Tensorflow 2.x

tensorflow2

	tf	keras	python	依赖
	TF2.0	Keras2.3.1	3.6,3.7
	TF2.1	Keras2.3.1	3.6,3.7
	TF2.2	Keras2.3.1	3.6-3.8
	TF2.3	Keras2.4.3	3.6-3.8	h5py=2.10.0 hdf5=1.10.4 numpy=1.19.2
	TF2.4	Keras2.4.3	3.6-3.8
	TF2.5		3.6-3.9
	TF2.6	Keras2.6	3.6-3.9

ENV

CUDA11.1＋cuDNN8.0.5＋tensorflow-gpu2.4.1

CUDA11.1＋cuDNN8.0.5＋tensorflow-gpu2.4.1

CUDA11.1＋cuDNN8.0.5＋tensorflow-gpu2.4.1

load

tf.saved_model.load(
    export_dir, tags=None, options=None
)

Loading Keras models

model = tf.keras.Model(...)
tf.saved_model.save(model, path)
imported = tf.saved_model.load(path)
outputs = imported(inputs)

tf.gather tf.gather_nd

tf.gather和tf.gather_nd都是从tensor中取出index标注的部分，不同之处在于，gather一般只使用一个index来标注，而gather_nd可以使用多个index。

import tensorflow.compat.v1 as tf
tf.disable_eager_execution()
# import tensorflow as tf
params = tf.constant([["a", "b"], ["c", "d"], ["e", "f"]])
gather = tf.constant([0, 2])
gather_nd = tf.constant([[0,0], [1, 1]])


with tf.Session() as sess:
    print(tf.gather(params, gather).eval())
    print()
    print(tf.gather_nd(params, gather_nd).eval())

[[b'a' b'b']
 [b'e' b'f']]

[b'a' b'd']

Keras

tf.keras.layers.Dense()
tf.keras.layers.LSTM()

LSTM

https://www.tensorflow.org/versions/r2.3/api_docs/python/tf/keras/layers/LSTM

inputs = tf.random.normal([32, 10, 8])
lstm = tf.keras.layers.LSTM(4)
output = lstm(inputs)
print(output.shape)  #(32,4)

lstm = tf.keras.layers.LSTM(4, return_sequences=True, return_state=True)
whole_seq_output, final_memory_state, final_carry_state = lstm(inputs)
print(whole_seq_output.shape)#(32,10,4)

print(final_memory_state.shape) #(32,4)

print(final_carry_state.shape) #(32,4)

BUGS

解决报错“ImportError: cannot import name 'get_config' from 'tensorflow.python.eager.context'”
安装的tensorflow版本和keras版本可能不合适的时候，会出现报错

module 'tensorflow_core.compat.v2' has no attribute '__internal__'

from tensorflow import keras
from keras import layers
-->
from tensorflow import keras
from tensorflow.keras import layers

Cannot convert a symbolic Tensor (lstm/strided_slice:0) to a numpy array.

conda install -c anaconda h5py

LSTM+Dense报错

NotImplementedError: Cannot convert a symbolic Tensor 
(lstm/strided_slice:0) to a numpy array. This error 
may indicate that you're trying to pass a Tensor to a NumPy call, 
which is not supported

解决：安装 numpy==1.19.2，太高的numpy版本不行