def __init__(self,
env_spec,
o_ph,
a_ph,
name="q_function",
hidden_sizes=[
32,
32,
],
activation=tf.nn.relu,
output_activation=None):
self._env_spec = env_spec
self._name = name
# TODO: Dynamically support disc. or cont action space
with tf.variable_scope(self._name):
x = tf.concat([o_ph, a_ph], axis=1)
x = mlp(x,
hidden_sizes + [1],
activation=activation,
output_activation=output_activation)
x = tf.squeeze(x)
self._qf = x
def test_mlp(devs, kv_type):
# guarantee the same weight init for each run
mx.random.seed(0)
logging.basicConfig(level=logging.DEBUG)
(train, val) = common.mnist(batch_size=100, input_shape=(784,))
# train
model = mx.model.FeedForward.create(
symbol=common.mlp(), ctx=devs, X=train, num_epoch=4, learning_rate=0.1, wd=0.0004, momentum=0.9, kvstore=kv_type
)
return common.accuracy(model, val)
def test_mlp(devs, kv_type):
# guarantee the same weight init for each run
mx.random.seed(0)
logging.basicConfig(level=logging.DEBUG)
(train, val) = common.mnist(batch_size=100, input_shape=(784, ))
# train
model = mx.model.FeedForward.create(symbol=common.mlp(),
ctx=devs,
X=train,
num_epoch=4,
learning_rate=0.1,
wd=0.0004,
momentum=0.9,
kvstore=kv_type)
return common.accuracy(model, val)
def __init__( self,
env_spec,
o_ph,
hidden_sizes=[32,32,],
activation = tf.nn.relu,
output_activation = None,
name='value_function',
):
self._env_spec = env_spec
self._name = name
# Note: Do we neeed to create internal referemces for hidden_sizes,
# activation, output_activation and such
# Building the VF graph
with tf.variable_scope( self._name):
x = o_ph
x = mlp( x, hidden_sizes+[1], activation = activation,
output_activation = output_activation)
x = tf.squeeze( x)
# TODO: Unlegant, upgrade
self._vf = x
#!/usr/bin/env python
import mxnet as mx
import logging
import common
mx.random.seed(0)
logging.basicConfig(level=logging.DEBUG)
kv = mx.kvstore.create('dist_async')
(train, val) = common.mnist(num_parts = kv.num_workers,
part_index = kv.rank,
batch_size = 100,
input_shape = (784,))
# train
model = mx.model.FeedForward.create(
symbol = common.mlp(),
ctx = mx.cpu(),
X = train,
num_round = 4,
learning_rate = 0.05,
wd = 0.0004,
momentum = 0.9,
kvstore = kv)
common.accuracy(model, val)
#!/usr/bin/env python
import mxnet as mx
import logging
import common
mx.random.seed(0)
logging.basicConfig(level=logging.DEBUG)
kv = mx.kvstore.create('dist_sync')
# feed each machine the whole data
(train, val) = common.mnist(batch_size=100, input_shape=(784, ))
# train
model = mx.model.FeedForward.create(symbol=common.mlp(),
ctx=mx.cpu(),
X=train,
num_epoch=4,
learning_rate=0.1,
wd=0.0004,
momentum=0.9,
kvstore=kv)
common.accuracy(model, val)
#!/usr/bin/env python
import common
import mxnet as mx
import logging
mx.random.seed(0)
logging.basicConfig(level=logging.DEBUG)
kv = mx.kvstore.create('dist_sync')
# feed each machine the whole data
(train, val) = common.mnist(batch_size = 100,
input_shape = (784,))
# train
model = mx.model.FeedForward.create(
symbol = common.mlp(),
ctx = mx.cpu(),
X = train,
num_epoch = 4,
epoch_size = 60000 / 100,
learning_rate = 0.1,
wd = 0.0004,
momentum = 0.9,
kvstore = kv)
common.accuracy(model, val)