使用 tf.data.Dataset.from_generator 時出錯

Question

我正在嘗試使用 tensorflow from_generator 制作 tensorflow 數據集，我很確定我已經制作了一個運行良好的 python 生成器，但是當我嘗試將其傳遞給 from_generator 時總是出錯。 這是我用來創建數據集的一段代碼

def dataset_generator(X, Y):
    for idx in range(X.shape[0]):
        img = X[idx, :, :, :]
        labels = Y[idx, :]
        yield img, labels

import tensorflow as tf
ds_generator = dataset_generator(X_data, Y_data)
ds = tf.data.Dataset.from_generator(ds_generator, output_signature=(tf.TensorSpec(shape=[None, 720, 720, 3], dtype=tf.int32), tf.TensorSpec(shape=[None, 30], dtype=tf.float16)))

但是當我運行它時，它總是會產生錯誤

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-63-af75191f4a28> in <module>
      1 import tensorflow as tf
      2 ds_generator = dataset_generator(X_data, Y_data)
----> 3 ds = tf.data.Dataset.from_generator(ds_generator, output_signature=(tf.TensorSpec(shape=[None, 720, 720, 3], dtype=tf.int32), tf.TensorSpec(shape=[None, 30], dtype=tf.float16)))

~/.local/lib/python3.6/site-packages/tensorflow/python/util/deprecation.py in new_func(*args, **kwargs)

~/.local/lib/python3.6/site-packages/tensorflow/python/data/ops/dataset_ops.py in from_generator(generator, output_types, output_shapes, args, output_signature)

TypeError: `generator` must be callable.

Answer 1

嗨，您的 gen function 的問題是您必須通過 args 命令來傳遞它，而不是像 function 這樣

import tensorflow as tf
import numpy as np

# Gen Function
def dataset_generator(X, Y):
    for idx in range(X.shape[0]):
        img = X[idx, :, :, :]
        labels = Y[idx, :]
        yield img, labels

# Created random data for testing
X_data = np.random.randn(100, 720, 720, 3).astype(np.float32)
Y_data = tf.one_hot(np.random.randint(0, 30, (100, )), 30)

# Testing function
ds = tf.data.Dataset.from_generator(
    dataset_generator,
    args=(X_data, Y_data), 
    output_types=(tf.float32, tf.uint8)
)

# Get output
next(iter(ds.batch(10).take(1)))