This is the last article on the Tensorflow 2.0 Advent Calendar. (Large late) Tensorflow 2.0 has been released! I thought that Tensorflow 2.1 had already arrived, so Write the changes and evolved points here based on the Release Note.
From Tensorflow2.1, it is no longer necessary to put tensorflow-gpu on purpose.
Just pip install tensorflow will make tensorflow GPU support!
Now that there is no confusion, it gives a smarter impression.
GCP has a feature called Cloud TPU. Colabratory also has a TPU item for the time being.
Experimental, but with keras APIs such as keras.fit, keras.predict
Cloud TPU can now be used.
In the old days, I had to do various things before keras.fit
In the first place, it can not be used in Colabratory ([Information here](https://stackoverflow.com/questions/55541881/how-to-convert-tf-keras-model-to-tpu-using-tensorflow-2-0- in-google-colab)))
It was troublesome in various ways. I hope this will be improved and become TPU friendly!
Ragged Tensor can finally be handled by Dataset, You can now do the following:
Ragged_Tensor
rt_x = tf.ragged.constant([[1,2,3],[],[1,2]])
x_data = tf.data.Dataset.from_tensor_slices(rt_x)
for x in x_data:
print(x)
result
tf.Tensor([1 2 3], shape=(3,), dtype=int32)
tf.Tensor([], shape=(0,), dtype=int32)
tf.Tensor([1 2], shape=(2,), dtype=int32)
This may increase the number of situations where you can make use of various things such as natural language processing and Ragged Tensor.
TensorFlow 2.1 will be the last TF release supporting Python 2. Python 2 support officially ends an January 1, 2020. As announced earlier, TensorFlow will also stop supporting Python 2 starting January 1, 2020, and no more releases are expected in 2019.
With that said, support for Python 2 will end in Tensorflow 2.1. Well, I've lived so long that Python2 has a long life. Even though official support has increased, it was the last in 2019, so I want to move to Python 3 soon. (Before that, I'd like Google to do Python 3 nativeization of the Gcloud SDK ...)
This time, I have only raised the main topic. If there is a Release in the future, I will post it in the article in this way, so thank you. (I really wanted to write more details, but I'm sorry for my lack of knowledge ...)