make tensorflow-addons dependency optional

[bertsky]

In an HPC environment (where I want to calamari-train), with Power9 CPU (ppc64le architecture), I get

pip install tensorflow-addons
ERROR: Could not find a version that satisfies the requirement tensorflow-addons (from versions: none)
ERROR: No matching distribution found for tensorflow-addons

Judging by this issue I guess it will be quite painful to build this from source in the HPC environment (bazel etc).

Since this seems to be needed only for under certain conditions (when the MovingAverageOptimizer gets used), I wonder if this could be made optional here.

[bertsky]

Correction: in fact it was quite simple.

pip install https://github.com/tensorflow/addons/archive/refs/tags/v0.16.1.tar.gz

So the Github release artifacts seem to be enough.

[bertsky]

Hmm, perhaps this recipe does not work, after all. When TF is loaded, it immediately complains…

UserWarning: Tensorflow Addons supports using Python ops for all Tensorflow versions above or equal to 2.6.0 and strictly below 2.9.0 (nightly versions are not supported). 
 The versions of TensorFlow you are currently using is 2.4.1 and is not supported. 
Some things might work, some things might not.
If you were to encounter a bug, do not file an issue.
If you want to make sure you're using a tested and supported configuration, either change the TensorFlow version or the TensorFlow Addons's version. 
You can find the compatibility matrix in TensorFlow Addon's readme:
https://github.com/tensorflow/addons

Then later-on when starting Calamari training, TF fails with:

  File "/home/h1/rosa992c/my-kernel/powerai-kernel2/lib/python3.7/site-packages/calamari_ocr/scripts/train.py", line 29, in main
    return trainer.train()
  File "/home/h1/rosa992c/my-kernel/powerai-kernel2/lib/python3.7/site-packages/calamari_ocr/ocr/training/trainer.py", line 190, in train
    last_logs = super().train(callbacks=callbacks)
  File "/home/h1/rosa992c/my-kernel/powerai-kernel2/lib/python3.7/site-packages/tfaip/device/device_config.py", line 134, in wrapper
    return func(*args, **kwargs)
  File "/home/h1/rosa992c/my-kernel/powerai-kernel2/lib/python3.7/site-packages/tfaip/trainer/trainer.py", line 209, in train
    training_graph_only=self._training_graph_only,
  File "/home/h1/rosa992c/my-kernel/powerai-kernel2/lib/python3.7/site-packages/tfaip/scenario/scenariobase.py", line 627, in setup_training
    self._keras_train_model((real_inputs, wrapped_targets, real_meta))
  File "/sw/installed/TensorFlow/2.4.1-fosscuda-2019b-Python-3.7.4/lib/python3.7/site-packages/tensorflow/python/keras/engine/base_layer.py", line 952, in __call__
    input_list)
  File "/sw/installed/TensorFlow/2.4.1-fosscuda-2019b-Python-3.7.4/lib/python3.7/site-packages/tensorflow/python/keras/engine/base_layer.py", line 1091, in _functional_construction_call
    inputs, input_masks, args, kwargs)
  File "/sw/installed/TensorFlow/2.4.1-fosscuda-2019b-Python-3.7.4/lib/python3.7/site-packages/tensorflow/python/keras/engine/base_layer.py", line 822, in _keras_tensor_symbolic_call
    return self._infer_output_signature(inputs, args, kwargs, input_masks)
  File "/sw/installed/TensorFlow/2.4.1-fosscuda-2019b-Python-3.7.4/lib/python3.7/site-packages/tensorflow/python/keras/engine/base_layer.py", line 863, in _infer_output_signature
    outputs = call_fn(inputs, *args, **kwargs)
  File "/sw/installed/TensorFlow/2.4.1-fosscuda-2019b-Python-3.7.4/lib/python3.7/site-packages/tensorflow/python/autograph/impl/api.py", line 670, in wrapper
    raise e.ag_error_metadata.to_exception(e)
NotImplementedError: in user code:

    /home/h1/rosa992c/my-kernel/powerai-kernel2/lib/python3.7/site-packages/tfaip/model/graphbase.py:72 call  *
        outputs = self.graph.train(inputs, pre_proc_targets)
    /home/h1/rosa992c/my-kernel/powerai-kernel2/lib/python3.7/site-packages/tfaip/model/graphbase.py:272 train  *
        return self(wrap((inputs, targets)))
    /home/h1/rosa992c/my-kernel/powerai-kernel2/lib/python3.7/site-packages/tfaip/model/graphbase.py:267 call  *
        return self._graph(inputs_targets)
    /home/h1/rosa992c/my-kernel/powerai-kernel2/lib/python3.7/site-packages/tfaip/model/graphbase.py:312 call  *
        outputs = self.build_train_graph(inputs, targets, training=training)
    /home/h1/rosa992c/my-kernel/powerai-kernel2/lib/python3.7/site-packages/tfaip/model/graphbase.py:343 build_train_graph  *
        return self.build_graph(inputs, training)
    /home/h1/rosa992c/my-kernel/powerai-kernel2/lib/python3.7/site-packages/calamari_ocr/ocr/model/graph.py:83 build_graph  *
        last_layer_output = layer(last_layer_output)
    /home/h1/rosa992c/my-kernel/powerai-kernel2/lib/python3.7/site-packages/calamari_ocr/ocr/model/layers/layer.py:77 call  *
        return self._call(inputs, **kwargs)
    /home/h1/rosa992c/my-kernel/powerai-kernel2/lib/python3.7/site-packages/calamari_ocr/ocr/model/layers/bilstm.py:50 _call  *
        return self.lstm(inputs)
    /sw/installed/TensorFlow/2.4.1-fosscuda-2019b-Python-3.7.4/lib/python3.7/site-packages/tensorflow/python/keras/layers/wrappers.py:539 __call__  **
        return super(Bidirectional, self).__call__(inputs, **kwargs)
    /sw/installed/TensorFlow/2.4.1-fosscuda-2019b-Python-3.7.4/lib/python3.7/site-packages/tensorflow/python/keras/engine/base_layer.py:1012 __call__
        outputs = call_fn(inputs, *args, **kwargs)
    /sw/installed/TensorFlow/2.4.1-fosscuda-2019b-Python-3.7.4/lib/python3.7/site-packages/tensorflow/python/keras/layers/wrappers.py:653 call
        initial_state=forward_state, **kwargs)
    /sw/installed/TensorFlow/2.4.1-fosscuda-2019b-Python-3.7.4/lib/python3.7/site-packages/tensorflow/python/keras/layers/recurrent.py:660 __call__
        return super(RNN, self).__call__(inputs, **kwargs)
    /sw/installed/TensorFlow/2.4.1-fosscuda-2019b-Python-3.7.4/lib/python3.7/site-packages/tensorflow/python/keras/engine/base_layer.py:1012 __call__
        outputs = call_fn(inputs, *args, **kwargs)
    /sw/installed/TensorFlow/2.4.1-fosscuda-2019b-Python-3.7.4/lib/python3.7/site-packages/tensorflow/python/keras/layers/recurrent_v2.py:1157 call
        inputs, initial_state, _ = self._process_inputs(inputs, initial_state, None)
    /sw/installed/TensorFlow/2.4.1-fosscuda-2019b-Python-3.7.4/lib/python3.7/site-packages/tensorflow/python/keras/layers/recurrent.py:859 _process_inputs
        initial_state = self.get_initial_state(inputs)
    /sw/installed/TensorFlow/2.4.1-fosscuda-2019b-Python-3.7.4/lib/python3.7/site-packages/tensorflow/python/keras/layers/recurrent.py:643 get_initial_state
        inputs=None, batch_size=batch_size, dtype=dtype)
    /sw/installed/TensorFlow/2.4.1-fosscuda-2019b-Python-3.7.4/lib/python3.7/site-packages/tensorflow/python/keras/layers/recurrent.py:2507 get_initial_state
        self, inputs, batch_size, dtype))
    /sw/installed/TensorFlow/2.4.1-fosscuda-2019b-Python-3.7.4/lib/python3.7/site-packages/tensorflow/python/keras/layers/recurrent.py:2987 _generate_zero_filled_state_for_cell
        return _generate_zero_filled_state(batch_size, cell.state_size, dtype)
    /sw/installed/TensorFlow/2.4.1-fosscuda-2019b-Python-3.7.4/lib/python3.7/site-packages/tensorflow/python/keras/layers/recurrent.py:3003 _generate_zero_filled_state
        return nest.map_structure(create_zeros, state_size)
    /sw/installed/TensorFlow/2.4.1-fosscuda-2019b-Python-3.7.4/lib/python3.7/site-packages/tensorflow/python/util/nest.py:659 map_structure
        structure[0], [func(*x) for x in entries],
    /sw/installed/TensorFlow/2.4.1-fosscuda-2019b-Python-3.7.4/lib/python3.7/site-packages/tensorflow/python/util/nest.py:659 <listcomp>
        structure[0], [func(*x) for x in entries],
    /sw/installed/TensorFlow/2.4.1-fosscuda-2019b-Python-3.7.4/lib/python3.7/site-packages/tensorflow/python/keras/layers/recurrent.py:3000 create_zeros
        return array_ops.zeros(init_state_size, dtype=dtype)
    /sw/installed/TensorFlow/2.4.1-fosscuda-2019b-Python-3.7.4/lib/python3.7/site-packages/tensorflow/python/util/dispatch.py:201 wrapper
        return target(*args, **kwargs)
    /sw/installed/TensorFlow/2.4.1-fosscuda-2019b-Python-3.7.4/lib/python3.7/site-packages/tensorflow/python/ops/array_ops.py:2819 wrapped
        tensor = fun(*args, **kwargs)
    /sw/installed/TensorFlow/2.4.1-fosscuda-2019b-Python-3.7.4/lib/python3.7/site-packages/tensorflow/python/ops/array_ops.py:2868 zeros
        output = _constant_if_small(zero, shape, dtype, name)
    /sw/installed/TensorFlow/2.4.1-fosscuda-2019b-Python-3.7.4/lib/python3.7/site-packages/tensorflow/python/ops/array_ops.py:2804 _constant_if_small
        if np.prod(shape) < 1000:
    <__array_function__ internals>:6 prod
        
    /home/h1/rosa992c/my-kernel/powerai-kernel2/lib/python3.7/site-packages/numpy/core/fromnumeric.py:3052 prod
        keepdims=keepdims, initial=initial, where=where)
    /home/h1/rosa992c/my-kernel/powerai-kernel2/lib/python3.7/site-packages/numpy/core/fromnumeric.py:86 _wrapreduction
        return ufunc.reduce(obj, axis, dtype, out, **passkwargs)
    /sw/installed/TensorFlow/2.4.1-fosscuda-2019b-Python-3.7.4/lib/python3.7/site-packages/tensorflow/python/framework/ops.py:855 __array__
        " a NumPy call, which is not supported".format(self.name))

    NotImplementedError: Cannot convert a symbolic Tensor (root/root/CalamariGraph/lstm_0/bidirectional/forward_lstm/strided_slice:0) to a numpy array. This error may indicate that you're trying to pass a Tensor to a NumPy call, which is not supported

(Before you ask: no, unfortunately I cannot install TF 2.6 here.)

[bertsky]

No, that's actually not related (I managed to get rid of the warning by using TF addons v. 0.12.1 built against TF 2.4.1, but the exception remains).