tensorflow==2.13 version incompatible in sleap[pypi]==1.3.2

[roomrys]

Discussed in #1483

^{Originally posted by Jaspermoray September 11, 2023}
Hi!

I'm using colab to do all of sleap's functions, and recently everything broke, I assume because of recent updates. At the moment, I'm getting a tensorflow issue:

2023-09-11 15:16:22.581609: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT
Traceback (most recent call last):
  File "/usr/local/bin/sleap-train", line 5, in <module>
    from sleap.nn.training import main
  File "/usr/local/lib/python3.10/dist-packages/sleap/__init__.py", line 14, in <module>
    import sleap.nn
  File "/usr/local/lib/python3.10/dist-packages/sleap/nn/__init__.py", line 1, in <module>
    import sleap.nn.architectures
  File "/usr/local/lib/python3.10/dist-packages/sleap/nn/architectures/__init__.py", line 7, in <module>
    from sleap.nn.architectures import pretrained_encoders
  File "/usr/local/lib/python3.10/dist-packages/sleap/nn/architectures/pretrained_encoders.py", line 47, in <module>
    from sleap.nn.data.normalization import ensure_rgb
  File "/usr/local/lib/python3.10/dist-packages/sleap/nn/data/__init__.py", line 2, in <module>
    from sleap.nn.data import confidence_maps
  File "/usr/local/lib/python3.10/dist-packages/sleap/nn/data/confidence_maps.py", line 6, in <module>
    from sleap.nn.data.utils import make_grid_vectors
  File "/usr/local/lib/python3.10/dist-packages/sleap/nn/data/utils.py", line 4, in <module>
    from keras.utils import tf_utils
ImportError: cannot import name 'tf_utils' from 'keras.utils' (/usr/local/lib/python3.10/dist-packages/keras/utils/__init__.py)
INFO:numexpr.utils:NumExpr defaulting to 2 threads.
2023-09-11 15:16:27.665644: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
To enable the following instructions: AVX2 FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.
2023-09-11 15:16:29.125970: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT
Traceback (most recent call last):
  File "/usr/local/bin/sleap-train", line 5, in <module>
    from sleap.nn.training import main
  File "/usr/local/lib/python3.10/dist-packages/sleap/__init__.py", line 14, in <module>
    import sleap.nn
  File "/usr/local/lib/python3.10/dist-packages/sleap/nn/__init__.py", line 1, in <module>
    import sleap.nn.architectures
  File "/usr/local/lib/python3.10/dist-packages/sleap/nn/architectures/__init__.py", line 7, in <module>
    from sleap.nn.architectures import pretrained_encoders
  File "/usr/local/lib/python3.10/dist-packages/sleap/nn/architectures/pretrained_encoders.py", line 47, in <module>
    from sleap.nn.data.normalization import ensure_rgb
  File "/usr/local/lib/python3.10/dist-packages/sleap/nn/data/__init__.py", line 2, in <module>
    from sleap.nn.data import confidence_maps
  File "/usr/local/lib/python3.10/dist-packages/sleap/nn/data/confidence_maps.py", line 6, in <module>
    from sleap.nn.data.utils import make_grid_vectors
  File "/usr/local/lib/python3.10/dist-packages/sleap/nn/data/utils.py", line 4, in <module>
    from keras.utils import tf_utils
ImportError: cannot import name 'tf_utils' from 'keras.utils' (/usr/local/lib/python3.10/dist-packages/keras/utils/__init__.py)

A cursory google told me that its probably a incompatibility with tensorflow and keras, but I don't want to download and use versions that may not be compatible with sleap's other code. So can anybody help?

---

I'm looking into the issue. To keep from halting, change the pip install sleap[pypi] command to the previous release for the time being pip install sleap==1.3.1. I've also updated all the notebooks online to reflect this in e4fca4f.

---

Some changes between 1.3.2 and 1.3.1:

pip install sleap[pypi]	pip install sleap==1.3.1
keras==2.8.0	keras==2.8.0
Keras-Applications==1.0.8	Keras-Applications==1.0.8
Keras-Preprocessing==1.1.2	Keras-Preprocessing==1.1.2
tensorboard==2.13.0	tensorboard==2.8.0
tensorboard-data-server==0.7.1	tensorboard-data-server==0.6.1
---	tensorboard-plugin-wit==1.8.1
tensorflow==2.13.0	tensorflow==2.8.4
tensorflow-datasets==4.9.2	tensorflow-datasets==4.9.2
tensorflow-estimator==2.13.0	tensorflow-estimator==2.8.0
tensorflow-gcs-config==2.13.0	tensorflow-gcs-config==2.13.0
tensorflow-hub==0.14.0	tensorflow-hub==0.14.0
tensorflow-io-gcs-filesystem==0.33.0	tensorflow-io-gcs-filesystem==0.33.0
tensorflow-metadata==1.14.0	tensorflow-metadata==1.14.0
tensorflow-probability==0.20.1	tensorflow-probability==0.20.1
tensorstore==0.1.41	tensorstore==0.1.41

Doing a !pip install tensorflow==2.8.4 after the SLEAP installation fixes the problem... Wondering how the wheel test missed this as we definitely run a training test. Looking at the logs, the wheel test prefers tensorflow==2.11 whereas colab seems to come pre-installed with the latest tensorflow==2.13. We are adding some restrictions in #1485.

[roomrys]

Hi @Jaspermoray,

We released a new version last week which aimed to address your issue. Things should be back in order now!

Thanks,
Liezl