I am trying to use a custom Estimator together with a feed_dict. Based on several relevant questions, such as this one, I have arrived at the following code. Note that I am returning dataset from input_fn, and not next_example, next_label. The error is
tensorflow.python.framework.errors_impl.InvalidArgumentError: You must feed a value for placeholder tensor 'Placeholder' with dtype float and shape [?,784]
[[{{node Placeholder}}]]
and the full stack trace is further down.
I am missing some fundamental concept about how where exactly how the data X and y are being fed to the graph. Can anyone shed light on what I am doing wrong? Thanks!
Code:
import numpy as np
import tensorflow as tf
class IteratorInitializerHook(tf.compat.v1.train.SessionRunHook):
def __init__(self):
super(IteratorInitializerHook, self).__init__()
self.iterator_initializer_func = None # Will be set in the input_fn
def after_create_session(self, session, coord):
# Initialize the iterator with the data feed_dict
self.iterator_initializer_func(session)
def get_inputs(X, y):
iterator_initializer_hook = IteratorInitializerHook()
def input_fn():
X_pl = tf.compat.v1.placeholder(X.dtype, [None, X.shape[1]])
y_pl = tf.compat.v1.placeholder(y.dtype, [None, y.shape[1]])
dataset = tf.compat.v1.data.Dataset.from_tensor_slices((X_pl, y_pl))
iterator = dataset.make_initializable_iterator()
dataset = dataset.batch(32)
iterator_initializer_hook.iterator_initializer_func =
lambda sess: sess.run(iterator.initializer,
feed_dict={X_pl: X, y_pl: y})
return dataset
return input_fn, iterator_initializer_hook
class MyMnist:
def __init__(self, params, **kwargs):
self.loss = 0
self.optimizer = tf.compat.v1.train.AdamOptimizer()
self.W = tf.compat.v1.Variable(tf.zeros([784,10]), trainable=True,
name="W")
self.b = tf.compat.v1.Variable(tf.zeros([10]), trainable=True,
name="b")
def build_model(self, features, labels, mode):
"""
Build model and return output
"""
is_training = mode == tf.estimator.ModeKeys.TRAIN
output = tf.compat.v1.nn.softmax(
tf.matmul(features, self.W) + self.b
)
return output
def build_total_loss(self, model_outputs, features, labels, mode):
"""
Return computed loss
"""
loss = tf.compat.v1.losses.softmax_cross_entropy(
labels, model_outputs
)
return loss
def build_optimizer(self):
"""
Setup the optimizer.
:returns: The optimizer
"""
print("build_optimizer")
lr = 0.01
optimizer = tf.compat.v1.train.AdamOptimizer(
learning_rate=lr,
name="Adam"
)
return optimizer
def build_train_ops(self, loss):
"""
Setup optimizer and build train ops.
:param Tensor loss: The loss tensor
:return: Train ops
"""
print("build_train_ops")
self.optimizer = self.build_optimizer()
return self.optimizer.minimize(
loss, global_step=tf.compat.v1.train.get_global_step()
)
def model_fn(features, labels, mode, params):
print('model_fn')
model = MyMnist(params)
output = model.build_model(features, labels, mode)
loss = model.build_total_loss(output, features, labels, mode)
if mode == tf.estimator.ModeKeys.TRAIN:
train_op = model.build_train_ops(loss)
log_hook =
tf.compat.v1.train.LoggingTensorHook(
{"W is": model.W, "b is": model.b},
every_n_iter=1)
return tf.estimator.EstimatorSpec(
mode=mode, loss=loss, train_op=train_op,
training_hooks=[log_hook]
)
def train(argv=None):
print("train")
params = { 'mode': 'train', 'model_dir': './model_dir',
'training': {'steps': 10 },
'size': 100
}
est = tf.estimator.Estimator(
model_fn,
model_dir=params['model_dir'],
params=params,
)
(X_train, l_train), (X_test, l_test) = tf.keras.datasets.mnist.load_data()
y_train = np.zeros((l_train.shape[0], l_train.max()+1), dtype=np.float32)
y_train[np.arange(l_train.shape[0]), l_train] = 1
y_test = np.zeros((l_test.shape[0], l_test.max()+1), dtype=np.float32)
y_test[np.arange(l_test.shape[0]), l_test] = 1
X_train = X_train.reshape((X_train.shape[0],-1)).astype(np.float32)
X_test = X_test.reshape((X_test.shape[0],-1))
train_input_fn, train_iterator_initializer_hook =
get_inputs(X_train, y_train)
test_input_fn, test_iterator_initializer_hook = get_inputs(X_test, y_test)
if params['mode'] == 'train':
est.train(
input_fn=train_input_fn,
hooks=[train_iterator_initializer_hook],
steps=params['training']['steps']
)
if __name__ == "__main__":
tf.compat.v1.disable_eager_execution()
tf.compat.v1.app.run(main=train)
Stack trace:
$ python main.py
train
INFO:tensorflow:Using default config.
I0128 18:51:12.061203 139970842568512 estimator.py:1822] Using default config.
INFO:tensorflow:Using config: {'_model_dir': './model_dir', '_tf_random_seed': None, '_save_summary_steps': 100, '_save_checkpoints_steps': None, '_save_checkpoints_secs': 600, '_session_config': allow_soft_placement: true
graph_options {
rewrite_options {
meta_optimizer_iterations: ONE
}
}
, '_keep_checkpoint_max': 5, '_keep_checkpoint_every_n_hours': 10000, '_log_step_count_steps': 100, '_train_distribute': None, '_device_fn': None, '_protocol': None, '_eval_distribute': None, '_experimental_distribute': None, '_experimental_max_worker_delay_secs': None, '_session_creation_timeout_secs': 7200, '_service': None, '_cluster_spec': ClusterSpec({}), '_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}
I0128 18:51:12.061676 139970842568512 estimator.py:191] Using config: {'_model_dir': './model_dir', '_tf_random_seed': None, '_save_summary_steps': 100, '_save_checkpoints_steps': None, '_save_checkpoints_secs': 600, '_session_config': allow_soft_placement: true
graph_options {
rewrite_options {
meta_optimizer_iterations: ONE
}
}
, '_keep_checkpoint_max': 5, '_keep_checkpoint_every_n_hours': 10000, '_log_step_count_steps': 100, '_train_distribute': None, '_device_fn': None, '_protocol': None, '_eval_distribute': None, '_experimental_distribute': None, '_experimental_max_worker_delay_secs': None, '_session_creation_timeout_secs': 7200, '_service': None, '_cluster_spec': ClusterSpec({}), '_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}
WARNING:tensorflow:From /srv/scratch/packages/spack/opt/spack/linux-rhel8-skylake_avx512/gcc-8.3.1/anaconda3-2020.07-weugqkfkxd6zmn2irm7lpmujzczwebiw/envs/graphsaint_env/lib/python3.8/site-packages/tensorflow/python/training/training_util.py:235: Variable.initialized_value (from tensorflow.python.ops.variables) is deprecated and will be removed in a future version.
Instructions for updating:
Use Variable.read_value. Variables in 2.X are initialized automatically both in eager and graph (inside tf.defun) contexts.
W0128 18:51:12.383132 139970842568512 deprecation.py:317] From /srv/scratch/packages/spack/opt/spack/linux-rhel8-skylake_avx512/gcc-8.3.1/anaconda3-2020.07-weugqkfkxd6zmn2irm7lpmujzczwebiw/envs/graphsaint_env/lib/python3.8/site-packages/tensorflow/python/training/training_util.py:235: Variable.initialized_value (from tensorflow.python.ops.variables) is deprecated and will be removed in a future version.
Instructions for updating:
Use Variable.read_value. Variables in 2.X are initialized automatically both in eager and graph (inside tf.defun) contexts.
WARNING:tensorflow:From main.py:21: DatasetV1.make_initializable_iterator (from tensorflow.python.data.ops.dataset_ops) is deprecated and will be removed in a future version.
Instructions for updating:
This is a deprecated API that should only be used in TF 1 graph mode and legacy TF 2 graph mode available through `tf.compat.v1`. In all other situations -- namely, eager mode and inside `tf.function` -- you can consume dataset elements using `for elem in dataset: ...` or by explicitly creating iterator via `iterator = iter(dataset)` and fetching its elements via `values = next(iterator)`. Furthermore, this API is not available in TF 2. During the transition from TF 1 to TF 2 you can use `tf.compat.v1.data.make_initializable_iterator(dataset)` to create a TF 1 graph mode style iterator for a dataset created through TF 2 APIs. Note that this should be a transient state of your code base as there are in general no guarantees about the interoperability of TF 1 and TF 2 code.
W0128 18:51:12.397508 139970842568512 deprecation.py:317] From main.py:21: DatasetV1.make_initializable_iterator (from tensorflow.python.data.ops.dataset_ops) is deprecated and will be removed in a future version.
Instructions for updating:
This is a deprecated API that should only be used in TF 1 graph mode and legacy TF 2 graph mode available through `tf.compat.v1`. In all other situations -- namely, eager mode and inside `tf.function` -- you can consume dataset elements using `for elem in dataset: ...` or by explicitly creating iterator via `iterator = iter(dataset)` and fetching its elements via `values = next(iterator)`. Furthermore, this API is not available in TF 2. During the transition from TF 1 to TF 2 you can use `tf.compat.v1.data.make_initializable_iterator(dataset)` to create a TF 1 graph mode style iterator for a dataset created through TF 2 APIs. Note that this should be a transient state of your code base as there are in general no guarantees about the interoperability of TF 1 and TF 2 code.
INFO:tensorflow:Calling model_fn.
I0128 18:51:12.404662 139970842568512 estimator.py:1162] Calling model_fn.
model_fn
build_train_ops
build_optimizer
INFO:tensorflow:Done calling model_fn.
I0128 18:51:12.476285 139970842568512 estimator.py:1164] Done calling model_fn.
INFO:tensorflow:Create CheckpointSaverHook.
I0128 18:51:12.477094 139970842568512 basic_session_run_hooks.py:546] Create CheckpointSaverHook.
data size:0.033848 MB
data size:0.033848 MB
INFO:tensorflow:Graph was finalized.
I0128 18:51:12.530334 139970842568512 monitored_session.py:246] Graph was finalized.
2021-01-28 18:51:12.530598: I tensorflow/core/platform/cpu_feature_guard.cc:142] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN)to use the following CPU instructions in performance-critical operations: SSE4.1 SSE4.2 AVX AVX2 AVX512F FMA
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
2021-01-28 18:51:12.539575: I tensorflow/core/platform/profile_utils/cpu_utils.cc:104] CPU Frequency: 2400000000 Hz
2021-01-28 18:51:12.543713: I tensorflow/compiler/xla/servic
