def prediction_all(test_dir=FilePath.validation_data_path[0],
train_file_dir=FilePath.data_path[0],
prediction_file_path=None,
config=None,
temp_path=None):
labels = get_all_labels(train_file_dir)
test_text = Dataset.read_original_data((test_dir, FilePath.data_path[1]),
mode="test")
Dataset.save_splitted_file(Dataset.read_original_data(
(train_file_dir, FilePath.data_path[1]), mode="test"),
Dataset.read_one_hot_label(
(train_file_dir, FilePath.label_path[1]),
mode="test"),
dir_path=train_file_dir)
if temp_path is None:
temp_path = root_dir()
for label in labels:
print(label)
train_data = Dataset.read_splitted_train_file(
os.path.join(train_file_dir, label + ".csv"))
cls = TransformerClassifier(**config).train(train_data["data"],
train_data["label"])
prob = cls.predict_prob(test_text["data"].values.tolist())
cls.save_prob_prediction_result(prob, label, temp_path)
label_prediction_result = cls.prob_convert_to_label(prob)
Dataset.save_label_prediction_result(label_prediction_result, label,
temp_path)
final = Dataset.merge_all_prediction_in_dir(temp_path, labels)
save_result_in_dir(final, prediction_file_path)
def classifier_ensemble(prediction_paths, output_path):
from functools import reduce
preds = [
Dataset.read_one_hot_label(label_path=(i, ),
mode="test").drop(columns=[''])
for i in prediction_paths
]
results = reduce(lambda a, b: a + b, preds)
results = results > 0
labels = reverse_transform(results)
pd.DataFrame(labels).to_csv(output_path,
header=False,
index=False,
encoding="utf-8")