python类get_output()的实例源码

def dist_info_sym(self, obs_var, state_info_vars):
        n_batches, n_steps = obs_var.shape[:2]
        obs_var = obs_var.reshape((n_batches, n_steps, -1))
        if self.state_include_action:
            prev_action_var = state_info_vars["prev_action"]
            all_input_var = TT.concatenate(
                [obs_var, prev_action_var],
                axis=2
            )
        else:
            all_input_var = obs_var

        if self.feature_network is None:
            return dict(
                prob=L.get_output(
                    self.prob_network.output_layer,
                    {self.l_input: all_input_var}
                )
            )
        else:
            flat_input_var = TT.reshape(all_input_var, (-1, self.input_dim))
            return dict(
                prob=L.get_output(
                    self.prob_network.output_layer,
                    {self.l_input: all_input_var, self.feature_network.input_layer: flat_input_var}
                )
            )

def dist_info_sym(self, obs_var, state_info_vars):
        n_batches, n_steps = obs_var.shape[:2]
        obs_var = obs_var.reshape((n_batches, n_steps, -1))
        if self._state_include_action:
            prev_action_var = state_info_vars["prev_action"]
            all_input_var = TT.concatenate(
                [obs_var, prev_action_var],
                axis=2
            )
        else:
            all_input_var = obs_var
        means, log_stds = L.get_output([self._mean_network.output_layer, self._l_log_std], all_input_var)
        return dict(mean=means, log_std=log_stds)

def dist_info_sym(self, obs_var, state_info_vars=None):
        return dict(
            prob=L.get_output(
                self._l_prob,
                {self._l_obs: obs_var}
            )
        )

def __init__(
            self,
            env_spec,
            hidden_sizes=(32, 32),
            hidden_nonlinearity=NL.tanh,
            num_seq_inputs=1,
            prob_network=None,
    ):
        """
        :param env_spec: A spec for the mdp.
        :param hidden_sizes: list of sizes for the fully connected hidden layers
        :param hidden_nonlinearity: nonlinearity used for each hidden layer
        :param prob_network: manually specified network for this policy, other network params
        are ignored
        :return:
        """
        Serializable.quick_init(self, locals())

        assert isinstance(env_spec.action_space, Discrete)

        if prob_network is None:
            prob_network = MLP(
                input_shape=(env_spec.observation_space.flat_dim * num_seq_inputs,),
                output_dim=env_spec.action_space.n,
                hidden_sizes=hidden_sizes,
                hidden_nonlinearity=hidden_nonlinearity,
                output_nonlinearity=NL.softmax,
            )

        self._l_prob = prob_network.output_layer
        self._l_obs = prob_network.input_layer
        self._f_prob = ext.compile_function([prob_network.input_layer.input_var], L.get_output(
            prob_network.output_layer))

        self._dist = Categorical(env_spec.action_space.n)

        super(CategoricalMLPPolicy, self).__init__(env_spec)
        LasagnePowered.__init__(self, [prob_network.output_layer])

def dist_info_sym(self, obs_var, state_info_vars=None):
        return dict(prob=L.get_output(self._l_prob, {self._l_obs: obs_var}))

def log_likelihood_sym(self, x_var, y_var):
        normalized_xs_var = (x_var - self._x_mean_var) / self._x_std_var

        normalized_means_var, normalized_log_stds_var = \
            L.get_output([self._l_mean, self._l_log_std], {
                self._mean_network.input_layer: normalized_xs_var})

        means_var = normalized_means_var * self._y_std_var + self._y_mean_var
        log_stds_var = normalized_log_stds_var + TT.log(self._y_std_var)

        return self._dist.log_likelihood_sym(y_var, dict(mean=means_var, log_std=log_stds_var))

def log_likelihood_sym(self, x_var, y_var):
        normalized_xs_var = (x_var - self._x_mean_var) / self._x_std_var

        normalized_means_var, normalized_log_stds_var = \
            L.get_output([self._l_mean, self._l_log_std], {
                self._mean_network.input_layer: normalized_xs_var})

        means_var = normalized_means_var * self._y_std_var + self._y_mean_var
        log_stds_var = normalized_log_stds_var + TT.log(self._y_std_var)

        return self._dist.log_likelihood_sym(y_var, dict(mean=means_var, log_std=log_stds_var))

def test_gru_network():
    from rllab.core.network import GRUNetwork
    import lasagne.layers as L
    from rllab.misc import ext
    import numpy as np
    network = GRUNetwork(
        input_shape=(2, 3),
        output_dim=5,
        hidden_dim=4,
    )
    f_output = ext.compile_function(
        inputs=[network.input_layer.input_var],
        outputs=L.get_output(network.output_layer)
    )
    assert f_output(np.zeros((6, 8, 2, 3))).shape == (6, 8, 5)

def __init__(self, x, y, args):
        self.params_theta = []
        self.params_lambda = []
        self.params_weight = []
        if args.dataset == 'mnist':
            input_size = (None, 1, 28, 28)
        elif args.dataset == 'cifar10':
            input_size = (None, 3, 32, 32)
        else:
            raise AssertionError
        layers = [ll.InputLayer(input_size)]
        self.penalty = theano.shared(np.array(0.))

        #conv1
        layers.append(Conv2DLayerWithReg(args, layers[-1], 20, 5))
        self.add_params_to_self(args, layers[-1])
        layers.append(ll.MaxPool2DLayer(layers[-1], pool_size=2, stride=2))
        #conv1
        layers.append(Conv2DLayerWithReg(args, layers[-1], 50, 5))
        self.add_params_to_self(args, layers[-1])
        layers.append(ll.MaxPool2DLayer(layers[-1], pool_size=2, stride=2))
        #fc1
        layers.append(DenseLayerWithReg(args, layers[-1], num_units=500))
        self.add_params_to_self(args, layers[-1])
        #softmax
        layers.append(DenseLayerWithReg(args, layers[-1], num_units=10, nonlinearity=nonlinearities.softmax))
        self.add_params_to_self(args, layers[-1])

        self.layers = layers
        self.y = ll.get_output(layers[-1], x, deterministic=False)
        self.prediction = T.argmax(self.y, axis=1)
        # self.penalty = penalty if penalty != 0. else T.constant(0.)
        print(self.params_lambda)
        # time.sleep(20)
        # cost function
        self.loss = T.mean(categorical_crossentropy(self.y, y))
        self.lossWithPenalty = T.add(self.loss, self.penalty)
        print "loss and losswithpenalty", type(self.loss), type(self.lossWithPenalty)

def predict(self,candidates):
    print(colored('Predicting {} samples...'.format(len(candidates)), 'green'))
    inputx     = [n.input_var for n in self.input_layers]
    output     = [layers.get_output(n) for n in self.nets] # Actual output
    gen_output = [theano.function([inputx[i]], output[i]) for i in range(len(self.nets))]
    vs = [gen_output[i](candidates) for i in range(len(self.nets))]
    vs = np.transpose(vs)

    # TAODEBUG: 
    print(vs[0])

    return vs[0]

  # NOTE: 
  # Sample of [save] / [load] of Lasagne CNN model
  # can be found at: 
  # https://github.com/Lasagne/Lasagne/blob/master/examples/mnist.py

  # def save(self, path):
  #   print(colored('Saving the models at {}'.format(path),'green'))
  #   i = 0
  #   for net in self.nets:
  #     print('...Saving {}'.format(path + str(i)))
  #     np.savez(path + str(i), *lasagne.layers.get_all_param_values(self.nets[i]))
  #     i += 1
  #   print('...Done')

def get_action_sym(self, obs_var):
        return L.get_output(self._output_layer, obs_var)

def dist_info_sym(self, obs_var, state_info_vars):
        n_batches, n_steps = obs_var.shape[:2]
        obs_var = obs_var.reshape((n_batches, n_steps, -1))
        if self.state_include_action:
            prev_action_var = state_info_vars["prev_action"]
            all_input_var = TT.concatenate(
                [obs_var, prev_action_var],
                axis=2
            )
        else:
            all_input_var = obs_var

        if self.feature_network is None:
            return dict(
                prob=L.get_output(
                    self.prob_network.output_layer,
                    {self.l_input: all_input_var}
                )
            )
        else:
            flat_input_var = TT.reshape(all_input_var, (-1, self.input_dim))
            return dict(
                prob=L.get_output(
                    self.prob_network.output_layer,
                    {self.l_input: all_input_var,
                        self.feature_network.input_layer: flat_input_var}
                )
            )

def dist_info_sym(self, obs_var, state_info_vars=None):
        mean_var, log_std_var = L.get_output(
            [self._l_mean, self._l_log_std], obs_var)
        if self.min_std is not None:
            log_std_var = TT.maximum(log_std_var, np.log(self.min_std))
        return dict(mean=mean_var, log_std=log_std_var)

def dist_info_sym(self, obs_var, state_info_vars):
        n_batches, n_steps = obs_var.shape[:2]
        obs_var = obs_var.reshape((n_batches, n_steps, -1))
        if self._state_include_action:
            prev_action_var = state_info_vars["prev_action"]
            all_input_var = TT.concatenate(
                [obs_var, prev_action_var],
                axis=2
            )
        else:
            all_input_var = obs_var
        means, log_stds = L.get_output(
            [self._mean_network.output_layer, self._l_log_std], all_input_var)
        return dict(mean=means, log_std=log_stds)

def dist_info_sym(self, obs_var, state_info_vars=None):
        return dict(
            prob=L.get_output(
                self._l_prob,
                {self._l_obs: obs_var}
            )
        )

def __init__(
            self,
            env_spec,
            hidden_sizes=(32, 32),
            hidden_nonlinearity=NL.tanh,
            num_seq_inputs=1,
            prob_network=None,
    ):
        """
        :param env_spec: A spec for the mdp.
        :param hidden_sizes: list of sizes for the fully connected hidden layers
        :param hidden_nonlinearity: nonlinearity used for each hidden layer
        :param prob_network: manually specified network for this policy, other network params
        are ignored
        :return:
        """
        Serializable.quick_init(self, locals())

        assert isinstance(env_spec.action_space, Discrete)

        if prob_network is None:
            prob_network = MLP(
                input_shape=(
                    env_spec.observation_space.flat_dim * num_seq_inputs,),
                output_dim=env_spec.action_space.n,
                hidden_sizes=hidden_sizes,
                hidden_nonlinearity=hidden_nonlinearity,
                output_nonlinearity=NL.softmax,
            )

        self._l_prob = prob_network.output_layer
        self._l_obs = prob_network.input_layer
        self._f_prob = ext.compile_function([prob_network.input_layer.input_var], L.get_output(
            prob_network.output_layer))

        self._dist = Categorical(env_spec.action_space.n)

        super(CategoricalMLPPolicy, self).__init__(env_spec)
        LasagnePowered.__init__(self, [prob_network.output_layer])

def dist_info_sym(self, obs_var, state_info_vars=None):
        return dict(prob=L.get_output(self._l_prob, {self._l_obs: obs_var}))

def log_likelihood_sym(self, x_var, y_var):
        normalized_xs_var = (x_var - self._x_mean_var) / self._x_std_var
        prob = L.get_output(
            self._l_prob, {self._prob_network.input_layer: normalized_xs_var})
        return self._dist.log_likelihood_sym(TT.cast(y_var, 'int32'), dict(prob=prob))

def log_likelihood_sym(self, x_var, y_var):
        normalized_xs_var = (x_var - self._x_mean_var) / self._x_std_var

        normalized_means_var, normalized_log_stds_var = \
            L.get_output([self._l_mean, self._l_log_std], {
                self._mean_network.input_layer: normalized_xs_var})

        means_var = normalized_means_var * self._y_std_var + self._y_mean_var
        log_stds_var = normalized_log_stds_var + TT.log(self._y_std_var)

        return self._dist.log_likelihood_sym(y_var, dict(mean=means_var, log_std=log_stds_var))

def prep(self, deterministic=False):
        layer_pairs = list(self.layer_iter)
        layers = [v for k,v in layer_pairs]
        names = [k for k,v in layer_pairs]
        outputs = get_output(layers, deterministic=deterministic)
        for name, output in zip(names, outputs):
            out_name = "{}_out".format(name)
            self.__dict__[out_name] = output