Beispiel #1
0
    def to_kanga(self, keys=None):
        keys = set(keys or self.vals.keys()) & set(
            ['sample', 'target_val', 'grad_val', 'accepted'])

        vals = {}

        for key in keys:
            if key == 'sample':
                vals[key] = self.get_samples().detach().cpu().numpy()
            elif key == 'target_val':
                vals[key] = self.get_target_vals().detach().cpu().numpy()
            elif key == 'grad_val':
                vals[key] = self.get_grad_vals().detach().cpu().numpy()
            elif key == 'accepted':
                vals[key] = np.array(self.vals['accepted'])

        return ChainArray(vals)
# %% Import packages

import kanga.plots as ps

from kanga.chains import ChainArray

from bnn_mcmc_examples.examples.mlp.pima.setting2.constants import diagnostic_iter_thres
from bnn_mcmc_examples.examples.mlp.pima.setting2.metropolis_hastings.constants import sampler_output_pilot_path
from bnn_mcmc_examples.examples.mlp.pima.setting2.model import model

# %% Load chain array

chain_array = ChainArray.from_file(keys=['sample', 'accepted'],
                                   path=sampler_output_pilot_path)

# %% Drop burn-in samples

chain_array.vals['sample'] = chain_array.vals['sample'][
    diagnostic_iter_thres:, :]
chain_array.vals['accepted'] = chain_array.vals['accepted'][
    diagnostic_iter_thres:]

# %% Plot traces of simulated chain

for i in range(model.num_params()):
    ps.trace(chain_array.get_param(i),
             title=r'Traceplot of $\theta_{{{}}}$'.format(i + 1),
             xlabel='Iteration',
             ylabel='Parameter value')

# %% Plot running means of simulated chain