def model(self, X, w_h1, w_h2, w_o):
h1 = tf.nn.sigmoid(melt.matmul(
X,
w_h1)) # this is a basic mlp, think 2 stacked logistic regressions
h2 = tf.nn.sigmoid(melt.matmul(
h1,
w_h2)) # this is a basic mlp, think 2 stacked logistic regressions
return tf.matmul(
h2, w_o
) # note that we dont take the softmax at the end because our cost fn does that for us
def model(self, X, w_h, b_h, w_o, b_o, gpu):
with tf.device('/gpu:%d' % gpu):
h = tf.nn.sigmoid(
melt.matmul(X, w_h) + b_h
) # this is a basic mlp, think 2 stacked logistic regressions
return tf.matmul(
h, w_o
) + b_o # note that we dont take the softmax at the end because our cost fn does that for us
def fully_connected(x, output_size, activation=tf.nn.relu, scope=None):
#@TODO -1 or last dim ? NotImplementedError("Negative indices are currently unsupported")
#input_dim = tf.shape(x)[-1]
#@TODO how is slim.fully_connected get inputdim and use..
#below will now work int() argument must be a string or a number, not 'Tensor' [input_dim, output_size])
#input_dim = tf.shape(x)[1]
#check contrib\layers\python\layers\layers.py
scope = 'fc' if scope is None else scope
with tf.variable_scope(scope):
input_dim = utils.last_dimension(x.get_shape(), min_rank=2)
if isinstance(x, tf.Tensor):
w_h = melt.get_weights('w_h', [input_dim, output_size])
else:
with tf.device('/cpu:0'):
w_h = melt.get_weights('w_h', [input_dim, output_size])
b_h = melt.get_bias('b_h', [output_size])
return activation(melt.matmul(x, w_h) + b_h)
def model(self, X, w_h, b_h, w_h2, b_h2, w_o, b_o):
h = self.activation(melt.matmul(X, w_h) + b_h) # this is a basic mlp, think 2 stacked logistic regressions
h2 = self.activation(tf.matmul(h, w_h2) + b_h2)
return tf.matmul(h2, w_o) + b_o # note that we dont take the softmax at the end because our cost fn does that for us
def model(self, X, w, b):
return melt.matmul(X,w) + b
def model(self, X, w_h, w_o): h = tf.nn.sigmoid(melt.matmul(X, w_h)) # this is a basic mlp, think 2 stacked logistic regressions return tf.matmul(h, w_o) # note that we dont take the softmax at the end because our cost fn does that for us
def model(self, X, w): return melt.matmul(X,w)
def fully_connected(inputs,
num_outputs,
input_dim=None,
activation_fn=nn.relu,
normalizer_fn=None,
normalizer_params=None,
weights_initializer=initializers.xavier_initializer(),
weights_regularizer=None,
biases_initializer=init_ops.zeros_initializer(),
biases_regularizer=None,
reuse=None,
variables_collections=None,
outputs_collections=None,
trainable=True,
scope=None):
use_bias = biases_initializer is not None
#--------TODO: use commented code as layers.fully_connected then, for app code you must manully pass scope like 'mlp' otherwise evaluate will fail try to use Mlp_1 not resue=True b
#--------by tf.variable_scope.reuse_variables() see http://stackoverflow.com/questions/40536665/tensorflow-varscope-reuse-variables
#with variable_scope.variable_scope(
# scope, 'Mlp', [inputs],
# reuse=reuse) as vs:
scope = 'fully_connected' if scope is None else scope
with tf.variable_scope(scope):
is_dense_input = True if isinstance(inputs, tf.Tensor) else False
dtype = inputs.dtype.base_dtype if is_dense_input else inputs[
1].values.dtype.base_dtype
#sparse input must tell input_dim
assert is_dense_input or input_dim is not None
if is_dense_input:
shape = inputs.get_shape().as_list()
input_dim = shape[-1].value
assert len(
shape
) == 2, "now only consider X shape dim as 2, TODO: make > 2 ok like layers.fully_connected"
#-----------deal first hidden
if is_dense_input:
w_h = tf.get_variable('weight_hidden',
shape=[input_dim, num_outputs],
initializer=weights_initializer,
regularizer=weights_regularizer,
dtype=dtype,
trainable=trainable)
else:
with tf.device('/cpu:0'):
w_h = tf.get_variable('weight_hidden',
shape=[input_dim, num_outputs],
initializer=weights_initializer,
regularizer=weights_regularizer,
dtype=dtype,
trainable=trainable)
if use_bias:
b_h = tf.get_variable('bias_hidden',
shape=[
num_outputs,
],
initializer=biases_initializer,
regularizer=biases_regularizer,
dtype=dtype,
trainable=trainable)
outputs = melt.matmul(inputs, w_h)
if use_bias:
outputs = nn.bias_add(outputs, b_h)
if activation_fn is not None:
outputs = activation_fn(outputs) # pylint: disable=not-callable
return outputs
def model(X, w):
return melt.matmul(X, w)