def max_add_feat(active_chl,rng,pf):
rng = rng
min_per=1
addfeat_data = 0
header = preproc.get_header(pf)
header_non_act = np.copy(header).tolist()
header_non_act.remove(active_chl)
header_non_act.remove('Elapsed time')
for chl in header_non_act:
sig_add = preproc.get_column(chl,pf)[2:]
sig_add = preproc.check(sig_add,return_series=1)[0]
chl_add = sig_add[rng[0]:rng[-1]]
if not discard_add_feat(sig_add,rng):
idx_isnan = np.isnan(chl_add)
idx = np.arange(len(chl_add))
idx_nan = idx[idx_isnan]
percent = len(idx_nan)/float(len(idx))
if min_per ==1:
min_per=percent
min_chl = sig_add
elif percent<=min_per:
min_per = percent
min_chl = sig_add
if min_per==0:
break
addfeat_data = min_chl[rng[0]-50000:rng[0]]
if addfeat_data!=0:
if np.isnan(addfeat_data).any():
idx = np.arange(len(addfeat_data))
idx_isnan = np.isnan(addfeat_data)
idx = idx[idx_isnan].tolist()
idx_start = idx[-1]
addfeat_data = addfeat_data[idx_start:]
return chl_add, addfeat_data
def add_feat(rng, active_chl,pf): ## name of active channel
active_chl= active_chl
rng = rng
pf = pf
tr=0
add_feat_pred,add_feat_tr = max_add_feat(active_chl,rng,pf)
print add_feat_pred,add_feat_tr
active_chl_array = preproc.get_column(active_chl,pf)[2:]
active_chl_array = preproc.check(active_chl_array,return_series=1)[0]
tr = gen_tr_data(rng,active_chl_array)
if tr!=0 and add_feat_tr!=0:
tr_len = min(len(tr),len(add_feat_tr))
tr = tr[-tr_len:]
add_feat_tr = add_feat_tr[-tr_len:]
return tr, add_feat_tr, add_feat_pred
else:
return 0
def interpolate(self):
sig_main = preproc.get_column(self.col,self.pf)
res = preproc.check(sig_main[2:],return_series=1)
if res[-1]>=0.9:
return sig_main
sig_main_gaps = res[3]
sig_array = res[0]
active_chl = self.col
for gap in sig_main_gaps:
rng = gap
data = gen_add_feat.add_feat(rng,active_chl,self.pf)
if data!=0:
tr, add_feat_tr,add_feat_pred = data[0],data[1],data[2]
nn = NeuralnetInterpolator(tr,add_feat_tr,add_feat_pred,self.window_length,self.bool_var)
new_sig = nn.interpolate()
sig_array[rng[0]:rng[-1]] = new_sig
else:
pass
sig_interpolated = sig_array
return sig_interpolated
def summarize(file_name):
file_path = '/Users/aubrey9012/Downloads/medical_ICU-features/data/mimic/'+str(file_name)+'.csv'
output_path = '/Users/aubrey9012/Downloads/medical_ICU-features/data/output/summary_of_'+str(file_name)+'.csv'
f = open(file_path, 'r')
output = open(output_path, 'wt')
writer = csv.writer(output)
reader = csv.reader(f)
h = reader.next()
header = [str(i)[1:-1] for i in h]
print header
f.close()
f = open(file_path, 'r')
key = ['black_cnt', 'gaps', 'gap_indeces', 'missing_value_proportion']
for col in header:
writer.writerow([col])
writer.writerow(key)
sig = preproc.get_column(col, file_path)
sig = sig[2:]
res = preproc.check(sig, return_series =False)
writer.writerow(res)
f.close()
output.close()
return
