读取TFRecord数据集时如何规范化多个y值

Question

我生成了训练数据，都是原始值，我想在读取TFRecord数据集时规范化数据。第一步，我从训练数据中读取所有原始值，循环通过每个iter来连接数组，使用minmaxscaler来拟合训练数据并得到scaler

第二步，我编写一个规范化函数来转换所有数据，代码如下：

def normalizeY(inputs, targets):
    targets['Y1'] = scaler_Y1.transform(tf.cast(targets['Y1'], tf.float32))
    targets['Y2'] = scaler_Y2.transform(tf.cast(targets['Y2'], tf.float32))
    targets['Y3'] = scaler_Y3.transform(tf.cast(targets['Y3'], tf.float32))
        
    return inputs, targets
    
dataset = tf.data.TFRecordDataset(filenames)
dataset = dataset.map(_parse_function, num_parallel_calls=AUTOTUNE)
dataset = dataset.map(normalizeY)
dataset = dataset.prefetch(AUTOTUNE)
dataset = dataset.batch(batch_size)

但是步骤dataset.map(normalizeY)给出了错误：TypeError: __array__() takes 1 positional argument but 2 were given

Answer 1

我想出了一个解决方案，用最小/最大值来规范化，而不是用定标器来变换

def normalizeY(inputs, targets):
   targets['Y1'] = (tf.cast(targets['Y1'], tf.float32) - Y1_min) / (Y1_max - Y1_min)
   targets['Y2'] = (tf.cast(targets['Y2'], tf.float32) - Y2_min) / (Y2_max - Y2_min)
   targets['Y3'] = (tf.cast(targets['Y3'], tf.float32) - Y3_min) / (Y3_max - Y3_min)