TensorFlow 2.6.x not compatible

[jmsaurel]

Works and tested on TensorFlow versions 2.3.x and 2.4.x.
TensorFlow version 2.6.x doesn't work.
Add precisions in the doc about that.

[matteoquintiliani]

I was interested in testing phaseworm but at the moment it seems that tensorflow 2.3 and 2.4 are no longer available.

PackagesNotFoundError: The following packages are not available from current channels:

   - tensorflow=2.4

Using tensorflow 2.5, the installation process completes successfully, but when I run it I get the following messages:

$ phaseworm -c ../ew_env/params/config_rt.cfg
Configuration file <../ew_env/params/config_rt.cfg> read
/opt/miniconda3/envs/phaseworm/lib/python3.8/site-packages/phasenet/model.py:138: UserWarning: `tf.layers.conv2d` is deprecated and will be removed in a future version. Please Use `tf.keras.layers.Conv2D` instead.
  net = tf.compat.v1.layers.conv2d(net,
/opt/miniconda3/envs/phaseworm/lib/python3.8/site-packages/phasenet/model.py:147: UserWarning: `tf.layers.batch_normalization` is deprecated and will be removed in a future version. Please use `tf.keras.layers.BatchNormalization` instead. In particular, `tf.control_dependencies(tf.GraphKeys.UPDATE_OPS)` should not be used (consult the `tf.keras.layers.BatchNormalization` documentation).
  net = tf.compat.v1.layers.batch_normalization(net,
/opt/miniconda3/envs/phaseworm/lib/python3.8/site-packages/phasenet/model.py:153: UserWarning: `tf.layers.dropout` is deprecated and will be removed in a future version. Please use `tf.keras.layers.Dropout` instead.
  net = tf.compat.v1.layers.dropout(net,
/opt/miniconda3/envs/phaseworm/lib/python3.8/site-packages/phasenet/model.py:163: UserWarning: `tf.layers.conv2d` is deprecated and will be removed in a future version. Please Use `tf.keras.layers.Conv2D` instead.
  net = tf.compat.v1.layers.conv2d(net,
/opt/miniconda3/envs/phaseworm/lib/python3.8/site-packages/phasenet/model.py:173: UserWarning: `tf.layers.batch_normalization` is deprecated and will be removed in a future version. Please use `tf.keras.layers.BatchNormalization` instead. In particular, `tf.control_dependencies(tf.GraphKeys.UPDATE_OPS)` should not be used (consult the `tf.keras.layers.BatchNormalization` documentation).
  net = tf.compat.v1.layers.batch_normalization(net,
/opt/miniconda3/envs/phaseworm/lib/python3.8/site-packages/phasenet/model.py:178: UserWarning: `tf.layers.dropout` is deprecated and will be removed in a future version. Please use `tf.keras.layers.Dropout` instead.
  net = tf.compat.v1.layers.dropout(net,
/opt/miniconda3/envs/phaseworm/lib/python3.8/site-packages/phasenet/model.py:186: UserWarning: `tf.layers.conv2d` is deprecated and will be removed in a future version. Please Use `tf.keras.layers.Conv2D` instead.
  net = tf.compat.v1.layers.conv2d(net,
/opt/miniconda3/envs/phaseworm/lib/python3.8/site-packages/phasenet/model.py:197: UserWarning: `tf.layers.batch_normalization` is deprecated and will be removed in a future version. Please use `tf.keras.layers.BatchNormalization` instead. In particular, `tf.control_dependencies(tf.GraphKeys.UPDATE_OPS)` should not be used (consult the `tf.keras.layers.BatchNormalization` documentation).
  net = tf.compat.v1.layers.batch_normalization(net,
/opt/miniconda3/envs/phaseworm/lib/python3.8/site-packages/phasenet/model.py:202: UserWarning: `tf.layers.dropout` is deprecated and will be removed in a future version. Please use `tf.keras.layers.Dropout` instead.
  net = tf.compat.v1.layers.dropout(net,
/opt/miniconda3/envs/phaseworm/lib/python3.8/site-packages/phasenet/model.py:212: UserWarning: `tf.layers.conv2d_transpose` is deprecated and will be removed in a future version. Please Use `tf.keras.layers.Conv2DTranspose` instead.
  net = tf.compat.v1.layers.conv2d_transpose(net,
/opt/miniconda3/envs/phaseworm/lib/python3.8/site-packages/phasenet/model.py:222: UserWarning: `tf.layers.batch_normalization` is deprecated and will be removed in a future version. Please use `tf.keras.layers.BatchNormalization` instead. In particular, `tf.control_dependencies(tf.GraphKeys.UPDATE_OPS)` should not be used (consult the `tf.keras.layers.BatchNormalization` documentation).
  net = tf.compat.v1.layers.batch_normalization(net,
/opt/miniconda3/envs/phaseworm/lib/python3.8/site-packages/phasenet/model.py:227: UserWarning: `tf.layers.dropout` is deprecated and will be removed in a future version. Please use `tf.keras.layers.Dropout` instead.
  net = tf.compat.v1.layers.dropout(net,
/opt/miniconda3/envs/phaseworm/lib/python3.8/site-packages/phasenet/model.py:237: UserWarning: `tf.layers.conv2d` is deprecated and will be removed in a future version. Please Use `tf.keras.layers.Conv2D` instead.
  net = tf.compat.v1.layers.conv2d(net,
/opt/miniconda3/envs/phaseworm/lib/python3.8/site-packages/phasenet/model.py:247: UserWarning: `tf.layers.batch_normalization` is deprecated and will be removed in a future version. Please use `tf.keras.layers.BatchNormalization` instead. In particular, `tf.control_dependencies(tf.GraphKeys.UPDATE_OPS)` should not be used (consult the `tf.keras.layers.BatchNormalization` documentation).
  net = tf.compat.v1.layers.batch_normalization(net,
/opt/miniconda3/envs/phaseworm/lib/python3.8/site-packages/phasenet/model.py:252: UserWarning: `tf.layers.dropout` is deprecated and will be removed in a future version. Please use `tf.keras.layers.Dropout` instead.
  net = tf.compat.v1.layers.dropout(net,
/opt/miniconda3/envs/phaseworm/lib/python3.8/site-packages/phasenet/model.py:260: UserWarning: `tf.layers.conv2d` is deprecated and will be removed in a future version. Please Use `tf.keras.layers.Conv2D` instead.
  net = tf.compat.v1.layers.conv2d(net,
Traceback (most recent call last):
  File "/opt/miniconda3/envs/phaseworm/bin/phaseworm", line 8, in <module>
    sys.exit(main())
  File "/opt/miniconda3/envs/phaseworm/lib/python3.8/site-packages/phaseworm_sources/phaseworm.py", line 493, in main
    run_loop()
  File "/opt/miniconda3/envs/phaseworm/lib/python3.8/site-packages/phaseworm_sources/phaseworm.py", line 436, in run_loop
    sess, model = init_pred(phasenet_config, conf.phasenet)
  File "/opt/miniconda3/envs/phaseworm/lib/python3.8/site-packages/phasenet/app.py", line 129, in init_pred
    saver.restore(sess, latest_check_point)
  File "/opt/miniconda3/envs/phaseworm/lib/python3.8/site-packages/tensorflow/python/training/saver.py", line 1406, in restore
    raise ValueError("Can't load save_path when it is None.")
ValueError: Can't load save_path when it is None.

Is anyone else facing this issue? Do you have any suggestions on how to proceed? I might be available to support development and debugging. Please let me know. Thanks.

[jmsaurel]

Hello @matteoquintiliani ,

Thanks for your interest and the report that TensorFlow 2.4 is no more available.
I just checked and saw that PhaseNet (https://github.com/AI4EPS/PhaseNet/) has been upgraded to TensorFlow 2.11 lately.

So I guess what we would need to do is to upgrade the PhaseNet code (in the directory Phasenet) to this newer version and hopefully it will work again.

[rlfesta]

"ValueError: Can't load save_path when it is None." , are you sure the directory to store picks exists and is correctly set in the configuration file ?
the others should be just warnings ;

tensorflow 2.3 is available if you install it with miniconda ,
install miniconda 3 , ( Miniconda3-py312_25.1.1-2-Linux-x86_64 )
activate miniconda with source [/installation_path/]miniconda3/activate
then follow the instruction for conda installation ,
it work well togheter with ew 7.1 on devuan daedalus ( debian 12 )