def find_or_retrieve_GMM_labels(dataset, eeg_date, eeg_gcoh_name, real_evs, iL, iH, fL, fH, armv_ver, gcoh_ver, which=0, try_K=[1, 2, 3, 4, 5, 6, 7], TRs=[1, 2, 4, 8, 16, 32, 64], manual_cluster=False, ignore_stored=False, do_pca=False, min_var_expld=0.95, dontsave=False, Bayesian=False):
###############
if not dontsave:
outdir = datconf.getResultFN(dataset, "%(rpsm)s/v%(av)d%(gv)d" % {"rpsm" : eeg_date, "w" : which, "av" : armv_ver, "gv" : gcoh_ver})
if not os.access(outdir, os.F_OK):
os.mkdir(outdir)
fn = "%(od)s/%(eeg)s_%(fL)d-%(fH)d_GMM_labels%(w)d" % {"od" : outdir, "eeg" : eeg_gcoh_name, "w" : which, "fL" : fL, "fH" : fH}
if (not ignore_stored) and os.access(fn, os.F_OK):
rmpd_lab = _N.loadtxt(fn, dtype=_N.int)
nStates = len(_N.unique(rmpd_lab))
else:
print("doing GMM")
minK = _N.min(try_K)
maxK = _N.max(try_K)
bics = _N.ones(((maxK-minK), _N.max(TRs)))*1000000
labs = _N.empty((maxK-minK, _N.max(TRs), real_evs.shape[0]), dtype=_N.int)
_features = _N.mean(real_evs[:, iL:iH], axis=1)
if do_pca:
pca = PCA()
pca.fit(_features)
proj = _N.einsum("ni,mi->nm", pca.components_, _features)
print(pca.explained_variance_ratio_)
maxC = _N.where(_N.cumsum(pca.explained_variance_ratio_) > min_var_expld)[0][0]
features = proj[0:maxC].T
print("Using %d features" % maxC)
else:
features = _features
if not Bayesian:
for K in range(minK, maxK):
for tr in range(TRs[K]):
gmm = mixture.GaussianMixture(n_components=K, covariance_type="full")
gmm.fit(features)
bics[K-minK, tr] = gmm.bic(features)
labs[K-minK, tr] = gmm.predict(features)
coords = _N.where(bics == _N.min(bics))
bestLab = labs[coords[0][0], coords[1][0]] # indices in 2-D array
nStates = list(range(minK, maxK))[coords[0][0]]
print("@!!!!! ")
print(coords)
else:
gmm = mixture.BayesianGaussianMixture(n_components=15, weight_concentration_prior_type="dirichlet_process", covariance_type="full")
gmm.fit(features)
bestLab = gmm.predict(features)
nStates = len(_N.unique(bestLab))
rmpd_lab = remap_label_by_similarity(nStates, bestLab, _features) # raw features
#rmpd_lab = increasing_labels_mapping(bestLab)
if not dontsave:
print("saving!!!!!!!!!!!!!!!! %s" % fn)
_N.savetxt(fn, rmpd_lab, fmt="%d")
print("manual_clustering %s" % str(manual_cluster))
if manual_cluster:
if os.access("%s_manual" % fn, os.F_OK): # open file of mapping
print("found manual cluster for %s" % fn)
mapping = _N.loadtxt("%s_manual" % fn, dtype=_N.int)
# col 0 is original cluster ID
# col 1 is remapped cluster ID
where_origs = []
for imaps in range(mapping.shape[0]):
orig_ID = mapping[imaps, 0]
new_ID = mapping[imaps, 1]
where_origs.append(_N.where(rmpd_lab == orig_ID)[0])
for imaps in range(mapping.shape[0]):
orig_ID = mapping[imaps, 0]
new_ID = mapping[imaps, 1]
rmpd_lab[where_origs[imaps]] = new_ID
else:
print("!! No manual cluster for %s. Using original mapping" % fn)
return nStates, rmpd_lab
all_pwr2 = _N.zeros(N_psd)
all_pwr12 = _N.zeros(N_psd)
fs_use = _N.linspace(0, 1.2, N_psd)
MULT = 1. # kernel width
lags_sec=50
ii = -1
for partID in partIDs:
ii += 1
rpsm_key = rpsms.rpsm_partID_as_key[partID]
sections = 4
pikdir = datconf.getResultFN(datconf._RPS, "%(dir)s/v%(av)d%(gv)d" % {"dir" : rpsm_key, "av" : armv_ver, "gv" : gcoh_ver})
label = 71
outdir = pikdir#"%(pd)s/%(lb)d_%(st)s_%(sec)d_%(toff)d" % {"pd" : pikdir, "lb" : label, "st" : shfl_type_str[shfl_type], "toff" : t_offset, "sec" : sections}
lm = depickle("%(od)s/%(rk)s_%(w)d_%(s)d_pkld_dat_v%(av)d%(gv)d_%(lb)d.dmp" % {"rk" : partID, "w" : win, "s" : slideby, "av" : armv_ver, "gv" : gcoh_ver, "od" : pikdir, "lb" : label})
if not os.access(outdir, os.F_OK):
os.mkdir(outdir)
_behv_sigs_all = _N.array(lm["behv"])
bigchg_behv_sig_ts = lm["behv_ts"][0:_behv_sigs_all.shape[1]-stop_early] - t_offset
win_gcoh = lm["win_gcoh"]
slide_gcoh = lm["slide_gcoh"]
win_spec = lm["win_spec"]
slide_spec = lm["slide_spec"]
"dsf": dat,
"av": armv_ver,
"gv": gcoh_ver,
"wn": win,
"sld": slideby,
"tr2tr": str2tr
}))
# #lm = depickle("../Neurable/DSi_dat/%(dat)s_gcoh_%(w)s_%(s)s.dmp" % {"dat" : dat, "w" : bin, "s" : slide})
# #A_gcoh_mat = _scio.loadmat("DSi_dat/%(dat)s_gcoh_%(w)d_%(sl)d.mat" % {"dat" : dat, "w" : bin, "sl" : slide})
# #A_gcoh = A_gcoh_mat["Cs"]
strt = 0 # if start at middle of experiment
A_gcoh = lm["Cs"][strt:]
n_fs = lm["fs"]
outdir1 = datconf.getResultFN(dataset, "%(dir)s" % {"dir": dat})
if not os.access(outdir1, os.F_OK):
os.mkdir(outdir1)
outdir = datconf.getResultFN(
dataset, "%(dir)s/v%(av)d%(gv)d" % {
"dir": dat,
"av": armv_ver,
"gv": gcoh_ver
})
if not os.access(outdir, os.F_OK):
os.mkdir(outdir)
# 12.10.2020
################ egenvectors
#imag_evs = A_gcoh_mat["VEC"][0]
imag_evs = lm["VEC"][strt:, :, ev_n]