def __init__(self, sess, ob_space, ac_space, nbatch, nsteps, reuse=False): #pylint: disable=W0613
nh, nw, nc = ob_space.shape
ob_shape = (nbatch, nh, nw, nc)
nact = ac_space.n
X = tf.placeholder(tf.uint8, ob_shape) #obs
with tf.variable_scope("model", reuse=reuse):
h = nature_cnn(X)
pi = fc(h, 'pi', nact, init_scale=0.01)
vf = fc(h, 'v', 1)[:, 0]
self.pdtype = make_pdtype(ac_space)
self.pd = self.pdtype.pdfromflat(pi)
a0 = self.pd.sample()
neglogp0 = self.pd.neglogp(a0)
self.initial_state = None
def step(ob, *_args, **_kwargs):
a, v, neglogp = sess.run([a0, vf, neglogp0], {X: ob})
return a, v, self.initial_state, neglogp
def value(ob, *_args, **_kwargs):
return sess.run(vf, {X: ob})
self.X = X
self.pi = pi
self.vf = vf
self.step = step
self.value = value
def __init__(self,
sess,
ob_space,
ac_space,
nbatch,
nsteps,
nlstm=256,
reuse=False):
nenv = nbatch // nsteps
nh, nw, nc = ob_space.shape
ob_shape = (nbatch, nh, nw, nc)
nact = ac_space.n
X = tf.placeholder(tf.uint8, ob_shape) #obs
M = tf.placeholder(tf.float32, [nbatch]) #mask (done t-1)
S = tf.placeholder(tf.float32, [nenv, nlstm * 2]) #states
with tf.variable_scope("model", reuse=reuse):
h = nature_cnn(X)
xs = batch_to_seq(h, nenv, nsteps)
ms = batch_to_seq(M, nenv, nsteps)
h5, snew = lstm(xs, ms, S, 'lstm1', nh=nlstm)
h5 = seq_to_batch(h5)
pi = fc(h5, 'pi', nact)
vf = fc(h5, 'v', 1)
self.pdtype = make_pdtype(ac_space)
self.pd = self.pdtype.pdfromflat(pi)
v0 = vf[:, 0]
a0 = self.pd.sample()
neglogp0 = self.pd.neglogp(a0)
self.initial_state = np.zeros((nenv, nlstm * 2), dtype=np.float32)
def step(ob, state, mask):
return sess.run([a0, v0, snew, neglogp0], {
X: ob,
S: state,
M: mask
})
def value(ob, state, mask):
return sess.run(v0, {X: ob, S: state, M: mask})
self.X = X
self.M = M
self.S = S
self.pi = pi
self.vf = vf
self.step = step
self.value = value
def __init__(self, sess, ob_space, ac_space, nbatch, nsteps, reuse=False): #pylint: disable=W0613
nenv = nbatch // nsteps
ob_shape = (nbatch, ) + ob_space.shape
nact = ac_space.n
X = tf.placeholder(tf.float32, ob_shape, name='Ob') #obs
with tf.variable_scope('intrinsic', reuse=reuse):
h3 = nature_cnn(X)
r_in0 = tf.tanh(fc(h3, 'r_in', nact))
v_ex0 = fc(h3, 'v_ex', 1)[:, 0]
with tf.variable_scope('policy', reuse=reuse):
h3 = nature_cnn(X)
pi = fc(h3, 'pi', nact, init_scale=0.01)
v_mix0 = fc(h3, 'v_mix', 1)[:, 0]
self.pdtype = make_pdtype(ac_space)
self.pd = self.pdtype.pdfromflat(pi)
a0 = self.pd.sample()
neglogp0 = self.pd.neglogp(a0)
self.init_policy_state = None
def step(ob, *_args, **_kwargs):
a, v_ex, v_mix, neglogp = sess.run([a0, v_ex0, v_mix0, neglogp0],
{X: ob})
return a, v_ex, v_mix, self.init_policy_state, neglogp
def value(ob, *_args, **_kwargs):
v_ex, v_mix = sess.run([v_ex0, v_mix0], {X: ob})
return v_ex, v_mix
def intrinsic_reward(ob, ac, *_args, **_kwargs):
r_in = sess.run(r_in0, {X: ob})
return r_in[np.arange(nbatch), ac]
self.X = X
self.r_in = r_in0
self.v_ex = v_ex0
self.pi = pi
self.v_mix = v_mix0
self.step = step
self.value = value
self.intrinsic_reward = intrinsic_reward
self.policy_params = tf.trainable_variables("policy")
self.intrinsic_params = tf.trainable_variables("intrinsic")
self.policy_new_fn = CnnPolicyNew
def nature_cnn(unscaled_images):
"""
CNN from Nature paper.
"""
scaled_images = tf.cast(unscaled_images, tf.float32) / 255.
activ = tf.nn.relu
h = activ(
conv(scaled_images, 'c1', nf=32, rf=8, stride=4,
init_scale=np.sqrt(2)))
h2 = activ(conv(h, 'c2', nf=64, rf=4, stride=2, init_scale=np.sqrt(2)))
h3 = activ(conv(h2, 'c3', nf=64, rf=3, stride=1, init_scale=np.sqrt(2)))
h3 = conv_to_fc(h3)
return activ(fc(h3, 'fc1', nh=512, init_scale=np.sqrt(2)))
def __init__(self, sess, ob_space, ac_space, nbatch, nsteps, reuse=False): #pylint: disable=W0613
ob_shape = (nbatch, ) + ob_space.shape
actdim = ac_space.shape[0]
X = tf.placeholder(tf.float32, ob_shape, name='Ob') #obs
with tf.variable_scope("model", reuse=reuse):
activ = tf.tanh
h1 = activ(fc(X, 'pi_fc1', nh=64, init_scale=np.sqrt(2)))
h2 = activ(fc(h1, 'pi_fc2', nh=64, init_scale=np.sqrt(2)))
pi = fc(h2, 'pi', actdim, init_scale=0.01)
h1 = activ(fc(X, 'vf_fc1', nh=64, init_scale=np.sqrt(2)))
h2 = activ(fc(h1, 'vf_fc2', nh=64, init_scale=np.sqrt(2)))
vf = fc(h2, 'vf', 1)[:, 0]
logstd = tf.get_variable(name="logstd",
shape=[1, actdim],
initializer=tf.zeros_initializer())
pdparam = tf.concat([pi, pi * 0.0 + logstd], axis=1)
self.pdtype = make_pdtype(ac_space)
self.pd = self.pdtype.pdfromflat(pdparam)
a0 = self.pd.sample()
neglogp0 = self.pd.neglogp(a0)
self.initial_state = None
def step(ob, *_args, **_kwargs):
a, v, neglogp = sess.run([a0, vf, neglogp0], {X: ob})
return a, v, self.initial_state, neglogp
def value(ob, *_args, **_kwargs):
return sess.run(vf, {X: ob})
self.X = X
self.pi = pi
self.vf = vf
self.step = step
self.value = value