def execute_round(ensemble, r, n, rule, n_clusts):
logger.info("- - - - - - - - - - - - - - - - - -")
logger.info("Sampled of {} patterns.".format(n))
logger.info("Rule: {}".format(rule.__doc__))
logger.info("kmin: {}, kmax: {}".format(n_clusts[0], n_clusts[1]))
logger.info("Type of mat: {}".format(type_mats[tm]))
logger.info("Type of mat tm: {}".format(tm))
logger.info("Round: {}".format(r))
logger.info("Building matrix...")
logger.info("Estimated req. mem. (MB):{}".format(mems[tm] / (1024.0**2)))
useDiskMST = False
if tm == 0: # full
if f_mat > mem_full_max:
logger.info("not enough memory")
break
eacEst = myEAC.EAC(n_samples=n, sparse=False, condensed=False)
t.reset()
t.tic()
eacEst.buildMatrix(ensemble)
t.tac()
eacEst.coassoc.getDegree()
degree = eacEst.coassoc.degree
nnz = eacEst.coassoc.nnz
n_max_degree = -1
elif tm == 1: # full condensed
if fc_mat > mem_full_max:
logger.info("not enough memory")
break
eacEst = myEAC.EAC(n_samples=n, sparse=False, condensed=True)
t.reset()
t.tic()
eacEst.buildMatrix(ensemble)
t.tac()
eacEst.coassoc.getDegree()
degree = eacEst.coassoc.degree
nnz = eacEst.coassoc.nnz
n_max_degree = -1
elif tm == 2: # sparse complete
if sp_const > mem_full_max:
logger.info("not enough memory")
break
if sp_const + sp_const_mst > mem_full_max:
useDiskMST = True
eacEst = myEAC.EAC(n_samples=n, sparse=True, condensed=False,
sparse_keep_degree=True, sl_disk=useDiskMST)
eacEst.sp_max_assocs_mode="constant"
eacEst.disk_dir = diskdir
t.reset()
t.tic()
eacEst.buildMatrix(ensemble)
t.tac()
degree = eacEst.coassoc.degree[:-1]
nnz = eacEst.coassoc.nnz
n_max_degree = (degree == ma).sum()
elif tm == 3: # sparse condensed const
if sp_const > mem_full_max:
logger.info("not enough memory")
break
if sp_const + sp_const_mst > mem_full_max:
useDiskMST = True
eacEst = myEAC.EAC(n_samples=n, sparse=True, condensed=True,
sparse_keep_degree=True, sl_disk=useDiskMST)
eacEst.sp_max_assocs_mode="constant"
eacEst.disk_dir = diskdir
t.reset()
t.tic()
eacEst.buildMatrix(ensemble)
t.tac()
degree = eacEst.coassoc.degree[:-1]
nnz = eacEst.coassoc.nnz
n_max_degree = (degree == ma).sum()
elif tm == 4: # sparse condensed linear
if sp_lin > mem_full_max:
logger.info("not enough memory")
break
if sp_lin + sp_lin_mst > mem_full_max:
useDiskMST = True
eacEst = myEAC.EAC(n_samples=n, sparse=True, condensed=True,
sparse_keep_degree=True, sl_disk=useDiskMST)
eacEst.sp_max_assocs_mode="linear"
eacEst.disk_dir = diskdir
t.reset()
t.tic()
eacEst.buildMatrix(ensemble)
t.tac()
degree = eacEst.coassoc.degree[:-1]
nnz = eacEst.coassoc.nnz
indptr = mySpEAC.indptr_linear(n,
eacEst.sp_max_assocs,
n_s, n_e, val_s, val_e)
max_degree = indptr[1:] - indptr[:-1]
n_max_degree = (degree == max_degree).sum()
else:
raise NotImplementedError("mat type {} not implemented".format(type_mats[tm]))
logger.info("Build time: {}".format(t.elapsed))
results.round[res_idx] = r # round number
results.n_samples[res_idx] = n # n_samples
results.rule[res_idx] = rule.__doc__ # rule
results.kmin[res_idx] = n_clusts[0] # kmin
results.kmax[res_idx] = n_clusts[1] # kmax
results.t_build[res_idx] = t.elapsed
results.type_mat[res_idx] = type_mats[tm] # type of matrix
results.disk[res_idx] = useDiskMST # SL disk
results.t_ensemble[res_idx] = t.elapsed # ensemble time
results.biggest_cluster[res_idx] = max_cluster_size # biggest_cluster
# number of associations
results.n_assocs[res_idx] = nnz
# stats number associations
results.max_degree[res_idx] = degree.max()
results.min_degree[res_idx] = degree.min()
results.mean_degree[res_idx] = degree.mean()
results.std_degree[res_idx] = degree.std()
results.n_max_degree[res_idx] = n_max_degree
logger.info("SL clustering...")
t.reset()
t.tic()
labels = eacEst.finalClustering(n_clusters=0)
t.tac()
logger.info("Clustering time: {}".format(t.elapsed))
labels_filename = "labels_{}_{}_{}".format(n, rule.__doc__, r)
np.save("")
results.t_sl[res_idx] = t.elapsed # build time
results.sl_clusts[res_idx] = eacEst.n_fclusts
t.reset()
t.tic()
# logger.info("Scoring accuracy (consistency)...")
# accEst = myAcc.ConsistencyIndex(n)
# accEst.score(gt_sampled, labels)
logger.info("Scoring accuracy (Hungarian)...")
accEst = myAcc.HungarianIndex(n)
accEst.score(gt_sampled, labels)
t.tac()
logger.info("Accuracy time: {}".format(t.elapsed))
results.t_accuracy[res_idx] = t.elapsed # accuracy time
results.accuracy[res_idx] = accEst.accuracy
# if the accuracy is zero of different from the last round
# then I want to check what's going on
if accEst.accuracy == 0 or accEst.accuracy != results.accuracy[res_idx-1]:
print "breakpoint"
results.to_csv(os.path.join(folder,"results_kmin.csv"))
res_idx += 1
del eacEst, accEst, degree, labels
gc.collect()
# end of inner most loop
useDiskMST = True
eacEst = myEAC.EAC(n_samples=n,
sparse=True,
condensed=True,
sparse_keep_degree=True)
eacEst.sp_max_assocs_mode = "linear"
eacEst.disk_dir = diskdir
t.reset()
t.tic()
eacEst.buildMatrix(ensemble)
t.tac()
degree = eacEst.coassoc.degree[:-1]
nnz = eacEst.coassoc.nnz
indptr = mySpEAC.indptr_linear(n, eacEst.sp_max_assocs,
n_s, n_e, val_s, val_e)
max_degree = indptr[1:] - indptr[:-1]
n_max_degree = (degree == max_degree).sum()
else:
raise NotImplementedError(
"mat type {} not implemented".format(type_mats[tm]))
logger.info("Build time: {}".format(t.elapsed))
results.round[res_idx] = r # round number
results.n_samples[res_idx] = n # n_samples
results.rule[res_idx] = rule.__doc__ # rule
results.kmin[res_idx] = n_clusts[0] # kmin
results.kmax[res_idx] = n_clusts[1] # kmax
results.t_build[res_idx] = t.elapsed
