Tensorflow & Keras version

[sungkwang]

Hello, I would like to know what version of Tensorflow and Keras should be used. For Tensorflow, I tried version 1.15 and 2.2, but I've got the following errors.
<1.15>
ImportError: Keras requires TensorFlow 2.2 or higher. Install TensorFlow via pip install tensorflow
<2.2>
ImportError: cannot import name 'is_tensor' from 'keras.backend' (C:\Program_Files\anaconda3\envs\tf2-gpu\lib\site-packages\keras\backend.py)

I am using Windows 10 and anaconda. Tensorflow works fine. And, I installed sciann using pip as recommended here. Thank you.

[imsenthur]

Tested with Tensorflow 2.2.0 and Keras 2.4.3

The issue seems to be with 'sciann/utils/utilities.py', replace the following lines of code.

sciann/sciann/utils/utilities.py

Line 16 in e31b25d

from keras.backend import is_tensor

to from tensorflow import is_tensor

sciann/sciann/utils/utilities.py

Lines 22 to 25 in e31b25d

	from keras.initializers import random_uniform as default_bias_initializer
	from keras.initializers import glorot_normal as default_kernel_initializer
	from keras.initializers import Constant as default_constant_initializer
	from keras.regularizers import l1_l2

to,

from keras.initializers import RandomUniform as default_bias_initializer
from keras.initializers import GlorotNormal as default_kernel_initializer
from keras.initializers import Constant as default_constant_initializer
from keras.regularizers import l1_l2

It should be up and running.

[sungkwang]

I faced another problem. The error looks related to keras.

Traceback (most recent call last):
File "SciANN-BurgersEquation.py", line 62, in
u = sn.Functional('u', [t,x], 8*[20], 'tanh')
File "C:\Program_Files\anaconda3\envs\tf2-gpu\lib\site-packages\sciann\functionals\functional.py", line 145, in init
layer.name = "conct_" + layer.name.split("_")[-1]
File "C:\Program_Files\anaconda3\envs\tf2-gpu\lib\site-packages\tensorflow\python\keras\engine\base_layer.py", line 2554, in setattr
'different name.').format(name))
AttributeError: Can't set the attribute "name", likely because it conflicts with an existing read-only @Property of the object. Please choose a different name.

Using Tensorflow 2.2.0 and Keras 2.4.3.

[imsenthur]

It should be layer._name = 'foo', since layer.name is read-only, according to https://stackoverflow.com/questions/49550182/keras-rename-model-and-layers

So I modified '\sciann\functionals\functional.py' and '\sciann\utils\math.py' which seemed to fix it. The 1D curve-fitting example: https://github.com/sciann/sciann/blob/master/examples/example-fitting-1d.py worked as expected, but others did not. There are few problems during eager execution tensorflow/tensorflow#34944.

Maybe it doesn't support Tensorflow 2.2.0. I'll try it out with >=1.14.

[imsenthur]

Works on Tensorflow 1.14.0. Downgrade your Keras version to 2.3.0 with pip install --upgrade keras==2.3.0. You can also try out Tensorflow 1.15.0 with Keras 2.3.1. Kindly refer to https://docs.floydhub.com/guides/environments/.

[dimitri-voytan]

I was able to get around this issue using Tensorflow 1.15.2. I am using Google Colab notebooks. Tensorflow v2.3 is imported by default and I get the same problem as @sungkwang . These imports appear to work, but I can't verify that they aren't broken in a subtle way.

%tensorflow_version 1.x import tensorflow as tf !pip install sciann

[sciann]

please use TF <= 2.2.0 and Keras==2.3.1. A new release is coming soon that support the latest release of tensorflow.

[sciann]

Should be resolved now - get the latest v0.5.0.0 version.