def getInps(spl = 0.3, Topt = np.array([1])):
inps = ld_obj('../data/inps_RCC')
ys = inps[0]+inps[2]
xs = inps[1]+inps[3]
xs_demo = np.stack(elt[1] for elt in xs)
inps_all = ld_obj('../data/inps')
new_inps = []
for val in [1,3,5]:
inds = []
for elt in inps_all[val]:
locn = np.where(np.all(elt[1] == xs_demo,axis=1))[0]
if locn.size == 1:
inds.append(locn[0])
inds=np.array(inds)
new_inps.append([bool(ys[elt]) for elt in inds])
new_inps.append([xs[elt] for elt in inds])
[trnLs, trnIms, tstLs, tstIms, valLs, valIms] = new_inps
[tstImsF, tstLsF] = ld_obj('../data/inps_new_data_Fuhrman_cc.pkl')
trnLs, trnIms = trnLs + tstLsF[:14], trnIms + tstImsF[:14]
tstLs, tstIms = tstLs + tstLsF[14:28], tstIms + tstImsF[14:28]
valLs, valIms = valLs + tstLsF[28:], valIms + tstImsF[28:]
[tstImsI, tstLsI] = ld_obj('../data/inps_new_data_ISUP_cc.pkl')
return trnLs, trnIms, tstLs, tstIms, valLs, valIms, tstLsI, tstImsI
def getInps(spl=0.3, Topt=np.array([1])):
[trnLs, trnIms, tstLs, tstIms, valLs, valIms] = ld_obj('../data/inps')
[tstImsF, tstLsF] = ld_obj('../data/inps_new_data_Fuhrman.pkl')
trnLs, trnIms = trnLs + tstLsF[:14], trnIms + tstImsF[:14]
tstLs, tstIms = tstLs + tstLsF[14:28], tstIms + tstImsF[14:28]
valLs, valIms = valLs + tstLsF[28:], valIms + tstImsF[28:]
[tstImsI, tstLsI] = ld_obj('../data/inps_new_data_ISUP.pkl')
return trnLs, trnIms, tstLs, tstIms, valLs, valIms, tstLsI, tstImsI
model.summary()
def with_substr(ls, sub_str):
return [elt for elt in ls if sub_str in elt]
data_list = os.listdir('.')
input_hds = with_substr(data_list, '.hd')
input_pkls = with_substr(data_list, '.pkl')
####################
[trnIms, valIms, tstIms, trnLs, valLs, tstLs, trnDs, valDs,
tstDs] = ld_obj(root + 'all_input_data')
#trnLs, trnIms, tstLs, tstIms, valLs, valIms = getInps()
x_train, y_train = trnIms[:, :, :, :, 0], trnLs
x_val, y_val = valIms[:, :, :, :, 0], valLs
x_tst, y_tst = tstIms[:, :, :, :, 0], tstLs
x_train = np.stack([(im - np.mean(im)) / np.std(im) for im in x_train], axis=0)
x_val = np.stack([(im - np.mean(im)) / np.std(im) for im in x_val], axis=0)
x_tst = np.stack([(im - np.mean(im)) / np.std(im) for im in x_tst], axis=0)
evl_tr = model.evaluate(x=x_train, y=y_train)
print('train: ' + str(evl_tr))
evl_vl = model.evaluate(x=x_val, y=y_val)
print('val: ' + str(evl_vl))
def getInps(spl=0.3, Topt=np.array([1])):
[trnLs, trnIms, tstLs, tstIms, valLs, valIms] = ld_obj('../data/inps')
return trnLs, trnIms, tstLs, tstIms, valLs, valIms
)
##just nav to directory
data_list = os.listdir('.')
for elt in data_list:
print(elt[:3] == 'all')
print(elt[3] != '_')
print([elt for elt in data_list if ('all' == elt[:3] and (elt[3] != '_'))])
input_hds = with_substr(data_list, '.hd')
input_pkls = with_substr(data_list, '.pkl')
[
history, mn, s, x_train, y_train, x_val, y_val, x_tst, y_tst, pred1_tr,
pred1_vl, pred1_ts
] = ld_obj(
[elt for elt in data_list if ('all' == elt[:3] and (elt[3] != '_'))][0])
model = LR(x_train.shape[1], 2)
chckptNm = with_substr(input_hds, 'all')[0]
model.compile(optimizer=optim,
loss='categorical_crossentropy',
metrics=['accuracy'])
model.load_weights(chckptNm)
evl_tr = model.evaluate(x=x_train, y=y_train)
print('train: ' + str(evl_tr))
evl_vl = model.evaluate(x=x_val, y=y_val)
print('val: ' + str(evl_vl))
