我在训练时遇到了以下错误：
“tensorflow.python.framework.errors_impl.InvalidArgumentError: Dense float feature must be a matrix.” 这个错误是在使用 tensorflow.contrib.boosted_trees.estimator_batch.estimator.GradientBoostedDecisionTreeClassifier 估计器时出现的。我使用的Tensorflow版本是1.4.0。如果我将估计器改为 tf.contrib.learn.DNNClassifier，同样的代码可以正常工作。在代码中，特征字典在 tf.contrib.learn.Experiment 的 “Train_input_fn” 中传递。
之前有人遇到过类似的错误吗？
#'tensorflow==1.4.0'import tensorflow as tfimport argparseimport sysimport osfrom tensorflow.contrib.boosted_trees.estimator_batch.estimator import GradientBoostedDecisionTreeClassifierfrom tensorflow.contrib.boosted_trees.proto import learner_pb2from tensorflow_transform.tf_metadata import metadata_iofrom tensorflow_transform.saved import input_fn_makerfrom tensorflow.contrib.learn.python.learn import learn_runnerRAW_METADATA_DIR="raw_metadata"CONTRACTED_METADATA_DIR="contracted_metadata"TRANSFORMED_METADATA_DIR="transformed_metadata"TRANSFORMED_TRAIN_DATA_FILE_PREFIX="train"TRANSFORMED_EVAL_DATA_FILE_PREFIX="eval"DATA_FILE_SUFFIX=".tfrecord.gz"TRANSFORM_FN_DIR="transform_fn"TARGET_FEATURE_COLUMN='target_field'FEATURE_NUMERICAL_COLUMN_NAMES = [ 'feature1', 'feature2', 'feature3', 'feature4', 'feature5' ]FEATURE_INTEGER_COLUMN_NAMES = [ # comment out fields that are not features 'feature6', 'feature7', 'feature8', 'feature9', 'feature10']def _parse_arguments(argv): """Parses command line arguments.""" parser = argparse.ArgumentParser( description="Runs training on data.") parser.add_argument( "--model_dir", required=True, type=str, help="The directory where model outputs will be written") parser.add_argument( "--input_dir", required=True, type=str, help=("GCS or local directory containing tensorflow-transform outputs.")) parser.add_argument( "--batch_size", default=30, required=False, type=int, help=("Batch size to use during training.")) parser.add_argument( "--num_epochs", default=100, required=False, type=int, help=("Number of epochs through the training set")) args, _ = parser.parse_known_args(args=argv[1:]) return argsdef get_eval_metrics(): return { "accuracy": tf.contrib.learn.MetricSpec( metric_fn=tf.contrib.metrics.streaming_accuracy, prediction_key=tf.contrib.learn.PredictionKey.CLASSES), "precision": tf.contrib.learn.MetricSpec( metric_fn=tf.contrib.metrics.streaming_precision, prediction_key=tf.contrib.learn.PredictionKey.CLASSES), "recall": tf.contrib.learn.MetricSpec( metric_fn=tf.contrib.metrics.streaming_recall, prediction_key=tf.contrib.learn.PredictionKey.CLASSES) }def read_and_decode_single_record(input_dir, num_epochs, mode=tf.contrib.learn.ModeKeys.TRAIN): if mode == tf.contrib.learn.ModeKeys.TRAIN: num_epochs = num_epochs file_prefix = TRANSFORMED_TRAIN_DATA_FILE_PREFIX else: num_epochs = 1 file_prefix = TRANSFORMED_EVAL_DATA_FILE_PREFIX transformed_metadata = metadata_io.read_metadata(os.path.join(input_dir, TRANSFORMED_METADATA_DIR)) input_file_names = tf.train.match_filenames_once(os.path.join(input_dir, '{}*{}'.format(file_prefix, DATA_FILE_SUFFIX))) filename_queue = tf.train.string_input_producer(input_file_names, num_epochs=num_epochs, shuffle=True) reader = tf.TFRecordReader(options=tf.python_io.TFRecordOptions( tf.python_io.TFRecordCompressionType.GZIP)) _, serialized_example = reader.read(filename_queue) features = tf.parse_single_example( serialized = serialized_example, features=transformed_metadata.schema.as_feature_spec() ) return featuresdef read_dataset(input_dir, num_epochs, batch_size, mode=tf.contrib.learn.ModeKeys.TRAIN): def _input_fn(): min_after_dequeue = 10000 features = read_and_decode_single_record(input_dir, num_epochs, mode) features = tf.train.shuffle_batch( tensors=features, batch_size=batch_size, min_after_dequeue=min_after_dequeue, capacity=(min_after_dequeue + 3) * batch_size) target = features.pop(TARGET_FEATURE_COLUMN) return features, target return _input_fndef specify_feature_columns(): feature_columns = [ tf.contrib.layers.real_valued_column(column_name = column_name) for column_name in FEATURE_NUMERICAL_COLUMN_NAMES] feature_columns.extend([ tf.contrib.layers.real_valued_column(column_name = column_name) for column_name in FEATURE_INTEGER_COLUMN_NAMES]) return feature_columnsdef build_estimator(model_dir, config, params): print "Using gradient boosted decision trees estimator \n" learner_config = learner_pb2.LearnerConfig() learner_config.learning_rate_tuner.fixed.learning_rate = 0.1 learner_config.regularization.l1 = 0.0 learner_config.regularization.l2 = 4.0 / params.batch_size learner_config.constraints.max_tree_depth = 4 learner_config.growing_mode = learner_pb2.LearnerConfig.WHOLE_TREE return GradientBoostedDecisionTreeClassifier( learner_config=learner_config, examples_per_layer=params.batch_size, num_trees=100, center_bias=False, feature_columns=specify_feature_columns() # feature_engineering_fn=feature_engineering_fn )def get_experiment_fn(args): config = tf.contrib.learn.RunConfig(save_checkpoints_steps=1000) def experiment_fn(output_dir): return tf.contrib.learn.Experiment( estimator = build_estimator(model_dir = output_dir, config = config, params = args), train_input_fn = read_dataset(args.input_dir, args.num_epochs, args.batch_size, mode=tf.contrib.learn.ModeKeys.TRAIN), eval_input_fn = read_dataset(args.input_dir, args.num_epochs, args.batch_size, mode=tf.contrib.learn.ModeKeys.EVAL), eval_metrics = get_eval_metrics()) return experiment_fndef run(args): learn_runner.run(get_experiment_fn(args), args.model_dir)if __name__ == '__main__': args = _parse_arguments(sys.argv) run(args)完整的错误跟踪：
WARNING:tensorflow:Using temporary folder as model directory: /var/folders/mg/sd4_qlyj4_lbh5ggfn6frvcr00fk8_/T/tmpPFhinsWARNING:tensorflow:From /Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/contrib/learn/python/learn/monitors.py:267: __init__ (from tensorflow.contrib.learn.python.learn.monitors) is deprecated and will be removed after 2016-12-05.Instructions for updating:Monitors are deprecated. Please use tf.train.SessionRunHook.WARNING:tensorflow:Casting <dtype: 'int64'> labels to bool.WARNING:tensorflow:Casting <dtype: 'int64'> labels to bool.WARNING:tensorflow:Error encountered when serializing resources.Type is unsupported, or the types of the items don't match field type in CollectionDef.'_Resource' object has no attribute 'name'2017-11-16 13:38:39.919664: I tensorflow/core/platform/cpu_feature_guard.cc:137] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2 AVX AVX2 FMAWARNING:tensorflow:Error encountered when serializing resources.Type is unsupported, or the types of the items don't match field type in CollectionDef.'_Resource' object has no attribute 'name'2017-11-16 13:38:48.810825: W tensorflow/core/framework/op_kernel.cc:1192] Invalid argument: Dense float feature must be a matrix.2017-11-16 13:38:48.810825: W tensorflow/core/framework/op_kernel.cc:1192] Invalid argument: Dense float feature must be a matrix.Traceback (most recent call last): File "./trainer/task.py", line 162, in <module> run(args) File "./trainer/task.py", line 157, in run learn_runner.run(get_experiment_fn(args), args.model_dir) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/contrib/learn/python/learn/learn_runner.py", line 218, in run return _execute_schedule(experiment, schedule) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/contrib/learn/python/learn/learn_runner.py", line 46, in _execute_schedule return task() File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/contrib/learn/python/learn/experiment.py", line 625, in train_and_evaluate self.train(delay_secs=0) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/contrib/learn/python/learn/experiment.py", line 367, in train hooks=self._train_monitors + extra_hooks) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/contrib/learn/python/learn/experiment.py", line 812, in _call_train monitors=hooks) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/python/util/deprecation.py", line 316, in new_func return func(*args, **kwargs) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/contrib/learn/python/learn/estimators/estimator.py", line 480, in fit loss = self._train_model(input_fn=input_fn, hooks=hooks) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/contrib/learn/python/learn/estimators/estimator.py", line 1040, in _train_model _, loss = mon_sess.run([model_fn_ops.train_op, model_fn_ops.loss]) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/python/training/monitored_session.py", line 521, in run run_metadata=run_metadata) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/python/training/monitored_session.py", line 892, in run run_metadata=run_metadata) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/python/training/monitored_session.py", line 967, in run raise six.reraise(*original_exc_info) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/python/training/monitored_session.py", line 952, in run return self._sess.run(*args, **kwargs) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/python/training/monitored_session.py", line 1024, in run run_metadata=run_metadata) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/python/training/monitored_session.py", line 827, in run return self._sess.run(*args, **kwargs) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 889, in run run_metadata_ptr) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 1120, in _run feed_dict_tensor, options, run_metadata) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 1317, in _do_run options, run_metadata) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 1336, in _do_call raise type(e)(node_def, op, message)tensorflow.python.framework.errors_impl.InvalidArgumentError: Dense float feature must be a matrix. [[Node: gbdt_1/GradientTreesPartitionExamples = GradientTreesPartitionExamples[num_dense_float_features=10, num_sparse_float_features=0, num_sparse_int_features=0, use_locking=true, _device="/job:localhost/replica:0/task:0/device:CPU:0"](ensemble_model, shuffle_batch:16, shuffle_batch:18, shuffle_batch:20, shuffle_batch:21, shuffle_batch:22, shuffle_batch:23, shuffle_batch:24, shuffle_batch:25, shuffle_batch:26, shuffle_batch:27, ^gbdt_1/TreeEnsembleStats)]]Caused by op u'gbdt_1/GradientTreesPartitionExamples', defined at: File "./trainer/task.py", line 162, in <module> run(args) File "./trainer/task.py", line 157, in run learn_runner.run(get_experiment_fn(args), args.model_dir) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/contrib/learn/python/learn/learn_runner.py", line 218, in run return _execute_schedule(experiment, schedule) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/contrib/learn/python/learn/learn_runner.py", line 46, in _execute_schedule return task() File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/contrib/learn/python/learn/experiment.py", line 625, in train_and_evaluate self.train(delay_secs=0) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/contrib/learn/python/learn/experiment.py", line 367, in train hooks=self._train_monitors + extra_hooks) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/contrib/learn/python/learn/experiment.py", line 812, in _call_train monitors=hooks) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/python/util/deprecation.py", line 316, in new_func return func(*args, **kwargs) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/contrib/learn/python/learn/estimators/estimator.py", line 480, in fit loss = self._train_model(input_fn=input_fn, hooks=hooks) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/contrib/learn/python/learn/estimators/estimator.py", line 986, in _train_model model_fn_ops = self._get_train_ops(features, labels) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/contrib/learn/python/learn/estimators/estimator.py", line 1202, in _get_train_ops return self._call_model_fn(features, labels, model_fn_lib.ModeKeys.TRAIN) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/contrib/learn/python/learn/estimators/estimator.py", line 1166, in _call_model_fn model_fn_results = self._model_fn(features, labels, **kwargs) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/contrib/boosted_trees/estimator_batch/model.py", line 98, in model_builder predictions_dict = gbdt_model.predict(mode) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/contrib/boosted_trees/python/training/functions/gbdt_batch.py", line 463, in predict ensemble_stamp, mode) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/contrib/boosted_trees/python/training/functions/gbdt_batch.py", line 392, in _predict_and_return_dict use_locking=True) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/contrib/boosted_trees/python/ops/gen_prediction_ops.py", line 117, in gradient_trees_partition_examples use_locking=use_locking, name=name) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/python/framework/op_def_library.py", line 787, in _apply_op_helper op_def=op_def) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 2956, in create_op op_def=op_def) File "/Users/amolsharma/anaconda/envs/oldpython/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 1470, in __init__ self._traceback = self._graph._extract_stack() # pylint: disable=protected-accessInvalidArgumentError (see above for traceback): Dense float feature must be a matrix. [[Node: gbdt_1/GradientTreesPartitionExamples = GradientTreesPartitionExamples[num_dense_float_features=10, num_sparse_float_features=0, num_sparse_int_features=0, use_locking=true, _device="/job:localhost/replica:0/task:0/device:CPU:0"](ensemble_model, shuffle_batch:16, shuffle_batch:18, shuffle_batch:20, shuffle_batch:21, shuffle_batch:22, shuffle_batch:23, shuffle_batch:24, shuffle_batch:25, shuffle_batch:26, shuffle_batch:27, ^gbdt_1/TreeEnsembleStats)]]回答：
我猜测由 tf.transform 创建的解析规格与我们通常得到的不同。您能分享 transformed_metadata.schema.as_feature_spec() 的输出吗？
作为一种解决方法，尝试在您的 input_fn 中添加以下这行代码，在 features = tf.train.shuffle_batch(…) 之后：
features = {feature_name: tf.reshape(feature_value, [-1, 1]) for feature_name, feature_value in features.items()}
