def create_models():
models = []
models.append(create_dense201())
models.append(create_dense161())
models.append(create_res101())
models.append(create_res152())
return models
def train_dense161():
print('training densenet 161')
model = create_dense161()
try:
load_best_weights(model)
except:
print('Failed to load weigths')
train(model, max_num=3)
def ensemble():
preds_raw = []
os.chdir(MODEL_DIR)
total_weight = 0
preds_w = None
for match_str in w_file_matcher:
w_files = glob.glob(match_str)
for w_file in w_files:
weight = 0
full_w_file = MODEL_DIR + '/' + w_file
if w_file.startswith('dense161'):
model, _ = create_dense161()
weight = 1
elif w_file.startswith('dense169'):
model, _ = create_dense169()
weight = 0.8
elif w_file.startswith('dense201'):
model, _ = create_dense201()
weight = 1
elif w_file.startswith('res50'):
model, _ = create_res50()
weight = 0.9
elif w_file.startswith('res101'):
model, _ = create_res101()
weight = 0.9
elif w_file.startswith('res152'):
model, _ = create_res152()
weight = 0.9
elif w_file.startswith('vgg16'):
model, _ = create_vgg16()
weight = 0.2
elif w_file.startswith('vgg19'):
model, _ = create_vgg19()
weight = 0.7
elif w_file.startswith('inceptionv3'):
model, _ = create_inceptionv3()
weight = 0.8
else:
pass
model.load_state_dict(torch.load(full_w_file))
print(full_w_file)
pred = make_preds(model, test_loader)
pred = np.array(pred)
preds_raw.append(pred)
if preds_w is None:
preds_w = np.zeros((pred.shape))
preds_w += pred * weight
total_weight += weight
del model
save_array(PRED_FILE_RAW, preds_raw)
preds = np.mean(preds_raw, axis=0)
#preds = preds_w / total_weight
save_array(PRED_FILE, preds)
def train_dense161():
print('training densenet 161')
model, w_file = create_dense161()
try:
model.load_state_dict(torch.load(w_file))
except:
print('{} not found, continue'.format(w_file))
pass
train(model, w_file, max_num=4)
def ensemble():
res101, _ = create_res101(True)
res152, _ = create_res152(True)
dense201, _ = create_dense201(True)
dense161, _ = create_dense161(True)
pred1 = np.array(make_preds(res101, test_loader))
pred2 = np.array(make_preds(res152, test_loader))
pred3 = np.array(make_preds(dense201, test_loader))
pred4 = np.array(make_preds(dense161, test_loader))
preds = np.mean([pred1, pred2, pred3, pred4], axis=0)
save_array(PRED_FILE, preds)
print(preds[:10])
def find_weights():
res101,_ = create_res101(True)
res152,_ = create_res152(True)
dense201,_ = create_dense201(True)
dense161,_ = create_dense161(True)
pred1, y1 = make_preds(res101, val_loader)
#print(y1)
#pred1 = np.random.random((600, 3))
#print(pred1[:5])
pred2, y2 = make_preds(res152, val_loader)
#print(y2)
pred3, y3 = make_preds(dense201, val_loader)
pred4, y4 = make_preds(dense161, val_loader)
#_, preds_test = torch.max(torch.from_numpy(pred4), 1)
#print(preds_test[:10])
print(pred1.shape)
#y1 = np.array(do_clip(one_hot(y1), 0.9))
print(y1[:5])
print(pred1[:5])
print(y1.shape, pred1.shape)
#print(y1.shape)
#print(log_loss(y1, y2))
#print(log_loss(y3, y2))
#print(log_loss(y3, y4))
print(log_loss(y1, pred1))
print(log_loss(y2, pred2))
#print(log_loss(np.array(one_hot(y2), float32), pred2))
#print(log_loss(np.array(one_hot(y3), float32), pred3))
#print(log_loss(np.array(one_hot(y4), float32), pred4))
y = []
for data, labels in val_loader:
for label in labels:
y.append(label)
print(np.array(y).shape)
#print(y)
return [pred1,pred2, pred3, pred4], y
def ensemble():
preds_raw = []
os.chdir(MODEL_DIR)
for match_str in w_file_matcher:
w_files = glob.glob(match_str)
for w_file in w_files:
full_w_file = MODEL_DIR + '/' + w_file
if w_file.startswith('dense161'):
model, _ = create_dense161()
elif w_file.startswith('dense169'):
model, _ = create_dense169()
elif w_file.startswith('dense201'):
model, _ = create_dense201()
elif w_file.startswith('res50'):
model, _ = create_res50()
elif w_file.startswith('res101'):
model, _ = create_res101()
elif w_file.startswith('res152'):
model, _ = create_res152()
elif w_file.startswith('vgg16'):
model, _ = create_vgg16()
elif w_file.startswith('vgg19'):
model, _ = create_vgg19()
elif w_file.startswith('inceptionv3'):
model, _ = create_inceptionv3()
else:
pass
model.load_state_dict(torch.load(full_w_file))
print(full_w_file)
pred = make_preds(model, test_loader)
pred = np.array(pred)
preds_raw.append(pred)
del model
save_array(PRED_FILE_RAW, preds_raw)
preds = np.mean(preds_raw, axis=0)
save_array(PRED_FILE, preds)
def ensemble(stage):
preds_weighted = None
total_weight = 0
preds_raw = []
preds_raw_weighted = []
for match_str in w_file_matcher:
#print(match_str)
os.chdir(MODEL_DIR)
w_files = glob.glob(match_str)
#print('cur:' + os.getcwd())
for w_file in w_files:
full_w_file = MODEL_DIR + '/' + w_file
print(full_w_file)
if w_file.startswith('dense161'):
model = create_dense161()
elif w_file.startswith('dense169'):
model = create_dense169()
elif w_file.startswith('dense201'):
model = create_dense201()
elif w_file.startswith('dense121'):
model = create_dense121()
elif w_file.startswith('res50'):
model = create_res50()
elif w_file.startswith('res101'):
model = create_res101()
elif w_file.startswith('res152'):
model = create_res152()
elif w_file.startswith('vgg16bn'):
model = create_vgg16bn()
elif w_file.startswith('vgg19bn'):
model = create_vgg19bn()
elif w_file.startswith('vgg19'):
model = create_vgg19()
elif w_file.startswith('inceptionv3'):
model = create_inceptionv3()
else:
print('No model for {}'.format(full_w_file))
continue
model.load_state_dict(torch.load(full_w_file))
pred, filenames = make_preds(model, stage)
pred = np.array(pred)
preds_raw.append(pred)
weight = get_weight(get_acc_from_w_filename(full_w_file))
preds_raw_weighted.append((pred, weight))
if preds_weighted is None:
preds_weighted = np.zeros(pred.shape)
preds_weighted = preds_weighted + pred * weight
total_weight += weight
del model
preds_weighted = preds_weighted / total_weight
save_array(PRED_FILE_RAW, preds_raw)
save_array(PRED_FILE_RAW_WEIGHTED, preds_raw_weighted)
preds = np.mean(preds_raw, axis=0)
save_array(PRED_FILE, preds)
save_array(PRED_FILE_WEIGHTED, preds_weighted)
save_array(TEST_FILE_NAMES, filenames)