def create_features(img):
img_gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
features, _ = train.create_features(img, img_gray, label=None, train=False)
return features
def gen_predictions(file, h_lick_p, args, model):
f_b_name = os.path.basename(file)
dst_f_path = os.path.join(output_dir, 'mask_pred_' + f_b_name)
dst_f_path_bin = os.path.join(output_dir,
'mask_pred_binarized_' + f_b_name)
#Get image and file
tmp_img, tmp_labl = get_mask_n_img_f(file)
#If file is present for prediction just read in and return else generate
#texture features and image from scratch where required
if os.path.isfile(dst_f_path + '_not_a_file_result'):
return imageio.imread(dst_f_path).flatten(), tmp_labl.flatten()
else:
#Generate features for model based on initial analysis.
features, labls = create_features(os.path.splitext(f_b_name)[0],
tmp_img,
tmp_labl,
h_lick_p,
args,
train=False)
#Minx max scaling same parameters taken from training script
features, _ = min_max_scaling(features)
predictions = compute_prediction(features, model)
#ipdb.set_trace()
write_img_to_file(predictions, tmp_img, dst_f_path)
#Writing binary logit to file for analysis.
gen_img_visual(tmp_img, predictions.reshape((256, 256)), tmp_labl,
dst_f_path_bin)
return predictions.flatten(), labls
def predict_all(model_file, input_fle, n):
w = load_model(model_file)
result = []
for sentence in input_fle:
phi = defaultdict(lambda: 0)
phi = create_features(sentence.lower(), n, phi)
result.append(predict_one(w, phi))
return result
def compute_prediction(fn, model):
border = int((h_neigh-1)/2)
img = cv2.imread(fn, cv2.IMREAD_COLOR)
cached_fn = os.path.join(cache_dir, fn.split(
os.path.sep)[-1].split('.')[0]+".pkl")
if os.path.isfile(cached_fn):
features = load_pickle(cached_fn)
else:
features = train.create_features(img)
dump2pickle(features, cached_fn)
scaler = load_pickle(os.path.join(model_dir, scaler_fn))
features = features.reshape(-1, features.shape[1])
features = scaler.transform(features)
model_predictions = model.predict_proba(features)
model_predictions = prob2class(model_predictions)
predictions_image = model_predictions.reshape(
[img.shape[0]-2*border, img.shape[1]-2*border, -1])
return predictions_image
from train import create_features
from train import predict_one
from collections import defaultdict
import sys
with open(sys.argv[1], 'r') as modelfile, open(sys.argv[2], 'r') as testfile:
w = defaultdict(lambda: 0)
for line in modelfile:
spl = line.strip().split('\t')
w[spl[0]] = int(spl[1])
for x in testfile:
phi = create_features(x)
y2 = predict_one(w, phi)
print(y2)
# test.py
# coding=utf-8
from collections import defaultdict
import sys, re, train
#重み
Weight = defaultdict(lambda: 0)
with open("ans", "r") as modelFile:
for line in modelFile:
key, value = line.split()
Weight[key] = float(value)
# reg_model = re.compile("^(-?1)(.*)$")
with open(sys.argv[1], "r") as inputFile:
for line in inputFile:
X = line.strip()
phi = train.create_features(X)
Y = train.predict_one(Weight, phi)
print("%d\t%s" % (Y, X))
# test.py
# coding=utf-8
from collections import defaultdict
import sys, re, train
#重み
Weight = defaultdict(lambda: 0)
with open("ans", "r") as modelFile:
for line in modelFile:
key, value = line.split()
Weight[key] = float(value)
# reg_model = re.compile("^(-?1)(.*)$")
with open(sys.argv[1], "r") as inputFile:
for line in inputFile:
X = line.strip()
phi = train.create_features(X)
Y = train.predict_one(Weight, phi)
print ("%d\t%s" % (Y, X))
