def PreprocessKernel(name):
img = DataHelper.ReadImage(name, gray=True)
grids = SVCClassifier.SVCPreprocess(img)
labels = np.array(
BoardHelper.FENtoOneHot(
DataHelper.GetCleanNameByPath(name))).argmax(axis=1)
return grids, labels
def func_generator(train_file_names):
for image_file_name in train_file_names:
img = DataHelper.ReadImage(image_file_name)
x = CNNClassifier.PreprocessImage(img)
y = np.array(
BoardHelper.FENtoOneHot(
DataHelper.GetCleanNameByPath(image_file_name)))
yield x, y
def TestAccuracy(self, test_file_names):
num_files = len(test_file_names)
predict_result = self.__model__.predict(
CNNClassifier.func_generator(test_file_names)).argmax(axis=1)
predict_result = predict_result.reshape(num_files, -1)
predicted_fen_arr = np.array([
BoardHelper.LtoFEN(BoardHelper.LabelArrayToL(labels))
for labels in predict_result
])
test_fens = np.array([
DataHelper.GetCleanNameByPath(file_name)
for file_name in test_file_names
])
final_accuracy = (predicted_fen_arr == test_fens).astype(
np.float).mean()
return final_accuracy
test_SVC = False
SVC_load_model = True
if test_ABC:
abc = Classifiers.ABClassifier()
train_names = DataHelper.GetFileNamesInDir(g_train_dir)
if ABC_load_model:
print("abc: loading model from " + abc_model_file)
abc.LoadModel(abc_model_file)
else:
abc.Train(train_names)
y_truth = BoardHelper.FENtoL(DataHelper.GetCleanNameByPath(a_random_file))
img = DataHelper.ReadImage(a_random_file, gray = True)
pred = abc.Predict(img)
print("truth: ", ''.join(y_truth))
print("pred : ", ''.join(pred))
# save model
if not ABC_load_model:
print("abc: saving model to " + abc_model_file)
abc.SaveModel(abc_model_file)
if test_SVC:
svc = Classifiers.SVCClassifier()
train_names = DataHelper.GetFileNamesInDir(g_train_dir)
