Optimizers.dp_optimizer_keras is not working under PlaidML-keras

Set-up:

tensorflow==2.4.0
tensorflow-privacy == 0.5.1
plaidml == 0.7.0
plaidml-keras == 0.7.0
Mac OS 10.15.7 (19H114) 16"
AMD Radeon Pro 5300M 4 GB

Problem:

Hello everyone.
I have been playing with some privacy-preserving mechanisms, and I got an error.
raise ValueError('Could not interpret optimizer identifier: ' + ValueError: Could not interpret optimizer identifier: <tensorflow_privacy.privacy.optimizers.dp_optimizer_keras.make_keras_optimizer_class.<locals>.DPOptimizerClass object at 0x10e9047f0>
That appears when I add the

optimizer = DPKerasSGDOptimizer(
    l2_norm_clip=l2_norm_clip,
    noise_multiplier=noise_multiplier,
    num_microbatches=num_microbatches,
    learning_rate=learning_rate)

from here: from tensorflow_privacy.privacy.optimizers.dp_optimizer_keras import DPKerasSGDOptimizer

Since in venv/lib/python3.8/site-packages/keras/optimizers.py line 788 if K.backend() == 'tensorflow': and in my case is: plaidml.keras.backend the exit on this function is the last one.
If I just return the identifier changing the if block in optimizers.py I got this:

ValueError                                Traceback (most recent call last)

<ipython-input-5-428b9b0ace60> in <module>
     35               optimizer=optimizer,
     36               metrics=['accuracy'])
---> 37 model.fit(x_train, y_train,
     38           batch_size=batch_size,
     39           epochs=epochs,

~venv/lib/python3.8/site-packages/keras/engine/training.py in fit(self, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, **kwargs)
   1008         else:
   1009             ins = x + y + sample_weights
-> 1010         self._make_train_function()
   1011         f = self.train_function
   1012 

~venv/lib/python3.8/site-packages/keras/engine/training.py in _make_train_function(self)
    505             with K.name_scope('training'):
    506                 with K.name_scope(self.optimizer.__class__.__name__):
--> 507                     training_updates = self.optimizer.get_updates(
    508                         params=self._collected_trainable_weights,
    509                         loss=self.total_loss)

~venv/lib/python3.8/site-packages/tensorflow/python/keras/optimizer_v2/optimizer_v2.py in get_updates(self, loss, params)
    725 
    726   def get_updates(self, loss, params):
--> 727     grads = self.get_gradients(loss, params)
    728     grads_and_vars = list(zip(grads, params))
    729     self._assert_valid_dtypes([

~venv/lib/python3.8/site-packages/tensorflow_privacy/privacy/optimizers/dp_optimizer_keras.py in get_gradients(self, loss, params)
    127       # in dp_optimizer.py, except that this returns only the gradients,
    128       # not the gradients and variables.
--> 129       microbatch_losses = tf.reshape(loss, [self._num_microbatches, -1])
    130       sample_params = (
    131           self._dp_sum_query.derive_sample_params(self._global_state))

~venv/lib/python3.8/site-packages/tensorflow/python/util/dispatch.py in wrapper(*args, **kwargs)
    199     """Call target, and fall back on dispatchers if there is a TypeError."""
    200     try:
--> 201       return target(*args, **kwargs)
    202     except (TypeError, ValueError):
    203       # Note: convert_to_eager_tensor currently raises a ValueError, not a

~venv/lib/python3.8/site-packages/tensorflow/python/ops/array_ops.py in reshape(tensor, shape, name)
    193     A `Tensor`. Has the same type as `tensor`.
    194   """
--> 195   result = gen_array_ops.reshape(tensor, shape, name)
    196   tensor_util.maybe_set_static_shape(result, shape)
    197   return result

~env/lib/python3.8/site-packages/tensorflow/python/ops/gen_array_ops.py in reshape(tensor, shape, name)
   8370       pass
   8371     try:
-> 8372       return reshape_eager_fallback(
   8373           tensor, shape, name=name, ctx=_ctx)
   8374     except _core._SymbolicException:

~venv/lib/python3.8/site-packages/tensorflow/python/ops/gen_array_ops.py in reshape_eager_fallback(tensor, shape, name, ctx)
   8391 
   8392 def reshape_eager_fallback(tensor, shape, name, ctx):
-> 8393   _attr_T, (tensor,) = _execute.args_to_matching_eager([tensor], ctx, [])
   8394   _attr_Tshape, (shape,) = _execute.args_to_matching_eager([shape], ctx, [_dtypes.int32, _dtypes.int64, ], _dtypes.int32)
   8395   _inputs_flat = [tensor, shape]

~venv/lib/python3.8/site-packages/tensorflow/python/eager/execute.py in args_to_matching_eager(l, ctx, allowed_dtypes, default_dtype)
    271 
    272       if tensor is None:
--> 273         tensor = ops.convert_to_tensor(
    274             t, dtype, preferred_dtype=default_dtype, ctx=ctx)
    275 

~venv/lib/python3.8/site-packages/tensorflow/python/profiler/trace.py in wrapped(*args, **kwargs)
    161         with Trace(trace_name, **trace_kwargs):
    162           return func(*args, **kwargs)
--> 163       return func(*args, **kwargs)
    164 
    165     return wrapped

~venv/lib/python3.8/site-packages/tensorflow/python/framework/ops.py in convert_to_tensor(value, dtype, name, as_ref, preferred_dtype, dtype_hint, ctx, accepted_result_types)
   1538 
   1539     if ret is None:
-> 1540       ret = conversion_func(value, dtype=dtype, name=name, as_ref=as_ref)
   1541 
   1542     if ret is NotImplemented:

~venv/lib/python3.8/site-packages/tensorflow/python/framework/constant_op.py in _constant_tensor_conversion_function(v, dtype, name, as_ref)
    337                                          as_ref=False):
    338   _ = as_ref
--> 339   return constant(v, dtype=dtype, name=name)
    340 
    341 

~venv/lib/python3.8/site-packages/tensorflow/python/framework/constant_op.py in constant(value, dtype, shape, name)
    262     ValueError: if called on a symbolic tensor.
    263   """
--> 264   return _constant_impl(value, dtype, shape, name, verify_shape=False,
    265                         allow_broadcast=True)
    266 

~venv/lib/python3.8/site-packages/tensorflow/python/framework/constant_op.py in _constant_impl(value, dtype, shape, name, verify_shape, allow_broadcast)
    274       with trace.Trace("tf.constant"):
    275         return _constant_eager_impl(ctx, value, dtype, shape, verify_shape)
--> 276     return _constant_eager_impl(ctx, value, dtype, shape, verify_shape)
    277 
    278   g = ops.get_default_graph()

~venv/lib/python3.8/site-packages/tensorflow/python/framework/constant_op.py in _constant_eager_impl(ctx, value, dtype, shape, verify_shape)
    299 def _constant_eager_impl(ctx, value, dtype, shape, verify_shape):
    300   """Implementation of eager constant."""
--> 301   t = convert_to_eager_tensor(value, ctx, dtype)
    302   if shape is None:
    303     return t

~venv/lib/python3.8/site-packages/tensorflow/python/framework/constant_op.py in convert_to_eager_tensor(value, ctx, dtype)
     96       dtype = dtypes.as_dtype(dtype).as_datatype_enum
     97   ctx.ensure_initialized()
---> 98   return ops.EagerTensor(value, ctx.device_name, dtype)
     99 
    100 

ValueError: TypeError: object of type 'Value' has no len()

Thanks

Start by reproducing the reported setup with tensorflow_privacy.privacy.optimizers.dp_optimizer_keras and the PlaidML Keras backend, then inspect the optimizer integration shown in dp_optimizer_keras.py and the reported Keras optimizer handling. Done means establishing whether this combination can work and covering the supported behavior or limitation with an appropriate test or documentation.