如何在tf.data.Dataset.map()中使用Keras的predict_on_batch? [英] How to use Keras' predict_on_batch in tf.data.Dataset.map()?

问题描述

我想找到一种在TF2.0.

假设我有一个numpy数据集

Let's say I have a numpy dataset

n_data = 10**5
my_data    = np.random.random((n_data,10,1))
my_targets = np.random.randint(0,2,(n_data,1))

data = ({'x_input':my_data}, {'target':my_targets})

和tf.keras模型

x_input = Input((None,1), name = 'x_input')
RNN     = SimpleRNN(100,  name = 'RNN')(x_input)
dense   = Dense(1, name = 'target')(RNN)

my_model = Model(inputs = [x_input], outputs = [dense])
my_model.compile(optimizer='SGD', loss = 'binary_crossentropy')

我可以使用

dataset = tf.data.Dataset.from_tensor_slices(data)
dataset = dataset.batch(10)
prediction_dataset = dataset.map(transform_predictions)

其中transform_predictions是用户定义的函数，可从predict_on_batch

where transform_predictions is a user defined function that gets the predictions from predict_on_batch

def transform_predictions(inputs, outputs):
    predictions = my_model.predict_on_batch(inputs)
    # predictions = do_transformations_here(predictions)
    return predictions

这给出了来自predict_on_batch的错误:

AttributeError: 'Tensor' object has no attribute 'numpy'

据我了解，predict_on_batch期望一个numpy数组，并且它正在从数据集中获取张量对象.

As far as I understand, predict_on_batch expects a numpy array, and it is getting a tensor object from the dataset.

似乎一个可能的解决方案是将predict_on_batch包装在`tf.py_function中，尽管我也无法使其正常工作.

It seems like one possible solution is to wrap predict_on_batch in a `tf.py_function, though I have not been able to get that working either.

有人知道该怎么做吗?

推荐答案

Dataset.map()返回没有numpy()方法的<class 'tensorflow.python.framework.ops.Tensor'>.

Dataset.map() returns <class 'tensorflow.python.framework.ops.Tensor'> which doesn't have numpy() method.

遍历数据集返回 <class 'tensorflow.python.framework.ops.EagerTensor'>具有numpy()方法.

Iterating over Dataset returns <class 'tensorflow.python.framework.ops.EagerTensor'> which has a numpy() method.

提供一个渴望的张量来预测()方法家族很好.

Feeding an eager tensor to predict() family of methods works fine.

您可以尝试这样的事情:

You could try something like this:

dataset = tf.data.Dataset.from_tensor_slices(data)
dataset = dataset.batch(10)

for x,y in dataset:
    predictions = my_model.predict_on_batch(x['x_input'])
    #or 
    predictions = my_model.predict_on_batch(x)

这篇关于如何在tf.data.Dataset.map()中使用Keras的predict_on_batch?的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！