def main():
# Prepare dataset(imgs, labels)
preprocess = Preprocessor('../dataset_test', batch_size=2, labels=['raccoon', 'test'])
num_classes = len(preprocess.labels)
imgs, labels = next(preprocess())
print(imgs.shape)
print(labels[0].shape)
print(labels[1].shape)
print(labels[2].shape)
# Training
yolov3 = YoloV3(input_shape=(416, 416, 3), num_classes=num_classes, training=True)
outputs = yolov3(imgs)
loss_func1 = YoloLoss(anchors_wh_mask[0], num_classes)
loss_func2 = YoloLoss(anchors_wh_mask[1], num_classes)
loss_func3 = YoloLoss(anchors_wh_mask[2], num_classes)
loss1, loss_breakdown1 = loss_func1(labels[0], outputs[0])
loss2, loss_breakdown2 = loss_func2(labels[1], outputs[1])
loss3, loss_breakdown3 = loss_func3(labels[2], outputs[2])
print(loss1)
print(loss2)
print(loss3)
def main():
data_dir = './dataset_test'
labels = ['raccoon']
num_classes = len(labels)
ckpt_dir = './checkpoints'
lr_rate = 0.0001
if not os.path.exists(ckpt_dir):
os.makedirs(ckpt_dir)
preprocessor = Preprocessor(data_dir=data_dir,
output_shape=(416, 416),
labels=labels,
batch_size=BATCH_SIZE)
model = YoloV3(input_shape=(416, 416, 3),
num_classes=num_classes,
training=True)
loss_objects = [
YoloLoss(valid_anchors_wh, num_classes)
for valid_anchors_wh in anchors_wh_mask
]
optimizer = tf.keras.optimizers.Adam(lr=lr_rate)
lowest_loss = 10000000
for epoch in range(EPOCH):
print('{} epoch start! : {}'.format(
epoch,
datetime.datetime.now().strftime("%Y.%m.%d %H:%M:%S")))
epoch_loss = train_one_epoch(model, loss_objects, preprocessor(),
optimizer)
if lowest_loss > epoch_loss:
lowest_loss = epoch_loss
save_path = ckpt_dir + '/cp-{:04d}-{:.4f}.ckpt'.format(
epoch, lowest_loss)
model.save_weights(save_path)
print('Save CKPT _ [loss : {:.4f}, save_path : {}]\n'.format(
lowest_loss, save_path))
import pandas as pd
import numpy as np
import sys
from sklearn.externals import joblib
from utils.preprocess import feature_expand_path
from utils.preprocess import Preprocessor
if __name__ == "__main__":
preprocessor = Preprocessor()
# 扩展特征
if len(sys.argv) == 3:
feature_expand_path(sys.argv[1], sys.argv[2])
else:
print('请指定需要预测的文件的路径以及特征扩展文件的路径')
test = pd.read_csv(sys.argv[2])
test_for_pre = test[test['pay_price'] > 0]
classify_test = test_for_pre.iloc[:, 1:]
classify_test = preprocessor.time_spliter(classify_test)
model = joblib.load('model_save/xgb_clf.model')
model_reg = joblib.load('model_save/xgb_reg.model')
# 准备测试集
classify_test = test_for_pre.iloc[:, 1:]
