How to use the torchx.layers.ReLU function in torchx

To help you get started, we’ve selected a few torchx examples, based on popular ways it is used in public projects.

Secure your code as it's written. Use Snyk Code to scan source code in minutes - no build needed - and fix issues immediately.

github SurrealAI / surreal / surreal / model / model_builders / builders.py View on Github external
def __init__(self, D_in, D_act, hidden_sizes=[300, 200], use_layernorm=True):
        super(ActorNetworkX, self).__init__()

        xp_input = L.Placeholder((None, D_in))
        xp = L.Linear(hidden_sizes[0])(xp_input)
        xp = L.ReLU()(xp)
        if use_layernorm:
            # Normalize 1 dimension
            xp = L.LayerNorm(1)(xp)
        xp = L.Linear(hidden_sizes[1])(xp)
        xp = L.ReLU()(xp)
        if use_layernorm:
            xp = L.LayerNorm(1)(xp)
        xp = L.Linear(D_act)(xp)
        xp = L.Tanh()(xp)

        self.model = L.Functional(inputs=xp_input, outputs=xp)
        self.model.build((None, D_in))
github SurrealAI / surreal / surreal / model / model_builders / builders.py View on Github external
def __init__(self, D_obs, D_act, hidden_sizes=[64, 64], init_log_sig=0):
        '''
            Constructor for PPO actor network
            Args: 
                D_obs: observation space dimension, scalar
                D_act: action space dimension, scalar
                hidden_sizes: list of fully connected dimension
                init_log_sig: initial value for log standard deviation parameter
        '''
        super(PPO_ActorNetwork, self).__init__()
        # assumes D_obs here is the correct RNN hidden dim
        xp_input = L.Placeholder((None, D_obs))
        xp = L.Linear(hidden_sizes[0])(xp_input)
        xp = L.ReLU()(xp)
        xp = L.Linear(hidden_sizes[1])(xp)
        xp = L.ReLU()(xp)
        xp = L.Linear(D_act)(xp)
        xp = L.Tanh()(xp)

        self.model = L.Functional(inputs=xp_input, outputs=xp)
        self.model.build((None, D_obs))

        self.log_var = nn.Parameter(torch.zeros(1, D_act) + init_log_sig)
github SurrealAI / surreal / surreal / model / model_builders / builders.py View on Github external
def __init__(self, D_in, D_act, hidden_sizes=[400, 300], use_layernorm=True):
        super(CriticNetworkX, self).__init__()

        xp_input_obs = L.Placeholder((None, D_in))
        xp = L.Linear(hidden_sizes[0])(xp_input_obs)
        xp = L.ReLU()(xp)
        if use_layernorm:
            xp = L.LayerNorm(1)(xp)
        self.model_obs = L.Functional(inputs=xp_input_obs, outputs=xp)
        self.model_obs.build((None, D_in))

        xp_input_concat = L.Placeholder((None, hidden_sizes[0] + D_act))
        xp = L.Linear(hidden_sizes[1])(xp_input_concat)
        xp = L.ReLU()(xp)
        if use_layernorm:
            xp = L.LayerNorm(1)(xp)
        xp = L.Linear(1)(xp)

        self.model_concat = L.Functional(inputs=xp_input_concat, outputs=xp)
        self.model_concat.build((None, D_act + hidden_sizes[0]))
github SurrealAI / surreal / surreal / model / model_builders / builders.py View on Github external
def __init__(self, D_obs, D_act, hidden_sizes=[64, 64], init_log_sig=0):
        '''
            Constructor for PPO actor network
            Args: 
                D_obs: observation space dimension, scalar
                D_act: action space dimension, scalar
                hidden_sizes: list of fully connected dimension
                init_log_sig: initial value for log standard deviation parameter
        '''
        super(PPO_ActorNetwork, self).__init__()
        # assumes D_obs here is the correct RNN hidden dim
        xp_input = L.Placeholder((None, D_obs))
        xp = L.Linear(hidden_sizes[0])(xp_input)
        xp = L.ReLU()(xp)
        xp = L.Linear(hidden_sizes[1])(xp)
        xp = L.ReLU()(xp)
        xp = L.Linear(D_act)(xp)
        xp = L.Tanh()(xp)

        self.model = L.Functional(inputs=xp_input, outputs=xp)
        self.model.build((None, D_obs))

        self.log_var = nn.Parameter(torch.zeros(1, D_act) + init_log_sig)
github SurrealAI / surreal / surreal / model / model_builders / builders.py View on Github external
def __init__(self, D_obs, D_out, conv_channels=[16, 32], kernel_sizes=[8, 4], strides=[4,2]):
        super(CNNStemNetwork, self).__init__()
        layers = []
        for i in range(len(conv_channels)):
            layers.append(L.Conv2d(conv_channels[i], kernel_size=kernel_sizes[i], stride=strides[i]))
            layers.append(L.ReLU())
        layers.append(L.Flatten())
        layers.append(L.Linear(D_out))
        layers.append(L.ReLU())
        self.model = L.Sequential(*layers)

        # instantiate parameters
        self.model.build((None, *D_obs))