如何在tf.keras中创建具有多个输出的回归模型？

Question

我正在尝试训练回归模型来预测音乐属性，例如BPM。该模型采用256x128px png文件的音频片段的频谱图，并输出几个连续值。到目前为止，我有以下代码，我是在tensorflow网站上基于this guide开发的：

import tensorflow as tf
import os
import random
import pathlib

AUTOTUNE = tf.data.experimental.AUTOTUNE
TRAINING_DATA_DIR = r'specgrams'

def gen_model():
    model = tf.keras.models.Sequential([
      tf.keras.layers.Flatten(input_shape=(256, 128, 3)),
      tf.keras.layers.Dense(256, activation='relu'),
      tf.keras.layers.Dense(2)
    ])

    model.compile(optimizer=tf.keras.optimizers.RMSprop(0.001),
                  loss='mse',
                  metrics=['mse', 'mae'])

    return model


def fetch_batch(batch_size=1000):
    all_image_paths = []
    all_image_labels = []

    data_root = pathlib.Path(TRAINING_DATA_DIR)
    files = data_root.iterdir()

    for file in files:
        file = str(file)
        all_image_paths.append(os.path.abspath(file))
        label = file[:-4].split('-')[2:]
        label = float(label[0]) / 200, int(label[1]) / 1000.0
        all_image_labels.append(label)

    def preprocess_image(path):
        img_raw = tf.io.read_file(path)
        image = tf.image.decode_png(img_raw, channels=3)
        image = tf.image.resize(image, [256, 128])
        image /= 255.0
        return image

    def preprocess(path, label):
        return preprocess_image(path), label

    path_ds = tf.data.Dataset.from_tensor_slices(all_image_paths)
    image_ds = path_ds.map(preprocess_image, num_parallel_calls=AUTOTUNE)
    label_ds = tf.data.Dataset.from_tensor_slices(all_image_labels)
    ds = tf.data.Dataset.zip((image_ds, label_ds))
    ds = ds.shuffle(buffer_size=len(os.listdir(TRAINING_DATA_DIR)))
    ds = ds.repeat()
    ds = ds.batch(batch_size)
    ds = ds.prefetch(buffer_size=AUTOTUNE)

    return ds

ds = fetch_batch()
model = gen_model()
model.fit(ds, epochs=1, steps_per_epoch=10)

但是我相信我的模型结构或者我如何预处理训练数据时犯了一个错误，因为我得到了关于错误维度的错误，但我正在努力缩小问题的确切位置。据我所知，我所遵循的指南是分类问题，而不是回归，而我的“标签”是一个2值的数组，这是导致问题的原因，但我不知道如何解决这个问题。

对于上下文，文件名的格式为xxx-xxx-A-B.png，其中A和B是模型的两个所需输出值。 A是70到180之间的浮点值，B是0到1000之间的整数值。因此，每个图像的label变量看起来像这样：(0.64, 0.319)。

这是我在尝试执行上述脚本时看到的错误：

Traceback (most recent call last):
  File "C:\Users\cainy\Desktop\BeatNet\training.py", line 60, in <module>
    model.fit(ds, epochs=1, steps_per_epoch=3)
  File "C:\Users\cainy\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow\python\keras\engine\training.py", line 791, in fit
    initial_epoch=initial_epoch)
  File "C:\Users\cainy\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow\python\keras\engine\training.py", line 1515, in fit_generator
    steps_name='steps_per_epoch')
  File "C:\Users\cainy\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow\python\keras\engine\training_generator.py", line 257, in model_iteration
    batch_outs = batch_function(*batch_data)
  File "C:\Users\cainy\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow\python\keras\engine\training.py", line 1259, in train_on_batch
    outputs = self._fit_function(ins)  # pylint: disable=not-callable
  File "C:\Users\cainy\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow\python\keras\backend.py", line 3217, in __call__
    outputs = self._graph_fn(*converted_inputs)
  File "C:\Users\cainy\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow\python\eager\function.py", line 558, in __call__
    return self._call_flat(args)
  File "C:\Users\cainy\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow\python\eager\function.py", line 627, in _call_flat
    outputs = self._inference_function.call(ctx, args)
  File "C:\Users\cainy\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow\python\eager\function.py", line 415, in call
    ctx=ctx)
  File "C:\Users\cainy\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow\python\eager\execute.py", line 66, in quick_execute
    six.raise_from(core._status_to_exception(e.code, message), None)
  File "<string>", line 3, in raise_from
tensorflow.python.framework.errors_impl.InvalidArgumentError: Can not squeeze dim[1], expected a dimension of 1, got 2
     [[{{node metrics/accuracy/Squeeze}}]] [Op:__inference_keras_scratch_graph_734]

编辑：我已将源代码上传到GitHub here。

Answer 1

您目前只有1个输出 - 长度为2的张量（每个批次元素）。如果你想使用/监控单独的损失，你需要在模型输出和标签中使用unstack。

我不确定models.Sequential是否合适，但您绝对可以使用功能API：

def gen_model():
    inputs = tf.keras.layers.Input(shape=(256, 128, 3), dtype=tf.float32)
    x = inputs
    x = tf.keras.layers.Dense(256, activation='relu')
    x = tf.keras.layers.Dense(2)
    a, b = tf.keras.layers.Lambda(tf.unstack, arguments=dict(axis=-1))(x)
    model = tf.keras.models.Model(inputs=inputs, outputs=[a, b])
    model.compile(optimizer=tf.keras.optimizers.RMSprop(0.001),
                  loss=['mse', 'mae'],
                  metrics=[['mse'], ['mae']])
    return model

在您的预处理中：

def preprocess(path, label):
    return preprocess_image(path), tf.unstack(label, axis=-1)