def model_fn(features_map, targets, mode):
if mode == tf.contrib.learn.ModeKeys.TRAIN:
predictions, loss = model_impl(hparams, mode, features_map,
targets)
if hparams.l1_reg > 0. or hparams.l2_reg > 0.:
# apply regularization
with tf.variable_scope("reg") as vs:
all_regularize = []
if hparams.l1_reg > 0.:
all_regularize.append(
tf.contrib.layers.l1_regularizer(hparams.l1_reg))
if hparams.l2_reg > 0.:
all_regularize.append(
tf.contrib.layers.l2_regularizer(hparams.l2_reg))
regularizer = fu.sum_regularizer(all_regularize, scope=vs)
regularization_penalty = tf.contrib.layers.apply_regularization(
regularizer)
loss += regularization_penalty
train_op = create_train_op(loss, hparams)
return model_fn_lib.ModelFnOps(mode=mode,
predictions={
'predictions': predictions,
'features':
features_map['features'],
'targets': targets
},
loss=loss,
train_op=train_op)
if mode == tf.contrib.learn.ModeKeys.INFER:
predictions, loss = model_impl(hparams, mode, features_map, None)
return model_fn_lib.ModelFnOps(mode=mode,
predictions={
'predictions': predictions,
'features':
features_map['features'],
'targets':
features_map['targets']
})
if mode == tf.contrib.learn.ModeKeys.EVAL:
predictions, loss = model_impl(hparams, mode, features_map,
targets)
return model_fn_lib.ModelFnOps(mode=mode,
predictions={
'predictions': predictions,
'features':
features_map['features'],
'targets': targets
},
loss=loss)
def model_fn(features, targets, mode, params):
# Features are board[81] + liberty[81] + group[81] + valid_move[81] + last_move[1] + ko[2].
# Note that turn and ko info are set to valid_move map.
board_features, _, _ = tf.split(features, [81 * 4, 1, 2], axis=1)
board = tf.reshape(board_features, [-1, 9, 9, 4])
# No relu, input includes negative. 4x25x64 = 6400
conv_in = tf.layers.conv2d(inputs=board,
filters=64,
kernel_size=[5, 5],
padding="same")
conv1 = tf.layers.conv2d(inputs=conv_in,
filters=16,
kernel_size=[3, 3],
padding="same") # 64*9*16 = 9216
conv2 = tf.layers.conv2d(inputs=conv1,
filters=16,
kernel_size=[3, 3],
padding="same") # 16*9*16 = 2304
conv_out = tf.layers.conv2d(inputs=conv2,
filters=1,
kernel_size=[1, 1],
padding="same") # To reduce size
# Flattens conv2d output and attaches last_move info.
conv_flat = tf.reshape(conv_out, [-1, 9 * 9 * 1])
# Dense layer and output.
# 81*64 = 5184
dense = tf.layers.dense(inputs=conv_flat, units=64)
output_layer = tf.contrib.layers.linear(dense, 1)
predictions = tf.reshape(output_layer, [-1])
# For predict mode.
if targets == None:
return model_fn_lib.ModelFnOps(mode=mode, predictions=predictions)
# For modes with targets.
loss = tf.losses.mean_squared_error(targets, predictions)
eval_metric_ops = {
"rmse":
tf.metrics.root_mean_squared_error(tf.cast(targets, tf.float32),
predictions)
}
train_op = tf.contrib.layers.optimize_loss(
loss=loss,
global_step=tf.contrib.framework.get_global_step(),
learning_rate=params["learning_rate"],
optimizer="SGD")
return model_fn_lib.ModelFnOps(mode=mode,
predictions=predictions,
loss=loss,
train_op=train_op,
eval_metric_ops=eval_metric_ops)
def model_fn(features, labels, params, mode):
#get inputs, targets from features, labels
source_tokens = tf.placeholder_with_default(
features["source_tokens"],
shape=[None, None],
name='sentence_inputs')
source_tokens_length = tf.placeholder_with_default(
features["source_len"],
shape=[None],
name='source_tokens_length')
if labels != None:
target_tokens = tf.placeholder_with_default(
labels["target_tokens"],
shape=[None, None],
name='targets')
target_tokens_length = tf.placeholder_with_default(
labels["target_len"],
shape=[None],
name='targets_token_length')
else:
target_tokens = None
target_tokens_length = None
# add placeholders
self._add_placeholders(source_tokens, source_tokens_length,
target_tokens, target_tokens_length, mode)
# set input_ids, target_ids, max_target_len
self._set_variables(mode)
# add an encoder to computation graph
self._add_encoder()
# add a decoder
self._add_decoder(mode)
if mode == tf.contrib.learn.ModeKeys.TRAIN: # train
return model_fn_lib.ModelFnOps(
mode=mode,
predictions={'predictions': self.predictions},
loss=self.loss,
train_op=self.train_op)
else:
return model_fn_lib.ModelFnOps(
mode=mode,
predictions={
'predictions':
self.target_id_to_vocab.lookup(
tf.to_int64(self.predictions)),
'source_tokens':
self.source_tokens
},
loss=self.loss)
def _model_fn(features, labels, mode):
"""Creates the prediction and its loss.
Args:
features: A dictionary of tensors keyed by the feature name.
labels: A tensor representing the labels.
mode: The execution mode, as defined in tf.contrib.learn.ModeKeys.
Returns:
A tuple consisting of the prediction, loss, and train_op.
"""
# Generate one embedding per sparse feature column and concatenate them.
concat_embeddings = tf.contrib.layers.input_from_feature_columns(
columns_to_tensors=features,
feature_columns=_get_feature_columns(include_target_column=False))
# Add one hidden layer.
hidden_layer_0 = tf.contrib.layers.relu(concat_embeddings,
FLAGS.hidden_units)
# Output and logistic loss.
logits = tf.contrib.layers.linear(hidden_layer_0, FLAGS.num_classes)
predictions = tf.contrib.layers.softmax(logits)
if mode == tf.contrib.learn.ModeKeys.INFER:
predictions = {
tf.contrib.learn.PredictionKey.PROBABILITIES: predictions,
PREDICTION_KEY: features[PREDICTION_KEY]
}
output_alternatives = {
DEFAULT_OUTPUT_ALTERNATIVE:
(tf.contrib.learn.ProblemType.UNSPECIFIED, predictions)
}
return model_fn.ModelFnOps(mode=mode,
predictions=predictions,
output_alternatives=output_alternatives)
target_one_hot = tf.one_hot(labels, FLAGS.num_classes)
target_one_hot = tf.reduce_sum(input_tensor=target_one_hot,
reduction_indices=[1])
loss = tf.losses.softmax_cross_entropy(target_one_hot, logits)
if mode == tf.contrib.learn.ModeKeys.EVAL:
return predictions, loss, None
opt = tf.train.MomentumOptimizer(FLAGS.learning_rate, FLAGS.momentum)
train_op = tf.contrib.layers.optimize_loss(
loss=loss,
global_step=tf.contrib.framework.get_global_step(),
learning_rate=FLAGS.learning_rate,
optimizer=opt)
return model_fn.ModelFnOps(mode=mode,
predictions=predictions,
loss=loss,
train_op=train_op)
def _model_fn(self, features, labels, mode):
x_reshaped = tf.reshape(features, [-1, 28, 28, 1])
h1 = tf.layers.conv2d(x_reshaped,
32, [5, 5],
padding="same",
activation=tf.nn.relu)
h2 = tf.layers.max_pooling2d(h1, 2, 2, padding="same")
h3 = tf.layers.conv2d(h2,
64, [5, 5],
padding="same",
activation=tf.nn.relu)
h4 = tf.layers.max_pooling2d(h3, 2, 2, padding="same")
h5 = tf.layers.dense(tf.reshape(h4, [-1, 7 * 7 * 64]),
1024,
activation=tf.nn.relu)
h6 = tf.layers.dropout(h5,
rate=0.3,
training=(mode == learn.ModeKeys.TRAIN))
ypred = tf.layers.dense(h6, 10)
predictions = {
"classes": tf.argmax(ypred, axis=1),
"probabilities": tf.nn.softmax(ypred, dim=1),
"logits": ypred
}
loss = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits(logits=ypred,
labels=tf.one_hot(
labels, depth=10)))
train_op = None
if mode == learn.ModeKeys.TRAIN:
train_op = tf.train.AdamOptimizer(1e-4).minimize(loss)
return model_fn_lib.ModelFnOps(predictions=predictions,
loss=loss,
train_op=train_op,
mode=mode)
def gmm_cluster_model_fn(features, labels, mode, params, config=None):
"""Model function."""
assert labels is None, labels
update_params = ''
for i in ["w", "m", 'c']:
if i in params["update_params"]:
update_params += i
(all_scores,
model_predictions,
losses, training_op) = gmm_ops.gmm(parse_tensor_or_dict(features),
"random",
params["num_clusters"],
params["random_seed"],
params["covariance_type"],
update_params,
)
incr_step = state_ops.assign_add(variables.get_global_step(), 1)
loss = math_ops.reduce_sum(losses)
training_op = with_dependencies([training_op, incr_step], loss)
predictions = {
'all_scores': all_scores[0],
'assignments': model_predictions[0][0],
}
eval_metric_ops = {
'scores': streaming_sum(loss),
}
return model_fn_lib.ModelFnOps(mode=mode, predictions=predictions,
eval_metric_ops=eval_metric_ops,
loss=loss, train_op=training_op)
def _model_fn(features, labels, mode, config):
"""Model function."""
assert labels is None, labels
(loss,
scores,
model_predictions,
training_op,
init_op,
is_initialized) = gmm_ops.gmm(self._parse_tensor_or_dict(features),
self._training_initial_clusters,
self._num_clusters, self._random_seed,
self._covariance_type,
self._params)
incr_step = state_ops.assign_add(training_util.get_global_step(), 1)
training_op = with_dependencies([training_op, incr_step], loss)
training_hooks = [_InitializeClustersHook(
init_op, is_initialized, config.is_chief)]
predictions = {
GMM.ASSIGNMENTS: model_predictions[0][0],
}
eval_metric_ops = {
GMM.SCORES: scores,
GMM.LOG_LIKELIHOOD: _streaming_sum(loss),
}
return model_fn_lib.ModelFnOps(mode=mode, predictions=predictions,
eval_metric_ops=eval_metric_ops,
loss=loss, train_op=training_op,
training_hooks=training_hooks)
def model_fn(features, targets, mode, params):
"""Model function for Estimator."""
first_hidden_layer = tf.contrib.layers.relu(features, 10)
second_hidden_layer = tf.contrib.layers.relu(first_hidden_layer, 10)
output_layer = tf.contrib.layers.linear(second_hidden_layer, 1)
predictions = tf.reshape(output_layer, [-1])
predictions_dict = {"ages": predictions}
loss = tf.losses.mean_squared_error(targets, predictions)
eval_metric_ops = {
"rmse":
tf.metrics.root_mean_squared_error(tf.cast(targets, tf.float64),
predictions)
}
train_op = tf.contrib.layers.optimize_loss(
loss=loss,
global_step=tf.contrib.framework.get_global_step(),
learning_rate=params["learning_rate"],
optimizer="SGD")
return model_fn_lib.ModelFnOps(mode=mode,
predictions=predictions_dict,
loss=loss,
train_op=train_op,
eval_metric_ops=eval_metric_ops)
def test_build_all_signature_defs_legacy_input_fn_not_supported(self):
"""Tests that legacy input_fn returning (features, labels) raises error.
serving_input_fn must return InputFnOps including a default input
alternative.
"""
input_features = constant_op.constant(["10"])
input_ops = ({"features": input_features}, None)
input_alternatives, _ = (
saved_model_export_utils.get_input_alternatives(input_ops))
output_1 = constant_op.constant(["1"])
output_2 = constant_op.constant(["2"])
output_3 = constant_op.constant(["3"])
provided_output_alternatives = {
"head-1": (constants.ProblemType.LINEAR_REGRESSION, {
"some_output_1": output_1
}),
"head-2": (constants.ProblemType.CLASSIFICATION, {
"some_output_2": output_2
}),
"head-3": (constants.ProblemType.UNSPECIFIED, {
"some_output_3": output_3
}),
}
model_fn_ops = model_fn.ModelFnOps(
model_fn.ModeKeys.INFER,
predictions={"some_output": constant_op.constant(["4"])},
output_alternatives=provided_output_alternatives)
output_alternatives, _ = (saved_model_export_utils.get_output_alternatives(
model_fn_ops, "head-1"))
with self.assertRaisesRegexp(
ValueError, "A default input_alternative must be provided"):
saved_model_export_utils.build_all_signature_defs(
input_alternatives, output_alternatives, "head-1")
def fnn_model_fn(features, labels, mode):
print(features)
print(labels)
# output_labels = tf.reshape(labels,[-1,1])
dense = tf.layers.dense(features,
units=nhidden,
activation=tf.nn.relu,
use_bias=True)
print(dense)
logits = tf.layers.dense(dense, units=1, use_bias=True)
print(logits)
onehot_labels = tf.one_hot(indices=tf.cast(labels, tf.int32), depth=1)
if mode != learn.ModeKeys.EVAL:
# loss = tf.losses.sigmoid_cross_entropy(output_labels,logits)
# loss = tf.losses.mean_squared_error(labels=output_labels,predictions=logits)
loss = tf.losses.softmax_cross_entropy(onehot_labels=onehot_labels,
logits=logits)
if mode == learn.ModeKeys.TRAIN:
train_op = tf.contrib.layers.optimize_loss(
loss=loss,
global_step=tf.contrib.framework.get_global_step(),
learning_rate=learning_rate,
optimizer="SGD")
predictions = {
"classes": tf.round(logits),
"probabilities": tf.nn.softmax(logits, name="softmax_tensor")
}
return model_fn.ModelFnOps(mode=mode,
predictions=predictions,
loss=loss,
train_op=train_op)
def model_fn(features, targets, mode, params):
learning_rate = 0.001
keep_prob = tf.placeholder(tf.float32) # dropout (keep probability)
# Construct model
pred = cnn_model_tfp(features, params, keep_prob)
pred_dict = {
"classes": tf.argmax(input=pred, axis=1),
"probabilities": tf.nn.softmax(pred, name="softmax_tensor")
}
loss = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits(logits=pred, labels=targets))
correct_pred = tf.equal(tf.argmax(pred, 1), tf.argmax(targets, 1))
eval_metric_ops = {
"accuracy": tf.reduce_mean(tf.cast(correct_pred, tf.float32)),
"model_loss": loss
}
train_op = tf.contrib.layers.optimize_loss(
loss=loss,
global_step=tf.contrib.framework.get_global_step(),
learning_rate=learning_rate,
optimizer="Adam")
print("Got model fn fxn. params:", params)
return model_fn_lib.ModelFnOps(mode=mode,
loss=loss,
train_op=train_op,
predictions=pred_dict,
eval_metric_ops=eval_metric_ops)
def _cnn_model_fn(features, labels, mode):
x = features['data']
layer1 = tf.layers.dense(x, 10)
layer2 = tf.layers.dense(layer1, 10)
layer3 = tf.layers.dense(layer2, 10)
logits = tf.layers.dense(layer3, 1)
loss = None
train_op = None
# Calculate Loss (for both TRAIN and EVAL modes)
if mode != learn.ModeKeys.INFER:
loss = tf.losses.sigmoid_cross_entropy(labels, logits)
# Configure the Training Op (for TRAIN mode)
if mode == learn.ModeKeys.TRAIN:
train_op = tf.contrib.layers.optimize_loss(
loss=loss,
global_step=tf.contrib.framework.get_global_step(),
learning_rate=0.001,
optimizer="Adam")
# Generate Predictions
predictions = {
"classes": tf.argmax(input=logits, axis=1),
"probabilities": tf.nn.sigmoid(logits)
}
# Return a ModelFnOps object
return model_fn_lib.ModelFnOps(mode=mode,
loss=loss,
train_op=train_op,
predictions=predictions)
def model_fn(features, targets, mode, params):
logits = Dense(10, input_dim=784)(features["x"])
loss = tf.losses.softmax_cross_entropy(
onehot_labels=targets, logits=logits)
train_op = tf.contrib.layers.optimize_loss(
loss=loss,
global_step=tf.contrib.framework.get_global_step(),
learning_rate=params["learning_rate"],
optimizer="SGD")
predictions = {
"classes": tf.argmax(input=logits, axis=1),
"probabilities": tf.nn.softmax(logits)
}
eval_metric_ops = {
"accuracy": tf.metrics.accuracy(
tf.argmax(input=logits, axis=1),
tf.argmax(input=targets, axis=1))
}
return model_fn_lib.ModelFnOps(
mode=mode,
predictions=predictions,
loss=loss,
train_op=train_op,
eval_metric_ops=eval_metric_ops)
def head_ops(self,
features,
labels,
mode,
train_op_fn,
logits=None,
logits_input=None):
"""See `_Head`."""
_check_mode_valid(mode)
_check_logits_input_not_supported(logits, logits_input)
predictions = self._predictions(logits)
if (mode == model_fn.ModeKeys.INFER) or (labels is None):
loss = None
train_op = None
eval_metric_ops = None
else:
loss = self._training_loss(features, labels, logits)
train_op = (None if train_op_fn is None
or mode == model_fn.ModeKeys.EVAL else self._train_op(
features, labels, train_op_fn, logits))
eval_metric_ops = self._eval_metric_ops(features, labels, logits)
signature_fn = self._signature_fn()
return model_fn.ModelFnOps(mode=mode,
predictions=predictions,
loss=loss,
train_op=train_op,
eval_metric_ops=eval_metric_ops,
signature_fn=signature_fn)
def model(features, labels, mode):
input_layer = tf.reshape(features, [-1, 999])
input_layer = tf.one_hot(input_layer, 1000)
layer_1 = layers.fully_connected(inputs=input_layer, num_outputs=4)
output_layer = layers.fully_connected(inputs=layer_1,
num_outputs=1,
activation_fn=None)
cross_entropy = None
optimizer = None
if mode != learn.ModeKeys.INFER:
cross_entropy = tf.losses.sigmoid_cross_entropy(
multi_class_labels=tf.reshape(labels, [-1]),
logits=tf.reshape(output_layer, [-1]))
if mode == learn.ModeKeys.TRAIN:
optimizer = tf.contrib.layers.optimize_loss(
loss=cross_entropy,
global_step=tf.contrib.framework.get_global_step(),
learning_rate=0.001,
optimizer="SGD")
activated = tf.sigmoid(output_layer, name='sigmoid_tensor')
predictions = {"classes": tf.round(activated), "probabilities": activated}
return model_fn_lib.ModelFnOps(mode=mode,
predictions=predictions,
loss=cross_entropy,
train_op=optimizer)
def _lstm_model(features, targets):
cell = tf.contrib.rnn.DropoutWrapper(
tf.contrib.rnn.BasicLSTMCell(5, state_is_tuple=True), 0.9)
stacked_lstm = tf.contrib.rnn.MultiRNNCell([cell for _ in range(1)],
state_is_tuple=True)
features = tf.unstack(features, axis=1, num=10)
output, layers = tf.contrib.rnn.static_rnn(stacked_lstm,
features,
dtype=dtypes.float32)
output = dnn_layers(output[-1], [10, 10])
prediction, loss = tflearn.models.linear_regression(output, targets)
train_op = tf.contrib.layers.optimize_loss(
loss,
tf.contrib.framework.get_global_step(),
optimizer='Adagrad',
learning_rate=params["learning_rate"])
eval_metric_ops = {
"rmse":
tf.metrics.root_mean_squared_error(tf.cast(targets, tf.float32),
prediction)
}
predictions_dict = {"classes": prediction}
return model_fn_lib.ModelFnOps(mode=mode,
predictions=predictions_dict,
loss=loss,
train_op=train_op,
eval_metric_ops=eval_metric_ops)
def test_build_all_signature_defs(self):
input_features = constant_op.constant(["10"])
input_example = constant_op.constant(["11"])
input_ops = input_fn_utils.InputFnOps({"features": input_features},
None,
{"default input": input_example})
input_alternatives, _ = (
saved_model_export_utils.get_input_alternatives(input_ops))
output_1 = constant_op.constant(["1"])
output_2 = constant_op.constant(["2"])
output_3 = constant_op.constant(["3"])
provided_output_alternatives = {
"head-1": (constants.ProblemType.LINEAR_REGRESSION, {
"some_output_1": output_1
}),
"head-2": (constants.ProblemType.CLASSIFICATION, {
"some_output_2": output_2
}),
"head-3": (constants.ProblemType.UNSPECIFIED, {
"some_output_3": output_3
}),
}
model_fn_ops = model_fn.ModelFnOps(
model_fn.ModeKeys.INFER,
predictions={"some_output": constant_op.constant(["4"])},
output_alternatives=provided_output_alternatives)
output_alternatives, _ = (
saved_model_export_utils.get_output_alternatives(
model_fn_ops, "head-1"))
signature_defs = saved_model_export_utils.build_all_signature_defs(
input_alternatives, output_alternatives, "head-1")
expected_signature_defs = {
"serving_default":
signature_def_utils.regression_signature_def(
input_example, output_1),
"default_input_alternative:head-1":
signature_def_utils.regression_signature_def(
input_example, output_1),
"default_input_alternative:head-2":
signature_def_utils.classification_signature_def(
input_example, output_2, None),
"default_input_alternative:head-3":
signature_def_utils.predict_signature_def({"input": input_example},
{"output": output_3}),
# "features_input_alternative:head-1":
# signature_def_utils.regression_signature_def(input_features,
# output_1),
# "features_input_alternative:head-2":
# signature_def_utils.classification_signature_def(input_features,
# output_2, None),
# "features_input_alternative:head-3":
# signature_def_utils.predict_signature_def({
# "input": input_features
# }, {"output": output_3}),
}
self.assertDictEqual(expected_signature_defs, signature_defs)
def _model_fn_scaffold(features, labels, mode):
_, _ = features, labels
return model_fn.ModelFnOps(
mode=mode,
predictions=tf.constant(0.),
loss=tf.constant(0.),
train_op=tf.constant(0.),
training_scaffold=tf.train.Scaffold(init_fn=_init_fn))
def _model_fn_scaffold(features, labels, mode):
_, _ = features, labels
return model_fn.ModelFnOps(
mode=mode,
predictions=constant_op.constant(0.),
loss=constant_op.constant(0.),
train_op=constant_op.constant(0.),
scaffold=monitored_session.Scaffold(init_fn=_init_fn))
def cnn_model_fn(features, labels, mode):
input_layer = tf.reshape(features, [-1, 28, 28, 1])
conv1 = tf.layers.conv2d(inputs=input_layer,
filters=20,
kernel_size=[5, 5],
padding='valid',
activation=tf.nn.relu)
print conv1.get_shape()
pool1 = tf.layers.max_pooling2d(inputs=conv1, pool_size=[2, 2], strides=2)
print pool1.get_shape()
conv2 = tf.layers.conv2d(inputs=pool1,
filters=40,
kernel_size=[5, 5],
padding='valid',
activation=tf.nn.relu)
print conv2.get_shape()
pool2 = tf.layers.max_pooling2d(inputs=conv2, pool_size=[2, 2], strides=2)
print pool2.get_shape()
# Dense Layer
pool2_flat = tf.reshape(pool2, [-1, 4 * 4 * 40])
dense = tf.layers.dense(inputs=pool2_flat,
units=1024,
activation=tf.nn.relu)
dropout = tf.layers.dropout(inputs=dense,
rate=0.5,
training=mode == learn.ModeKeys.TRAIN)
# Logits Layer
logits = tf.layers.dense(inputs=dropout, units=10)
loss = None
train_op = None
if mode != learn.ModeKeys.INFER:
onehot_labels = tf.one_hot(indices=tf.cast(labels, tf.int32), depth=10)
loss = tf.losses.softmax_cross_entropy(onehot_labels=onehot_labels,
logits=logits)
if mode == learn.ModeKeys.TRAIN:
train_op = tf.contrib.layers.optimize_loss(
loss=loss,
global_step=tf.contrib.framework.get_global_step(),
learning_rate=0.005,
optimizer='SGD')
# Generate Predictions
predictions = {
"classes": tf.argmax(input=logits, axis=1),
"probabilities": tf.nn.softmax(logits, name="softmax_tensor")
}
return model_fn_lib.ModelFnOps(mode=mode,
predictions=predictions,
loss=loss,
train_op=train_op)
def lenet_model_fn(features, labels, mode):
# input layer
input_layer = tf.reshape(features,
[-1, IMAGE_SIZE, IMAGE_SIZE, NUM_CHANNELS])
conv1 = tf.layers.conv2d(inputs=input_layer,
filters=32,
kernel_size=[5, 5],
padding='same',
activation=tf.nn.relu)
pool1 = tf.layers.max_pooling2d(inputs=conv1, pool_size=[2, 2], strides=2)
conv2 = tf.layers.conv2d(inputs=pool1,
filters=64,
kernel_size=[5, 5],
padding='same',
activation=tf.nn.relu)
pool2 = tf.layers.max_pooling2d(inputs=conv2, pool_size=[2, 2], strides=2)
pool2_flat = tf.reshape(pool2, [-1, 7 * 7 * 64])
dense = tf.layers.dense(inputs=pool2_flat,
units=1024,
activation=tf.nn.relu)
dropout = tf.layers.dropout(inputs=dense,
rate=0.4,
training=(mode == learn.ModeKeys.TRAIN))
logits = tf.layers.dense(inputs=dropout, units=10)
loss = None
train_op = None
if mode != learn.ModeKeys.INFER:
onehot_labels = tf.one_hot(indices=tf.cast(labels, tf.int32), depth=10)
loss = tf.losses.softmax_cross_entropy(onehot_labels=onehot_labels,
logits=logits)
if mode == learn.ModeKeys.TRAIN:
train_op = tf.contrib.layers.optimize_loss(
loss=loss,
global_step=tf.contrib.framework.get_global_step(),
learning_rate=0.001,
optimizer='SGD')
predictions = {
"classes": tf.argmax(input=logits, axis=1),
"probabilities": tf.nn.softmax(logits, name="softmax_tensor")
}
return model_fn_lib.ModelFnOps(mode=mode,
predictions=predictions,
loss=loss,
train_op=train_op)
def cnn_model_fn(features, labels, mode, params):
"""Model function for CNN."""
# Input Layer
input_layer = tf.reshape(features, [-1, IMG_SIZE, IMG_SIZE, COL_CHANNELS])
# Convolutional Layer #1
conv1 = tf.layers.conv2d(inputs=input_layer,
filters=32,
kernel_size=[5, 5],
padding="same",
activation=tf.nn.relu,
strides=2)
# Pooling Layer #1
pool1 = tf.layers.max_pooling2d(inputs=conv1, pool_size=[2, 2], strides=2)
# Convolutional Layer #2 and Pooling Layer #2
conv2 = tf.layers.conv2d(inputs=pool1,
filters=64,
kernel_size=[5, 5],
padding="same",
activation=tf.nn.relu,
strides=2)
pool2 = tf.layers.max_pooling2d(inputs=conv2, pool_size=[2, 2], strides=2)
# Dense Layer
height = int(pool2.get_shape()[1])
width = int(pool2.get_shape()[2])
pool2_flat = tf.reshape(pool2, [-1, width * height * 64])
dense = tf.layers.dense(inputs=pool2_flat, units=1024, activation=tf.nn.relu)
dropout = tf.layers.dropout(inputs=dense, rate=params["dropout"], training=mode == learn.ModeKeys.TRAIN)
# Logits Layer
logits = tf.layers.dense(inputs=dropout, units=10)
loss = None
train_op = None
# Calculate Loss (for both TRAIN and EVAL modes)
if mode != learn.ModeKeys.INFER:
onehot_labels = tf.one_hot(indices=tf.cast(labels, tf.int32), depth=10)
loss = tf.losses.softmax_cross_entropy(onehot_labels=onehot_labels, logits=logits)
# Configure the Training Op (for TRAIN mode)
if mode == learn.ModeKeys.TRAIN:
train_op = tf.contrib.layers.optimize_loss(
loss=loss,
global_step=tf.contrib.framework.get_global_step(),
learning_rate=params["learnrate"],
optimizer=params["optimizer"])
# Generate Predictions
predictions = {"classes": tf.argmax(input=logits, axis=1),
"probabilities": tf.nn.softmax(logits, name="softmax_tensor")}
# Return a ModelFnOps object
return model_fn_lib.ModelFnOps(mode=mode, predictions=predictions, loss=loss, train_op=train_op)
def tof_net_func(x, y, mode):
# it reverse engineer the kinect
x_shape = [-1, 384, 512, 9]
y_shape = [-1, 384, 512, 18]
l = 2
lr = 1e-5
# convert to the default data type
msks = tf.tile(x[:, :, :, 9::], [1, 1, 1, 2])
msks = tf.cast(msks, dtype)
x = tf.cast(x[:, :, :, 0:9], dtype)
v = dnnOpticalFlow(x)
# v = v * msks
#
loss = None
train_op = None
# compute loss (for TRAIN and EVAL modes)
if mode != learn.ModeKeys.INFER:
y = tf.cast(y, dtype)
v_true = y
loss = (tf.reduce_mean(\
tf.abs(\
(v-v_true)\
)**l)
)**(1/l)
# loss1 = tf.reduce_sum((inten0 * inten1))/tf.reduce_sum(tf.sqrt(inten0**2*inten1**2))
# loss2 = tf.reduce_sum((inten2 * inten1))/tf.reduce_sum(tf.sqrt(inten2**2*inten1**2))
loss = tf.identity(loss, name="loss")
# configure the training op (for TRAIN mode)
if mode == learn.ModeKeys.TRAIN:
train_op = tf.contrib.layers.optimize_loss(\
loss=loss,
global_step=tf.contrib.framework.get_global_step(),
learning_rate=lr,
optimizer="Adam"
)
# generate predictions
predictions = {
"v": v,
}
# output intermediate things
# ms = tf.identity(ms, name='ms')
# x_tilde = tf.identity(x_tilde, name='x_tilde')
tensors = []
for tensor in tensors:
predictions[tensor.name] = tensor
# return a ModelFnOps object
return model_fn_lib.ModelFnOps(
mode=mode,
predictions=predictions,
loss=loss,
train_op=train_op,
)
def cnn_model_fn(features, labels, mode, params):
nn = cnn(features, labels, mode, params['rank'], params['is_riemannian'])
# Return a ModelFnOps object
model = model_fn_lib.ModelFnOps(mode=mode,
predictions=nn['predictions'],
loss=nn['loss'],
train_op=nn['train_op'])
return model
def model_fn(features, targets, mode, params):
"""Model function for Estimator."""
# 1. Configure the model via TensorFlow operations
# First, build all the model, a good idea is using Keras or tf.layers
# since these are high-level API's
conv1 = Conv2D(32, (5, 5), activation='relu',
input_shape=(28, 28, 1))(features)
pool1 = MaxPooling2D(pool_size=(2, 2))(conv1)
conv2 = Conv2D(64, (5, 5), activation='relu')(pool1)
pool2 = MaxPooling2D(pool_size=(2, 2))(conv2)
flat = Flatten()(pool2)
dense = Dense(1024, activation='relu')(flat)
preds = Dense(10, activation='softmax')(dense)
# 2. Define the loss function for training/evaluation
loss = None
train_op = None
# Calculate Loss (for both TRAIN and EVAL modes)
if mode != learn.ModeKeys.INFER:
loss = tf.losses.softmax_cross_entropy(onehot_labels=targets,
logits=preds)
# 3. Define the training operation/optimizer
# Configure the Training Op (for TRAIN mode)
if mode == learn.ModeKeys.TRAIN:
train_op = tf.contrib.layers.optimize_loss(
loss=loss,
global_step=tf.contrib.framework.get_global_step(),
learning_rate=params["learning_rate"],
optimizer="Adam",
)
# 4. Generate predictions
predictions_dict = {
"classes": tf.argmax(input=preds, axis=1),
"probabilities": tf.nn.softmax(preds, name="softmax_tensor")
}
# 5. Define how you want to evaluate the model
metrics = {
"accuracy":
tf.metrics.accuracy(tf.argmax(input=preds, axis=1),
tf.argmax(input=targets, axis=1))
}
# 6. Return predictions/loss/train_op/eval_metric_ops in ModelFnOps object
return model_fn_lib.ModelFnOps(mode=mode,
predictions=predictions_dict,
loss=loss,
train_op=train_op,
eval_metric_ops=metrics)
def linear_model_fn_with_model_fn_ops(features, labels, mode):
"""Same as linear_model_fn, but returns `ModelFnOps`."""
assert mode in (model_fn.ModeKeys.TRAIN, model_fn.ModeKeys.EVAL,
model_fn.ModeKeys.INFER)
prediction, loss = (models.linear_regression_zero_init(features, labels))
train_op = optimizers.optimize_loss(
loss, variables.get_global_step(), optimizer='Adagrad', learning_rate=0.1)
return model_fn.ModelFnOps(
mode=mode, predictions=prediction, loss=loss, train_op=train_op)
def _model_fn(features, labels, mode):
"""Function that returns predictions, training loss, and training op."""
weights = None
if weights_name and weights_name in features:
weights = features.pop(weights_name)
graph_builder = graph_builder_class(params,
device_assigner=device_assigner)
inference = {}
if (mode == model_fn_lib.ModeKeys.EVAL
or mode == model_fn_lib.ModeKeys.INFER):
inference[eval_metrics.INFERENCE_PROB_NAME] = (
graph_builder.inference_graph(features))
if not params.regression:
inference[eval_metrics.INFERENCE_PRED_NAME] = math_ops.argmax(
inference[eval_metrics.INFERENCE_PROB_NAME], 1)
# labels might be None if we're doing prediction (which brings up the
# question of why we force everything to adhere to a single model_fn).
loss_deps = []
training_graph = None
if labels is not None and mode == model_fn_lib.ModeKeys.TRAIN:
training_graph = control_flow_ops.group(
graph_builder.training_graph(features,
labels,
input_weights=weights,
num_trainers=num_trainers,
trainer_id=trainer_id),
state_ops.assign_add(contrib_framework.get_global_step(), 1))
loss_deps.append(training_graph)
training_loss = None
if (mode == model_fn_lib.ModeKeys.EVAL
or mode == model_fn_lib.ModeKeys.TRAIN):
with ops.control_dependencies(loss_deps):
training_loss = graph_builder.training_loss(features,
labels,
name=LOSS_NAME)
if report_feature_importances and mode == model_fn_lib.ModeKeys.EVAL:
training_loss = logging_ops.Print(
training_loss, [graph_builder.feature_importances()],
summarize=1000)
# Put weights back in
if weights is not None:
features[weights_name] = weights
training_hooks = []
if early_stopping_rounds:
training_hooks.append(TensorForestLossHook(early_stopping_rounds))
return model_fn_lib.ModelFnOps(mode=mode,
predictions=inference,
loss=training_loss,
train_op=training_graph,
training_hooks=training_hooks)
def estimator_spec_to_model_fn_ops(estimator_spec):
alternatives = _export_outputs_to_output_alternatives(
estimator_spec.export_outputs)
return model_fn.ModelFnOps(mode=_core_mode_to_contrib_mode(
estimator_spec.mode),
predictions=estimator_spec.predictions,
loss=estimator_spec.loss,
train_op=estimator_spec.train_op,
eval_metric_ops=estimator_spec.eval_metric_ops,
output_alternatives=alternatives)
def cnn_model_fn(features, labels, mode):
#features = features.astype(dtype=np.float32)
# Input Layer
input_layer = tf.reshape(features, [batch_size, 121, 195, 1])
input_layer = tf.to_float(input_layer)
# Modèle simplifié avec un seul CNN
# Conv Layer
conv1 = tf.layers.conv2d(
inputs=input_layer,
filters=16,
strides=(10,10),
kernel_size=[10, 10],
padding="same",
activation=tf.nn.relu)
# Dense Layer
conv1_flat = tf.reshape(conv1, [batch_size, 4160])
dense = tf.layers.dense(inputs=conv1_flat, units=1024, activation=tf.nn.relu)
dropout = tf.layers.dropout(
inputs=dense, rate=0.4, training=mode == learn.ModeKeys.TRAIN)
# Logits Layer
logits = tf.layers.dense(inputs=dropout, units=5)
loss = None
train_op = None
# Calculate Loss (for both TRAIN and EVAL modes)
if mode != learn.ModeKeys.INFER:
onehot_labels = tf.one_hot(indices=tf.cast(labels, tf.int32), depth=5)
loss = tf.losses.softmax_cross_entropy(
onehot_labels=onehot_labels, logits=logits)
# Configure the Training Op (for TRAIN mode)
if mode == learn.ModeKeys.TRAIN:
train_op = tf.contrib.layers.optimize_loss(
loss=loss,
global_step=tf.contrib.framework.get_global_step(),
learning_rate=0.001,
optimizer="SGD")
# Generate Predictions
predictions = {
"classes": tf.argmax(
input=logits, axis=1),
"probabilities": tf.nn.softmax(
logits, name="softmax_tensor")
}
# Return a ModelFnOps object
return model_fn_lib.ModelFnOps(
mode=mode, predictions=predictions, loss=loss, train_op=train_op)
def cnn_model_fn(features, labels, mode):
"""Model function for CNN."""
input_one_hot = tf.one_hot(indices=tf.cast(features, tf.int32), depth=3000)
input_one_hot = tf.reshape(input_one_hot, [-1, 128, 1, 3000])
conv1a = tf.layers.conv2d(inputs=input_one_hot,
filters=1000,
kernel_size=[3, 1],
padding="same",
activation=tf.nn.relu)
conv1b = tf.layers.conv2d(inputs=input_one_hot,
filters=1000,
kernel_size=[2, 1],
padding="same",
activation=tf.nn.relu)
conv1 = tf.concat([conv1a, conv1b], 1)
pool1 = tf.layers.max_pooling2d(inputs=conv1, pool_size=[2, 1], strides=2)
pool1_flat = tf.reshape(pool1, [-1, 128 * 1 * 1000])
dense = tf.layers.dense(inputs=pool1_flat,
units=1024,
activation=tf.nn.relu)
dropout = tf.layers.dropout(inputs=dense,
rate=0.4,
training=mode == learn.ModeKeys.TRAIN)
ratings = tf.layers.dense(inputs=dropout, units=1, activation=tf.nn.relu6)
loss = None
train_op = None
# Calculate Loss (for both TRAIN and EVAL modes)
if mode != learn.ModeKeys.INFER:
labels = tf.reshape(labels, [-1, 1])
loss = tf.losses.mean_squared_error(labels=labels, predictions=ratings)
def f(lr, gs):
return tf.train.exponential_decay(lr, gs, 100, 0.85)
# Configure the Training Op (for TRAIN mode)
if mode == learn.ModeKeys.TRAIN:
train_op = tf.contrib.layers.optimize_loss(
loss=loss,
global_step=tf.contrib.framework.get_global_step(),
learning_rate=0.001,
learning_rate_decay_fn=f,
optimizer="SGD")
# Generate Predictions
predictions = {"ratings": ratings}
# Return a ModelFnOps object
return model_fn_lib.ModelFnOps(mode=mode,
predictions=predictions,
loss=loss,
train_op=train_op)