def refine_akmeans(data, datax2_clusterx, centroids, max_iters=5,
flann_params={}, cache_dir=None, cfgstr='',
use_data_hash=True, akmeans_cfgstr=None):
""" Refines the approximates centroids """
print('[akmeans.precompute] refining:')
if akmeans_cfgstr is None:
akmeans_cfgstr = nn.get_flann_cfgstr(data, flann_params, cfgstr, use_data_hash)
datax2_clusterx_old = datax2_clusterx
(datax2_clusterx, centroids) = _akmeans_iterate(data, centroids, datax2_clusterx_old, max_iters, flann_params, 0, 10)
ut.save_cache(cache_dir, CLUSTERS_FNAME, akmeans_cfgstr, centroids)
ut.save_cache(cache_dir, DATAX2CL_FNAME, akmeans_cfgstr, datax2_clusterx)
return (datax2_clusterx, centroids)
def precompute_akmeans(data, num_clusters, max_iters=5, flann_params={},
cache_dir=None, force_recomp=False, use_data_hash=True,
cfgstr='', refine=False, akmeans_cfgstr=None):
""" precompute aproximate kmeans with builtin caching """
print('[akmeans] pre_akmeans()')
# filename prefix constants
assert cache_dir is not None, 'choose a cache directory'
# Build a cfgstr if the full one is not specified
if akmeans_cfgstr is None:
# compute a hashstr based on the data
akmeans_cfgstr = nn.get_flann_cfgstr(data, flann_params, cfgstr, use_data_hash)
try:
# Try and load a previous clustering
if force_recomp:
raise UserWarning('forceing recommpute')
centroids = ut.load_cache(cache_dir, CLUSTERS_FNAME, akmeans_cfgstr)
datax2_clusterx = ut.load_cache(cache_dir, DATAX2CL_FNAME, akmeans_cfgstr)
print('[akmeans.precompute] load successful')
if refine:
# Refines the cluster centers if specified
(datax2_clusterx, centroids) =\
refine_akmeans(data, datax2_clusterx, centroids,
max_iters=max_iters, flann_params=flann_params,
cache_dir=cache_dir, akmeans_cfgstr=akmeans_cfgstr)
return (datax2_clusterx, centroids)
except IOError as ex:
ut.printex(ex, 'cache miss', iswarning=True)
except UserWarning:
pass
# First time computation
print('[akmeans.precompute] pre_akmeans(): calling akmeans')
(datax2_clusterx, centroids) = akmeans(data, num_clusters, max_iters, flann_params)
print('[akmeans.precompute] save and return')
ut.save_cache(cache_dir, CLUSTERS_FNAME, akmeans_cfgstr, centroids)
ut.save_cache(cache_dir, DATAX2CL_FNAME, akmeans_cfgstr, datax2_clusterx)
return (datax2_clusterx, centroids)