python中 np.repeat()

实现Python时，有的时候同一个功能，有的代码写得很难看，有的写得又很优雅。
列举工作中遇到的codesnap:

API 的方法的选择

• numpy.astype 实现类型转换
• np.minimum(np.maximum(0), 1) 实现最大最小值截断，
  而不是通过 data[data < 0] = 0 与 data[data > 255] = 255
• np.repeat() 方法 numpy 维度的扩展

代码示例

不优雅的代码：

    begin_time = time.time()

    # 将 HU值 通过 WWWL 变成 uint8 类型
    gray_image_data = (hu_image_data- (self.window_level - self.window_width / 2)) / self.window_width * 255
    gray_image_data [gray_image_data < 0] = 0
    gray_image_data [gray_image_data > 255] = 255
    gray_image_data_uint8 = gray_image_data.astype(np.uint8)
    print(r'WWWL  cost time {}'.format(time.time() - begin_time))

    # 将numpy 的shape 从 [width, height] 变成 shape [width*height, 3]
    gray_image_data_flattened = gray_image_data_uint8 .reshape(-1)
    gray_image_data_flattened = gray_image_data_flattened [:, np.newaxis]
    image_data_rgb = np.zeros(([gray_image_data_flattened .shape[0], 3]))
    image_data_rgb [:, 0] = gray_image_data_flattened [:, 0]
    image_data_rgb [:, 1] = gray_image_data_flattened [:, 0]
    image_data_rgb [:, 2] = gray_image_data_flattened [:, 0]

    print(r'convert cost time {}'.format(time.time() - begin_time))

优雅的代码：

    # 将 HU值 通过 WWWL 变成 uint8 类型
    gray_image_data = (hu_image_data- (self.window_level - self.window_width / 2)) / self.window_width
    gray_image_data = np.minimum(np.maximum(gray_image_data, 0), 1) * 255
    gray_image_data_uint8 = gray_image_data .astype(np.uint8)
    print(r'WWWL  cost time {}'.format(time.time() - begin_time))

    # 将numpy 的shape 从 [width, height] 变成 shape [width*height, 3]
    gray_image_data_flattened = gray_image_data_uint8 .reshape(-1)
    mage_data_rgb = gray_image_data_flattened [:, None].repeat(3, axis=1)
    print(r'convert cost time {}'.format(time.time() - begin_time))

可以看到，优雅的代码更简洁，同时减少拷贝。