def _init_estimate(self):
# mini-batch sets stride to None
stride = self.stride if self.stride else 1
###### init
self._cluster_centers_iter = []
self._init_centers_indices = {}
self._t_total = 0
traj_lengths = self.trajectory_lengths(stride=stride)
total_length = sum(traj_lengths)
if not self.n_clusters:
self.n_clusters = min(int(math.sqrt(total_length)), 5000)
self._logger.info("The number of cluster centers was not specified, "
"using min(sqrt(N), 5000)=%s as n_clusters." % self.n_clusters)
if self.init_strategy == 'kmeans++':
self._progress_register(self.n_clusters,
description="initialize kmeans++ centers", stage=0)
self._progress_register(self.max_iter, description="kmeans iterations", stage=1)
self._init_in_memory_chunks(total_length)
if self.init_strategy == 'uniform':
# gives random samples from each trajectory such that the cluster centers are distributed percentage-wise
# with respect to the trajectories length
with conditional(self.fixed_seed, random_seed(42)):
for idx, traj_len in enumerate(traj_lengths):
self._init_centers_indices[idx] = random.sample(list(range(0, traj_len)), int(
math.ceil((traj_len / float(total_length)) * self.n_clusters)))
return stride
def _init_estimate(self):
# mini-batch sets stride to None
stride = self.stride if self.stride else 1
###### init
self._init_centers_indices = {}
self._t_total = 0
traj_lengths = self.trajectory_lengths(stride=stride, skip=self.skip)
total_length = sum(traj_lengths)
if not self.n_clusters:
self.n_clusters = min(int(math.sqrt(total_length)), 5000)
self.logger.info("The number of cluster centers was not specified, "
"using min(sqrt(N), 5000)=%s as n_clusters." % self.n_clusters)
from pyemma.coordinates.data import DataInMemory
if not isinstance(self, MiniBatchKmeansClustering) and not isinstance(self.data_producer, DataInMemory):
n_chunks = self.data_producer.n_chunks(chunksize=self.chunksize, skip=self.skip, stride=self.stride)
self._progress_register(n_chunks, description="creating data array", stage='data')
if self.init_strategy == 'kmeans++':
self._progress_register(self.n_clusters,
description="initialize kmeans++ centers", stage=0)
self._progress_register(self.max_iter, description="kmeans iterations", stage=1)
self._init_in_memory_chunks(total_length)
if self.init_strategy == 'uniform':
# gives random samples from each trajectory such that the cluster centers are distributed percentage-wise
# with respect to the trajectories length
with random_seed(self.fixed_seed):
for idx, traj_len in enumerate(traj_lengths):
self._init_centers_indices[idx] = random.sample(list(range(0, traj_len)), int(
math.ceil((traj_len / float(total_length)) * self.n_clusters)))
from ._ext import kmeans as kmeans_mod
self._inst = kmeans_mod.Kmeans_f(self.n_clusters, self.metric, self.data_producer.ndim)
return stride