keras - 如果validation_data ValueError 在model.fit() 中引发错误：具有多个元素的数组的真值不明确。使用 a.any() 或 a.all()

Question

我正在尝试运行一个简单的自动编码器模型。我正在从包含单词嵌入的 csv 中读取训练数据。我有这段代码，但标题中的错误是在model.fit() 函数中引发的，并与我的validation data. 我尝试了很多东西，但错误仍然存​​在。我是 NLP 的新手，也许我的逻辑完全错误我不知道。因此，如果有人可以提供帮助，我将不胜感激。这是我的代码：

def train_predict(df):
X_train, X_validation = train_test_split(df, test_size=0.3, random_state=42, shuffle=True)
X = X_train.iloc[:, :-1].to_numpy()           #shape is (1880,220) in here
X = tf.expand_dims(X, axis=-1)                #shape is (1880,220,1)
X_val = X_validation.iloc[:,:-1].to_numpy()   #shape is (300,220)
X_val= tf.expand_dims(X_val, axis=-1)         #shape is (300,220,1)

inputs, decoder_output, visualization = autoEncoder(X)

model = Model(inputs=inputs, outputs=decoder_output)
encoder_model = Model(inputs=inputs, outputs=visualization)

batch_size = 128
train_steps = len(X) // batch_size
val_steps = len(X_val) // batch_size
model.summary()
model.compile(optimizer='adam', metrics=['accuracy'], loss='mean_squared_error')
model.fit(X, steps_per_epoch=train_steps, validation_data=X_val, validation_steps=val_steps,epochs=100) 
result = model.evaluate(X_val, steps=10)

我的自动编码器功能代码的详细信息如下：

def autoEncoder(X_train):
inputs = tf.keras.layers.Input(shape=(X_train.shape[1],1))
# parameters
conv_1 = Conv1D(filters=64, kernel_size=3, activation='relu', padding='same')(inputs)
max_pool_1 = MaxPool1D(pool_size=2)(conv_1)

conv_2 = Conv1D(filters=128, kernel_size=3, activation='relu', padding='same')(max_pool_1)
max_pool_2 = MaxPool1D(pool_size=2)(conv_2)

# BOTTLE NECK

bottle_neck = Conv1D(filters=256, kernel_size=3, activation='relu', padding='same')(max_pool_2)
visualization = Conv1D(filters=1, kernel_size=3, activation='sigmoid', padding='same')(bottle_neck)

# DECODER
conv_3 = Conv1D(filters=128, kernel_size=3, activation='relu', padding='same')(bottle_neck)
upsample_1 = UpSampling1D(size=2)(conv_3)

conv_4 = Conv1D(filters=64, kernel_size=3, activation='relu', padding='same')(upsample_1)
upsample_2 = UpSampling1D(size=2)(conv_4)

decoder_output = Conv1D(filters=1, kernel_size=3, activation='sigmoid', padding='same')(upsample_2)

return inputs, decoder_output, visualization

标签： kerastensorflownlpnumpypython-3.x

Answer 1

如果您可以复制粘贴代码产生的整个错误堆栈，那就太好了，每个人都应该遵循与错误相关的问题，因为这使调试变得更加容易。

这是使用虚拟数据集重现相同错误的尝试：

import numpy as np
import tensorflow as tf

np.random.seed(11)
np.set_printoptions(precision=2)

def autoEncoder(X_train):
    inputs = tf.keras.layers.Input(shape=(X_train.shape[1], 1))
    conv_1 = tf.keras.layers.Conv1D(filters=64, kernel_size=3, activation='relu', padding='same')(inputs)
    max_pool_1 = tf.keras.layers.MaxPool1D(pool_size=2)(conv_1)

    conv_2 = tf.keras.layers.Conv1D(filters=128, kernel_size=3, activation='relu', padding='same')(max_pool_1)
    max_pool_2 = tf.keras.layers.MaxPool1D(pool_size=2)(conv_2)

    bottle_neck = tf.keras.layers.Conv1D(filters=256, kernel_size=3, activation='relu', padding='same')(max_pool_2)
    visualization = tf.keras.layers.Conv1D(filters=1, kernel_size=3, activation='sigmoid', padding='same')(bottle_neck)

    conv_3 = tf.keras.layers.Conv1D(filters=128, kernel_size=3, activation='relu', padding='same')(bottle_neck)
    upsample_1 = tf.keras.layers.UpSampling1D(size=2)(conv_3)

    conv_4 = tf.keras.layers.Conv1D(filters=64, kernel_size=3, activation='relu', padding='same')(upsample_1)
    upsample_2 = tf.keras.layers.UpSampling1D(size=2)(conv_4)

    decoder_output = tf.keras.layers.Conv1D(filters=1, kernel_size=3, activation='sigmoid', padding='same')(upsample_2)

    return inputs, decoder_output, visualization

X = np.random.randn(1880, 220)
X_val = np.random.randn(300, 220)

X = np.expand_dims(X, axis=-1)
X = tf.convert_to_tensor(X)   # (1880, 220, 1)
X_val = np.expand_dims(X_val, axis=-1)
X_val = tf.convert_to_tensor(X_val)  # (300, 220, 1)

inputs, decoder_output, visualization = autoEncoder(X)
model = tf.keras.Model(inputs=inputs, outputs=decoder_output)
encoder_model = tf.keras.Model(inputs=inputs, outputs=visualization)

batch_size = 128
train_steps = len(X) // batch_size
val_steps = len(X_val) // batch_size
model.compile(optimizer='adam', metrics=['accuracy'], loss='mean_squared_error')
model.fit(X, steps_per_epoch=train_steps, validation_data = X_val, validation_steps=val_steps, epochs=100)

在 google-colab 上，这给出了以下错误：

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-29-a889c5a46f35> in <module>()
      3 val_steps = len(X_val) // batch_size
      4 model.compile(optimizer='adam', metrics=['accuracy'], loss='mean_squared_error')
----> 5 model.fit(X, steps_per_epoch=train_steps, validation_data = X_val, validation_steps=val_steps, epochs=100)

1 frames
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/engine/training.py in fit(self, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, validation_batch_size, validation_freq, max_queue_size, workers, use_multiprocessing)
   1041               (x, y, sample_weight), validation_split=validation_split))
   1042 
-> 1043     if validation_data:
   1044       val_x, val_y, val_sample_weight = (
   1045           data_adapter.unpack_x_y_sample_weight(validation_data))

/usr/local/lib/python3.7/dist-packages/tensorflow/python/framework/ops.py in __bool__(self)
    990 
    991   def __bool__(self):
--> 992     return bool(self._numpy())
    993 
    994   __nonzero__ = __bool__

ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()

这与您的 OP 相同。最好发布错误堆栈的原因是因为答案隐藏在这些行中，特别是：

1043     if validation_data:
1044       val_x, val_y, val_sample_weight = (
1045           data_adapter.unpack_x_y_sample_weight(validation_data))

的格式validation_data与(x, y, sample_weight). 这是fit 方法文档必须说的：

validation_data将覆盖validation_split. validation_data可能是： - (x_val, y_val)Numpy 数组或张量的元组 - (x_val, y_val, val_sample_weights)Numpy 数组的元组 - 数据集 对于前两种情况，必须提供 batch_size。对于最后一种情况，validation_steps可以提供。

我想你现在明白为什么你会出错，Y你的自动编码器没有。这不应该有任何问题，因为你X本身就是你的Y. 这是编码器教程中的一句话，可以在这种情况下帮助我们：

x_train使用作为输入和目标来训练模型。将encoder学习将数据集从 784 维压缩到潜在空间，decoder并将学习重建原始图像。

因此，您应该做的是编写以下内容：

model.fit(X, X, steps_per_epoch=train_steps, validation_data=(X_val, X_val), validation_steps=val_steps, epochs=100)

这确实开始了培训！