No module named 'tensorflow.python.keras.mixed_precision.experimental'

[jorgelahoz]

I'm suffering a strange behaviour when importing the gradients functions:

Set up
pip install see-rnn

test.py

from see_rnn import features_0D, features_1D, features_2D
from see_rnn import get_gradients
from see_rnn import get_layer_gradients

exit()

Error

Exception has occurred: ImportError
cannot import name 'get_gradients' from 'see_rnn' (/Users/.../opt/anaconda3/envs/fns/lib/python3.8/site-packages/see_rnn/init.py)

Sames happen with "get_layer_gradients".

[OverLordGoldDragon]

Most likely you don't have TensorFlow installed; I've made the error message more informative in the latest version. Try via

pip install git+https://github.com/OverLordGoldDragon/see-rnn

and paste the error here if it doesn't mention lack of TensorFlow install.

[jorgelahoz]

Same error, here is the output:

pip install git+https://github.com/OverLordGoldDragon/see-rnn
Collecting git+https://github.com/OverLordGoldDragon/see-rnn
  Cloning https://github.com/OverLordGoldDragon/see-rnn to /private/var/folders/vq/yhjbmwgj379bkvvj9csfpbx00000gn/T/pip-req-build-2_6s7aio
  Running command git clone -q https://github.com/OverLordGoldDragon/see-rnn /private/var/folders/vq/yhjbmwgj379bkvvj9csfpbx00000gn/T/pip-req-build-2_6s7aio
Requirement already satisfied: numpy in ./.local/lib/python3.8/site-packages (from see-rnn==1.15.2) (1.19.5)
Requirement already satisfied: matplotlib in ./.local/lib/python3.8/site-packages (from see-rnn==1.15.2) (3.3.4)
Requirement already satisfied: cycler>=0.10 in ./.local/lib/python3.8/site-packages (from matplotlib->see-rnn==1.15.2) (0.10.0)
Requirement already satisfied: python-dateutil>=2.1 in ./.local/lib/python3.8/site-packages (from matplotlib->see-rnn==1.15.2) (2.8.1)
Requirement already satisfied: kiwisolver>=1.0.1 in ./.local/lib/python3.8/site-packages (from matplotlib->see-rnn==1.15.2) (1.3.1)
Requirement already satisfied: pillow>=6.2.0 in ./.local/lib/python3.8/site-packages (from matplotlib->see-rnn==1.15.2) (8.1.2)
Requirement already satisfied: pyparsing!=2.0.4,!=2.1.2,!=2.1.6,>=2.0.3 in ./.local/lib/python3.8/site-packages (from matplotlib->see-rnn==1.15.2) (2.4.7)
Requirement already satisfied: six in ./.local/lib/python3.8/site-packages (from cycler>=0.10->matplotlib->see-rnn==1.15.2) (1.15.0)
Building wheels for collected packages: see-rnn
  Building wheel for see-rnn (setup.py) ... done
  Created wheel for see-rnn: filename=see_rnn-1.15.2-py3-none-any.whl size=36164 sha256=e8392951e9be5ec11505ffde1b3c33dedec4fcf633ba3405852d7bc5db0eb72e
  Stored in directory: /private/var/folders/vq/yhjbmwgj379bkvvj9csfpbx00000gn/T/pip-ephem-wheel-cache-qx3ny0uz/wheels/ea/6d/12/91ad126da05e06f67063d34baabf0ab1d149769df23f0affd4
Successfully built see-rnn
Installing collected packages: see-rnn
  Attempting uninstall: see-rnn
    Found existing installation: see-rnn 1.15.1
    Uninstalling see-rnn-1.15.1:
      Successfully uninstalled see-rnn-1.15.1
Successfully installed see-rnn-1.15.2

And packages already installed:

pip list | grep tensor
tensorboard              2.4.1
tensorboard-plugin-wit   1.8.0
tensorflow               2.4.1
tensorflow-estimator     2.4.0

pip list | grep Keras
Keras                    2.4.3
Keras-Preprocessing      1.1.2

[OverLordGoldDragon]

It cannot be the same error; try pip uninstall see-rnn then pip install git+https://github.com/OverLordGoldDragon/see-rnn.

[jorgelahoz]

After uninstalling and installing:

Traceback (most recent call last):
File "tests/test_jl.py", line 13, in
from see_rnn import get_gradients
ImportError: cannot import name 'get_gradients' from 'see_rnn' (/Users/.../opt/anaconda3/envs/fns/lib/python3.8/site-packages/see_rnn/init.py)

[OverLordGoldDragon]

Unsure what the problem is, need debug info:

1. In __init__.py (import see_rnn; print(see_rnn.__file__)), comment lines 26, 33-35 (try, except, pass), and unindent 27-32. Save file. 1a) restart kernel, then run from see_rnn import get_gradients, paste output here.
2. On first line of __init__.py (above import os), paste from .inspect_gen import get_gradients; repeat 1a.

[jorgelahoz]

Changing init.py
New code:

...
from . import visuals_gen
from .visuals_gen import *
#try:
from . import visuals_rnn
from .visuals_rnn import *
from . import inspect_gen
from .inspect_gen import *
from . import inspect_rnn
from .inspect_rnn import *
#except:
    # handled in _backend.py
#    pass
...

Output:

No module named 'tensorflow.python.keras.mixed_precision.experimental'

Changing first line of init.py
Same output:

No module named 'tensorflow.python.keras.mixed_precision.experimental'

[jorgelahoz]

Could this be linked somehow with the issue?

https://www.tensorflow.org/api_docs/python/tf/keras/mixed_precision/experimental/LossScaleOptimizer

Warning: This class is deprecated and will be removed in TensorFlow 2.5. Please use the non-experimental class tf.keras.mixed_precision.LossScaleOptimizer instead.

[OverLordGoldDragon]

Possibly, yes. The last tested TF version of see_rnn is 2.3.0, and I am no longer maintaining it for future compatibility. Version bugfix PRs are welcome.

Thanks for the report, but I'm afraid the only current solution is downgrading TensorFlow.

[Juan-estudiante-uam]

Seems to be working up to version 2.9.0 so far, 10 and 11 throw the exception..

[Siddp278]

Hey, so what I did was I went to the venv\Lib\site-packages\deepasr\model\deepasrnetwork1.py and venv\lib\site-packages\deepasr\model\deepspeech2.py and change the the line "from tensorflow.keras.mixed_precision import experimental as mixed_precision" to "from tensorflow.keras.mixed_precision import *"
Also remove the mixed_precision calls from the script in model/deepasrnetwork1.py and model/deepspeech2.py