def nn_mnist_model(graph):
"""Creates model for NN mnist.
Returns:
logits
"""
with graph.as_default():
is_train = tf.placeholder(tf.bool, shape=(), name='is_train')
n_inputs = 28 * 28
n_hidden1 = 300
n_hidden2 = 100
n_outputs = 10
with graph.as_default():
with tf.name_scope('Inputs'):
with tf.name_scope('X'):
X = tf.placeholder(DTYPE, shape=(None, n_inputs), name='X') # pylint: disable=invalid-name
with tf.name_scope('y'):
y = tf.placeholder(tf.int64, shape=(None), name='y') # pylint: disable=invalid-name
hidden1 = nn_layer(X, n_hidden1, name='hidden1', activation=tf.nn.relu) # pylint: disable=no-member
hidden2 = nn_layer(hidden1,
n_hidden2,
name='hidden2',
activation=tf.nn.relu) # pylint: disable=no-member
logits = nn_layer(hidden2, n_outputs, name='logits')
return X, y, is_train, logits
def nn_mnist_model_small_with_dropout(graph):
with graph.as_default():
is_train = tf.placeholder(tf.bool, shape=(), name='is_train')
n_inputs = 28 * 28
n_hidden1 = 5
n_hidden2 = 5
n_outputs = 10
with graph.as_default():
with tf.name_scope('Inputs'):
with tf.name_scope('X'):
X = tf.placeholder(DTYPE, shape=(None, n_inputs), name='X') # pylint: disable=invalid-name
with tf.name_scope('y'):
y = tf.placeholder(tf.int64, shape=(None), name='y') # pylint: disable=invalid-name
keep_prob = tf.placeholder(DTYPE, name='keep_prob')
hidden1 = nn_layer(X, n_hidden1, name='hidden1', activation=tf.nn.relu) # pylint: disable=no-member
hidden1_dropout = tf.nn.dropout(hidden1, keep_prob)
hidden2 = nn_layer(hidden1_dropout,
n_hidden2,
name='hidden2',
activation=tf.nn.relu) # pylint: disable=no-member
hidden2_dropout = tf.nn.dropout(hidden2, keep_prob)
logits = nn_layer(hidden2, n_outputs, name='logits')
return X, y, is_train, keep_prob, logits
def nn_mnist_model_dropout(graph): # pylint: disable=too-many-locals
"""Creates model for NN mnist.
Returns:
logits
"""
n_inputs = 28 * 28
n_hidden1 = 1024
n_hidden2 = 1024
n_hidden3 = 2048
n_outputs = 10
with graph.as_default():
is_train = tf.placeholder(tf.bool, shape=(), name='is_train')
with graph.as_default():
with tf.name_scope('Inputs'):
with tf.name_scope('X'):
X = tf.placeholder(DTYPE, shape=(None, n_inputs), name='X') # pylint: disable=invalid-name
with tf.name_scope('y'):
y = tf.placeholder(tf.int64, shape=(None), name='y') # pylint: disable=invalid-name
#with tf.name_scope('NN'):
keep_prob = tf.placeholder(DTYPE, name='keep_prob')
hidden1 = nn_layer(X, n_hidden1, name='hidden1', activation=tf.nn.relu) # pylint: disable=no-member
hidden1_dropout = tf.nn.dropout(hidden1, keep_prob)
hidden2 = nn_layer(hidden1_dropout,
n_hidden2,
name='hidden2',
activation=tf.nn.relu) # pylint: disable=no-member
hidden2_dropout = tf.nn.dropout(hidden2, keep_prob)
hidden3 = nn_layer(hidden2_dropout,
n_hidden3,
name='hidden3',
activation=tf.nn.relu) # pylint: disable=no-member
hidden3_dropout = tf.nn.dropout(hidden3, keep_prob)
logits = nn_layer(hidden3_dropout, n_outputs, name='logits')
return X, y, is_train, keep_prob, logits