Can we support Tensorflow 1.15?

[wangty6]

Currently it will give [module 'tensorflow.python.keras.api._v1.keras.backend' has no attribute 'is_keras_tensor'] error, and I have to replace is_keras_tensor to False.

[arogozhnikov]

TF < 2 was supported in earlier versions (0.2, maybe 0.3)

[wangty6]

Thanks for you reply, I wonder if there's a way we can let TF<2 benefit from latest update?

[arogozhnikov]

TF < 2 support is over. If you want to use newer features, you'll need to upgrade to TF2.

[wangty6]

Our distributed training infra strongly bind to TF 1.15, I think it cannot upgrade in a short period. Anyway, thank you for your quick reply. We will use earlier versions of einops or manually hack currently version.