() 727 finally: 728 self._sess = None

/home/yinan.li1/anaconda3/lib/python3.6/site-packages/tensorflow/python/training/monitored_session.py in close(self) 972 if self._sess: 973 try: --> 974 self._sess.close() 975 except _PREEMPTION_ERRORS: 976 pass

/home/yinan.li1/anaconda3/lib/python3.6/site-packages/tensorflow/python/training/monitored_session.py in close(self) 1116 self._coord.join( 1117
stop_grace_period_secs=self._stop_grace_period_secs, -> 1118 ignore_live_threads=True) 1119 finally: 1120 try:

/home/yinan.li1/anaconda3/lib/python3.6/site-packages/tensorflow/python/training/coordinator.py in join(self, threads, stop_grace_period_secs, ignore_live_threads) 387 self._registered_threads = set() 388 if self._exc_info_to_raise: --> 389 six.reraise(*self._exc_info_to_raise) 390 elif stragglers: 391 if ignore_live_threads:

/home/yinan.li1/anaconda3/lib/python3.6/site-packages/six.py in reraise(tp, value, tb) 683 value = tp() 684 if value. traceback is not tb: --> 685 raise value.with_traceback(tb) 686 raise value 687

/home/yinan.li1/anaconda3/lib/python3.6/site-packages/tensorflow/python/estimator/inputs/queues/feeding_queue_runner.py in _run(self, sess, enqueue_op, feed_fn, coord) 92 try: 93 feed_dict = None if feed_fn is None else feed_fn() ---> 94 sess.run(enqueue_op, feed_dict=feed_dict) 95 except (errors.OutOfRangeError, errors.CancelledError): 96 # This exception indicates that a queue was closed.

/home/yinan.li1/anaconda3/lib/python3.6/site-packages/tensorflow/python/client/session.py in run(self, fetches, feed_dict, options, run_metadata) 898 try: 899 result = self._run(None, fetches, feed_dict, options_ptr, --> 900 run_metadata_ptr) 901 if run_metadata: 902 proto_data = tf_session.TF_GetBuffer(run_metadata_ptr)

/home/yinan.li1/anaconda3/lib/python3.6/site-packages/tensorflow/python/client/session.py in _run(self, handle, fetches, feed_dict, options, run_metadata)
1133 if final_fetches or final_targets or (handle and feed_dict_tensor): 1134 results = self._do_run(handle, final_targets, final_fetches, -> 1135 feed_dict_tensor, options, run_metadata) 1136 else: 1137 results = []

/home/yinan.li1/anaconda3/lib/python3.6/site-packages/tensorflow/python/client/session.py in _do_run(self, handle, target_list, fetch_list, feed_dict, options, run_metadata) 1314 if handle is None: 1315 return self._do_call(_run_fn, feeds, fetches, targets, options, -> 1316 run_metadata) 1317 else: 1318 return self._do_call(_prun_fn, handle, feeds, fetches)

/home/yinan.li1/anaconda3/lib/python3.6/site-packages/tensorflow/python/client/session.py in _do_call(self, fn, *args) 1333 except KeyError: 1334 pass -> 1335 raise type(e)(node_def, op, message) 1336 1337 def _extend_graph(self):

InternalError: Unable to get element as bytes.

Answer 1

I think one of the problems in your case is that the type of column in pandas is actually object not string. If you converted that to separate string columns you would get rid of this error. Keep in mind that The basic TensorFlow tf.string dtype allows you to build tensors of byte strings. and as you are storing object in this column not a string you get an error.

The code below will get overcome the error you get above, but it will not solve your problem completely. The variable lenght of lists would have to be dealt with by padding or the list or something similar as again indicator_column might have problems dealing with missing values.

X2= pd.DataFrame(X['x'].values.tolist(), columns=['x1','x2'])

feat1 = tf.feature_column.categorical_column_with_vocabulary_list(
            key="x1", vocabulary_list=["a", "b", "c"]
        )
feat2 = tf.feature_column.categorical_column_with_vocabulary_list(
            key="x2", vocabulary_list=["a", "b", "c"]
        )
indicator_features = [
    tf.feature_column.indicator_column(
        categorical_column=feat1
    ),tf.feature_column.indicator_column(
        categorical_column=feat2
    )
]

training_input_fn = tf.estimator.inputs.pandas_input_fn(
    x=X2, y=Y, batch_size=64, shuffle=True, num_epochs=None
)