def test_dense(input_size, output_size, n_samples):
params = dict()
params["w"] = npr.normal(size=(input_size, output_size))
params["b"] = npr.normal(size=(output_size, ))
x = npr.normal(size=(n_samples, input_size))
output = dense(params, x)
assert output.shape == (n_samples, output_size)
def model(params, Fs, As):
Fs = mpnn(params["graph1"], As, Fs, nonlin=relu)
Fs = mpnn(params["graph2"], As, Fs, nonlin=relu)
out = gather(Fs)
out = dense(params["dense1"], out, nonlin=relu)
return out
def decoder(params, x):
a = dense(params["dec1"], x, nonlin=tanh)
a = dense(params["dec2"], a, nonlin=tanh)
output = dense(params["dec3"], a, nonlin=sigmoid)
return output
def model(p, x):
"""Forward neural network model."""
x = dense(p["dense1"], x, nonlin=relu)
x = dense(p["dense2"], x)
return x
def encoder(params, x):
a = dense(params["enc1"], x, nonlin=tanh)
a = dense(params["enc2"], a, nonlin=tanh)
z_mean = dense(params["mean"], a)
z_log_var = dense(params["logvar"], a)
return z_mean, z_log_var