def check_is_empty(annfile):
'''
Takes an annotation file and returns True is 'NA' is in the annotation
'''
annotation = utils.read_ann(annfile)
if "NA" in annotation:
return True
return False
def parse_ann():
for subject in subjects:
print(subject)
for x in range(24):
annfile = '/Users/francob/real_ann/{}/{}.ann'.format(subject, x)
lstfile = '/Users/francob/real_ann/{}/{}.lst'.format(subject, x)
if (os.path.exists(annfile) and os.path.exists(lstfile)):
words = [x.rstrip('\n').strip().upper() for x in open(lstfile)]
with open(annfile, 'rb') as ann:
real_hits = [
x['word'].upper() for x in utils.read_ann(annfile)
if x['word'].upper() in words
]
print(response)
def parse_all():
gs_total = []
ds_total = []
for subject in subjects:
print(subject)
tp = []
fp = []
fn = []
tl = []
tp2 = []
fp2 = []
fn2 = []
tl2 = []
for x in range(24):
annfile = '/Users/francob/real_ann/{}/{}.ann'.format(subject, x)
lstfile = '/Users/francob/real_ann/{}/{}.lst'.format(subject, x)
gspkl = '/Users/francob/{}/{}.pkl'.format(subject, x)
gslst = '/Users/francob/{}/{}.lst'.format(subject, x)
gs_seen = []
ds_seen = []
real_hits = []
pklfile = '/Users/francob/ram_analyses/ram_analyses/{}/{}.pkl'.format(
subject, x)
if os.path.exists(lstfile):
words = [x.rstrip('\n').strip().upper() for x in open(lstfile)]
if (os.path.exists(gspkl) and os.path.exists(gslst)):
gs_seen = [
x.upper() for x in parse(gspkl) if x.upper() in words
]
# print('Google Cloud')
# print(gs_seen)
if (os.path.exists(pklfile)):
words = [
x.rstrip('\n').strip().upper() for x in open(lstfile)
]
with open(pklfile, 'rb') as pkl:
ds_seen = [
x.upper() for x in pickle.load(pkl)
if x.upper() in words
]
if (os.path.exists(annfile) and os.path.exists(lstfile)):
with open(annfile, 'rb') as ann:
real_hits = [
x['word'].upper() for x in utils.read_ann(annfile)
if x['word'].upper() in words
]
if gs_seen:
# print('Real words')
# print(real_hits)
tp.append(
len(set(gs_seen).intersection(set(real_hits))))
fp.append(len(set(gs_seen) - set(real_hits)))
fn.append(len(set(real_hits) - set(gs_seen)))
tl.append(len(set(real_hits)))
if set(real_hits) != set(gs_seen):
print(subject, x)
print(real_hits)
print(gs_seen)
if ds_seen:
tp2.append(
len(set(ds_seen).intersection(set(real_hits))))
fp2.append(len(set(ds_seen) - set(real_hits)))
fn2.append(len(set(real_hits) - set(ds_seen)))
tl2.append(len(set(real_hits)))
# if nd gs_seen:
# print('Real words')
# print(real_hits)
# print('Deepspeech')
# print(ds_seen)
# print('Google Cloud')
# print(gs_seen)
gs_total.append([(sum(tl) - sum(fp) - sum(fn)) / max(sum(tl), 1),
sum(tp),
sum(fp),
sum(fn),
sum(tl)])
ds_total.append([(sum(tl2) - sum(fp2) - sum(fn2)) / max(sum(tl2), 1),
sum(tp2),
sum(fp2),
sum(fn2),
sum(tl2)])
print([x[0] for x in gs_total])
print([x[0] for x in ds_total])
data = [x[0] for x in gs_total if x[0] > 0]
def main():
'''
This is meant to be called from parsync.sh.
'''
rec_mod_dirs = find_recently_modified.find_cont_subdir(
'/home1/maint/parse_files/', 'session')
for dir in rec_mod_dirs:
#check if it has already been combined
with open() as comb_sess:
comb_sess_list = [x for x in comb_sess]
if dir not in comb_sess_list:
chunk_dir = os.path.join(dir, 'chunks/')
fta_dir = os.path.join(dir, 'files_to_annotate/')
#if it hasn't been combined, try to combine
if check_all_exist(fta_dir):
# in order to keep track of all the files, we look at the .times file - first, the one that shows the chunk files
time_files = glob.glob(dir + '*.times')
for tf in time_files:
base = os.path.basename(os.path.splitext(tf))
with open(tf) as time_list_file:
time_list = [
x.rstrip('\n').split(' ')[0]
for x in time_list_file
]
#now, we can find the ann
for i, chunk_start in enumerate(time_list):
current_chunk_ann_file = os.path.join(
chunk_dir, '{}_{}.ann'.format(chunk_start, i))
current_chunk_ann = utils.read_ann(
current_chunk_ann_file)
current_chunk_time_file = os.path.join(
chunk_dir, '{}_{}.times'.format(chunk_start, i))
#if the chunk file had unknown words then the file exists, otherwise it doesn't
if os.path.exists(current_chunk_time_file):
current_time_list = [
x.rstrip('\n').split(' ')[0]
for x in current_chunk_time_file
]
current_end_time_list = [
x.rstrip('\n').split(' ')[1]
for x in current_chunk_time_file
]
current_chunk_ann = [
x for x in current_chunk_ann if x.time < ft
]
#now, we remove the unks and surrounding words in the original ann file
unk_replacements = []
for i2, fta_start in enumerate(current_time_list):
#now, we find the user generated ann
fta_ann_file = '{}/{}_{}_{}.ann'.format(
x, i, i2, fta_start)
if check_is_empty(fta_ann_file):
pass
else:
fta_ann = utils.read_ann(fta_ann_file)
unk_replacements.extend(
utils.add_ann(fta_ann, fta_start))
current_chunk_ann.extend(unk_replacements)
write_par_file(current_chunk_ann,
os.path.join(dir, base + '.par'))
else:
email(
r'Some files did not exist in {}, was not automatically combined.',
'*****@*****.**')
def alt_main():
gs_total = []
fns_dict = defaultdict(int)
fps_dict = defaultdict(int)
subject_dict = {}
for subject in subjects:
if subject not in spanish:
print(subject)
tp = []
fp = []
fn = []
tl = []
for num in range(24):
annfile = '/data/eeg/{}/behavioral/FR1/session_0/{}.ann'.format(subject, num)
lstfile = '/data/eeg/{}/behavioral/FR1/session_0/{}.lst'.format(subject, num)
pklfile = '/scratch/francob/big_ram_analyses_alternatives/{}/{}.pkl'.format(subject, num)
gs_seen = []
ds_seen = []
real_hits = []
if os.path.exists(lstfile):
raw_words = [x.rstrip('\n').strip().lower() for x in open(lstfile)]
modify_dict = {modify(x):x for x in raw_words}
if (os.path.exists(pklfile)):
gs_seen = [x.lower() for x in parse(pklfile, raw_words, modify_dict) if x.lower() in raw_words]
if (os.path.exists(annfile)):
with open(annfile, 'rb') as ann:
real_hits = [x['word'].lower() for x in utils.read_ann(annfile) if x['word'].lower() in raw_words]
if gs_seen:
# print('Real words')
# print(real_hits)
# print('Cloud')
# print(gs_seen)
temp_tp = len(set(gs_seen).intersection(set(real_hits)))
temp_fp = len(set(gs_seen) - set(real_hits))
temp_fn = len(set(real_hits) - set(gs_seen))
temp_tl = len(set(real_hits))
tp.append(temp_tp)
fp.append(temp_fp)
fn.append(temp_fn)
tl.append(temp_tl)
temp_fns = [x for x in set(real_hits) - set(gs_seen)]
for x in temp_fns:
fns_dict[x] += 1
temp_fps = [x for x in set(gs_seen) - set(real_hits)]
for x in temp_fps:
fps_dict[x] += 1
if temp_fn + temp_fp > 0:
print(subject, num)
print('real_hits')
print(list(real_hits))
print('gs_seen')
print(list(gs_seen))
# print('raw_phonemes')
# print(list(raw_phonemes))
total_diff = ((sum(tl) - sum(fp) - sum(fn)) / float(max(sum(tl), 1)))
if sum(tl):
gs_total.append([total_diff, sum(tp), sum(fp), sum(fn), sum(tl)])
subject_dict[subject] = total_diff
diffs = [x[0] for x in gs_total]
tps = [x[1] for x in gs_total]
fps = [x[2] for x in gs_total]
fns = [x[3] for x in gs_total]
tls = [x[4] for x in gs_total]
print(fns_dict)
print(fps_dict)
print([(x, subject_dict[x]) for x in sorted(subject_dict, key=subject_dict.get, reverse=True)])
print(diffs)
# print(tls)
print((sum(tls) - sum(fps) - sum(fns)) / float(sum(tls)))
print(sum(fps) / float(sum(tls)))
print(sum(fps))
print(sum(fns))
print(sum(fns) / float(sum(tls)))
print(np.mean([subject_dict[x] for x in subject_dict]))
# print(subject_dict)
location_dict = {}
for subject in subject_dict:
if subject[-1] in location_dict:
location_dict[subject[-1]].append(subject_dict[subject])
else:
location_dict[subject[-1]] = [subject_dict[subject]]
print(location_dict)
for location in location_dict:
print(location)
print(np.mean(location_dict[location]))