sub_labels_inds = []
for reg in regions:
for lab_idx, lab in enumerate(labels):
if reg == lab.name:
sub_labels.append(lab)
sub_labels_inds.append(lab_idx)
fdr_thresh = 0.05
filelist = listdir(proc_dir + freq)
models = [{} for reg in regions]
for filename in filelist:
if not re.search("[0-9].pickle", filename):
continue
source_idx = int(re.search("src_([0-9]*)", filename).groups()[0])
dest_idx = int(re.search("dest_([0-9]*)", filename).groups()[0])
models[source_idx][dest_idx] = MixedLMResults.load("{}{}/{}".format(
proc_dir, freq, filename))
indep_vars = [
"Block[T.visselten]", "Block[T.visual]", "Wav[T.4000cheby]",
"Wav[T.4000fftf]", "Wav[T.7000Hz]", "Laut", "Angenehm", "visual-visselten"
]
iv_disp = {
"Block[T.visselten]": "audio-visselten",
"Block[T.visual]": "audio-visual",
"Wav[T.4000cheby]": "4000Hz-4000cheby",
"Wav[T.4000fftf]": "4000Hz-4000fftf",
"Wav[T.7000Hz]": "4000Hz-7000Hz",
"Laut": "Laut",
"Angenehm": "Angenehm",
"visual-visselten": "visual-visselten"
}
var_base = "C(Block, Treatment('rest'))" # stem of the condition names in statsmodels format
stat_conds = [var_base + "[T." + cond + "]" for cond in conds[1:]]
if calc_aic:
# get permutations
with open("/home/jev/ATT_dat/proc/dics_aic_perm.pickle", "rb") as f:
perms = pickle.load(f)
aics = {mod: np.empty(node_n) for mod in models}
aics_params = {mod: [None for n in range(node_n)] for mod in models}
aics_confint = {mod: [None for n in range(node_n)] for mod in models}
aics_pvals = {mod: [None for n in range(node_n)] for mod in models}
for mod in models:
for n_idx in range(node_n):
print(n_idx)
this_mod = MixedLMResults.load(
"{}{}/{}_reg70_lmm_{}.pickle".format(proc_dir, band, mod,
n_idx))
aics[mod][n_idx] = this_mod.aic
aics_pvals[mod][n_idx] = this_mod.pvalues
aics_params[mod][n_idx] = this_mod.params
aics_confint[mod][n_idx] = this_mod.conf_int()
# calculate the AIC delta thresholds from the permutations
null_tile = np.tile(np.expand_dims(aics["null"], 1), (1, 1024))
perm_simp_diff = perms["simple"] - null_tile
perm_simp_maxima, perm_simp_minima = (perm_simp_diff.max(axis=1),
perm_simp_diff.min(axis=1))
simp_thresh = np.quantile(perm_simp_minima, threshold / 2)
perm_avg_tile = np.tile(np.mean(perms["simple"], axis=1, keepdims=True),
(1, 1024))
perm_count = 200
perm_num = 5
vec_mat_len = 2415
effect_names = [
"C(Block, Treatment('rest'))[T.audio]",
"C(Block, Treatment('rest'))[T.visselten]",
"C(Block, Treatment('rest'))[T.visual]",
"C(Block, Treatment('rest'))[T.zaehlen]"
]
effects = ["audio", "visselten", "visual", "zaehlen"]
# load up the main result
mod_fits = []
for vml in range(vec_mat_len):
mod_fits.append(
MixedLMResults.load("{}{}/reg70_lmm_{}.pickle".format(
proc_dir, band, vml)))
# extract p and z values
eff_dict = {
e: {
"effect_name": en,
"pvals": [],
"zvals": []
}
for e, en in zip(effects, effect_names)
}
for effect in effects:
for mf in mod_fits:
pval = mf.pvalues.get(eff_dict[effect]["effect_name"])
zval = mf.tvalues.get(eff_dict[effect]["effect_name"])
eff_dict[effect]["pvals"].append(pval)
with open("{}{}".format(perm_dir, pf), "rb") as f:
this_perm = pickle.load(f)
perms["simple"] = np.hstack((perms["simple"], this_perm["simple"]))
perms["cond"] = np.hstack((perms["cond"], this_perm["cond"]))
aics = {mod: np.zeros(node_n) for mod in models}
aics_pvals = {mod: [None for n in range(node_n)] for mod in models}
aics_params = {mod: [None for n in range(node_n)] for mod in models}
aics_confint = {mod: [None for n in range(node_n)] for mod in models}
aics_predicted = {mod: [None for n in range(node_n)] for mod in models}
for mod in models:
for n_idx in range(node_n):
print(n_idx)
try:
this_mod = MixedLMResults.load(
"{}{}/{}_reg70_byresp_lmm_{}{}.pickle".format(
lmm_dir, band, mod, n_idx, z_name))
except:
continue
aics[mod][n_idx] = this_mod.aic
aics_pvals[mod][n_idx] = this_mod.pvalues
aics_params[mod][n_idx] = this_mod.params
aics_confint[mod][n_idx] = this_mod.conf_int()
predicts = pd.Series({en: 0
for en in this_mod.model.exog_names},
dtype=float)
for en_idx, en in enumerate(this_mod.model.exog_names):
vector = np.zeros(len(this_mod.model.exog_names))
vector[0] = 1
vector[en_idx] = 1
predicts[en] = this_mod.model.predict(this_mod.params, vector)