def train_mog_made(n_hiddens, act_fun, n_comps, mode):
assert is_data_loaded(), 'Dataset hasn\'t been loaded'
model = mades.MixtureOfGaussiansMade(data.n_dims,
n_hiddens,
act_fun,
n_comps,
mode=mode)
train(model, a_made)
save_model(model, 'made', mode, n_hiddens, act_fun, n_comps, False)
def __init__(self, n_inputs, n_hiddens, act_fun, n_layers, n_comps, batch_norm=True, input_order='sequential', mode='sequential', input=None, rng=np.random):
"""
Constructor.
:param n_inputs: number of inputs
:param n_hiddens: list with number of hidden units for each hidden layer
:param act_fun: name of activation function
:param n_layers: number of layers in the flow
:param n_comps: number of gaussians per conditional for the target made
:param batch_norm: whether to use batch normalization between layers
:param input_order: order of inputs of last made
:param mode: strategy for assigning degrees to hidden nodes: can be 'random' or 'sequential'
:param input: theano variable to serve as input; if None, a new variable is created
"""
# save input arguments
self.n_inputs = n_inputs
self.n_hiddens = n_hiddens
self.act_fun = act_fun
self.n_layers = n_layers
self.n_comps = n_comps
self.batch_norm = batch_norm
self.mode = mode
self.input = tt.matrix('x', dtype=dtype) if input is None else input
self.parms = []
# maf
self.maf = MaskedAutoregressiveFlow(n_inputs, n_hiddens, act_fun, n_layers, batch_norm, input_order, mode, self.input, rng)
self.bns = self.maf.bns
self.parms += self.maf.parms
self.input_order = self.maf.input_order
# mog made
input_order = input_order if input_order == 'random' else self.maf.mades[-1].input_order[::-1]
self.made = mades.MixtureOfGaussiansMade(n_inputs, n_hiddens, act_fun, n_comps, input_order, mode, self.maf.u, rng)
self.parms += self.made.parms
# log likelihoods
self.L = self.made.L + self.maf.logdet_dudx
self.L.name = 'L'
# train objective
self.trn_loss = -tt.mean(self.L)
self.trn_loss.name = 'trn_loss'
# theano evaluation functions, will be compiled when first needed
self.eval_lprob_f = None
self.eval_grad_f = None
self.eval_us_f = None