def random(n_units, mean=None): h_network = nnet.random_linear_then_tanh_chain(n_units[:-1]) mean_network = nnet.Linear.random(n_units[-2], n_units[-1]) if mean is not None: mean_network.b.set_value(mean.astype(floatX)) sigma_network = nnet.NNet().add_layer(nnet.Linear.random(n_units[-2], n_units[-1])).add_layer(nnet.Exponential()) return GaussianSampler(h_network, mean_network, sigma_network)
def random(n_units, mean=None): h_network = nnet.random_linear_then_tanh_chain(n_units[:-1]) mean_network = nnet.Linear.random(n_units[-2], n_units[-1]) if mean is not None: mean_network.b.set_value(mean.astype(floatX)) sigma_network = nnet.NNet().add_layer( nnet.Linear.random(n_units[-2], n_units[-1])).add_layer(nnet.Exponential()) return GaussianSampler(h_network, mean_network, sigma_network)
def random(n_units, bias=None): mean_network = nnet.random_linear_then_tanh_chain(n_units[:-1]) mean_network.add_layer(nnet.Linear.random(n_units[-2], n_units[-1])) if bias is not None: mean_network.layers[-1].b.set_value(bias.astype(theano.config.floatX)) mean_network.add_layer(nnet.Sigmoid()) return BernoulliSampler(mean_network)
def random(n_units, bias=None): mean_network = nnet.random_linear_then_tanh_chain(n_units[:-1]) mean_network.add_layer(nnet.Linear.random(n_units[-2], n_units[-1])) if bias is not None: mean_network.layers[-1].b.set_value( bias.astype(theano.config.floatX)) mean_network.add_layer(nnet.Sigmoid()) return BernoulliSampler(mean_network)