def evaluate_cluster_purities_all_states_(self):
env = LehnertGridworld()
all_states, all_state_labels = env.get_all_states()
all_states = np.array(all_states)
all_state_labels = np.array(all_state_labels)
_, log_probs = self.model.encode(all_states)
probs = np.exp(log_probs)
dataset = ArrayDataset({constants.STATE_LABELS: all_state_labels})
cluster = Cluster(None, dataset)
cluster.num_clusters = self.num_components
cluster_map = cluster.assign_label_to_each_cluster_soft(probs)
purities, sizes, mean_purity = cluster.evaluate_purity_soft(probs)
mean_i_purity, i_scores, i_totals = cluster.evaluate_inverse_purity_soft(
probs, cluster_map)
self.logger.info("all states mixture soft purity: {:.2f}%".format(
mean_purity * 100))
self.logger.info(
"all states mixture inverse soft purity: {:.2f}%".format(
mean_i_purity * 100))
self.saver.save_array_as_txt([mean_purity, mean_i_purity],
"cluster_purity_all_states")
def evaluate_purities_all_states_(self):
dataset = ArrayDataset({constants.STATE_LABELS: self.all_state_labels})
cluster = Cluster(None, dataset)
cluster.num_clusters = len(self.partition)
probs = np.zeros((len(self.all_states), len(self.partition)))
for block_idx, block in enumerate(self.partition):
for state_idx in block:
probs[state_idx, block_idx] = 1.0
cluster_map = cluster.assign_label_to_each_cluster_soft(probs)
purities, sizes, mean_purity = cluster.evaluate_purity_soft(probs)
mean_i_purity, i_scores, i_totals = cluster.evaluate_inverse_purity_soft(
probs, cluster_map)
self.logger.info("state mixture soft purity: {:.2f}%".format(
mean_purity * 100))
self.logger.info("state mixture inverse soft purity: {:.2f}%".format(
mean_i_purity * 100))
self.saver.save_array_as_txt(
[mean_purity, mean_i_purity,
len(self.partition)], "cluster_purity")