Question

我想使用Keras＆＃39;运行外部优化器。 TFOptimizer包装并给它tf.contrib.opt.ScipyOptimizerInterface，但我没有成功。

我的模型是这个（非常小的例子）：

model = Sequential()
model.add(InputLayer(input_shape=(6,))
model.add(Dense(3, activation='sigmoid'))
model.add(Dense(5, activation='linear'))

我尝试编译：

model.compile(optimizer=TFOptimizer(ScipyOptimizerInterface()))

但是，当然，它不起作用，因为ScipyOptimizerInterface构造函数在初始化期间需要loss参数。

所以，我创建了两个自己的类：

class CGTFOptimizer(object):
    def compute_gradients(self, loss, params):
        optimizer = ScipyOptimizerInterface(loss, method='cg')
        result = optimizer.minimize(loss)
        return result

class CGTF(TFOptimizer):
    """ Wrapper for TFOptimizer """
    def __init__(self):
       super(CGTF, self).__init__(optimizer=CGTFOptimizer())

使用model.compile(optimizer=CGTF())编译模型成功但经过培训后，我从ScipyOptimizerInterface调用收到以下错误：

Traceback (most recent call last):
  File "venv/py3/lib/python3.6/site-packages/tensorflow/python/framework/op_def_library.py", line 455, in _apply_op_helper
    as_ref=input_arg.is_ref)
  File "venv/py3/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 991, in internal_convert_n_to_tensor
    ctx=ctx))
  File "venv/py3/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 926, in internal_convert_to_tensor
    ret = conversion_func(value, dtype=dtype, name=name, as_ref=as_ref)
  File "venv/py3/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 774, in _TensorTensorConversionFunction
    (dtype.name, t.dtype.name, str(t)))
ValueError: Tensor conversion requested dtype float32 for Tensor with dtype int64: 'Tensor("training/TFOptimizer/Reshape_25:0", shape=(1,), dtype=int64)'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "directly_with_tensorflow.py", line 136, in <module>
    cgd_model.fit(X_reference, y_reference, epochs=50, batch_size=5)
  File "venv/py3/lib/python3.6/site-packages/tensorflow/python/keras/_impl/keras/models.py", line 841, in fit
    initial_epoch=initial_epoch)
  File "venv/py3/lib/python3.6/site-packages/tensorflow/python/keras/_impl/keras/engine/training.py", line 1590, in fit
    self._make_train_function()
  File "venv/py3/lib/python3.6/site-packages/tensorflow/python/keras/_impl/keras/engine/training.py", line 952, in _make_train_function
    params=self._collected_trainable_weights, loss=self.total_loss)
  File "venv/py3/lib/python3.6/site-packages/tensorflow/python/keras/_impl/keras/optimizers.py", line 674, in get_updates
    grads = self.optimizer.compute_gradients(loss, params)
  File "directly_with_tensorflow.py", line 113, in compute_gradients
    optimizer = ScipyOptimizerInterface(loss, method='cg')
  File "venv/py3/lib/python3.6/site-packages/tensorflow/contrib/opt/python/training/external_optimizer.py", line 126, in __init__
    self._packed_var = self._pack(self._vars)
  File "venv/py3/lib/python3.6/site-packages/tensorflow/contrib/opt/python/training/external_optimizer.py", line 259, in _pack
    return array_ops.concat(flattened, 0)
  File "venv/py3/lib/python3.6/site-packages/tensorflow/python/ops/array_ops.py", line 1099, in concat
    return gen_array_ops._concat_v2(values=values, axis=axis, name=name)
  File "venv/py3/lib/python3.6/site-packages/tensorflow/python/ops/gen_array_ops.py", line 706, in _concat_v2
    "ConcatV2", values=values, axis=axis, name=name)
  File "venv/py3/lib/python3.6/site-packages/tensorflow/python/framework/op_def_library.py", line 483, in _apply_op_helper
    raise TypeError("%s that don't all match." % prefix)
TypeError: Tensors in list passed to 'values' of 'ConcatV2' Op have types [float32, float32, float32, float32, float32, float32, float32, float32, float32, float32, float32, float32, int64] that don't all match.

这可能是TensorFlow中的问题，而不是Keras？我尝试了将loss转换为float32，float64或int64的不同方法，但没有任何帮助。

X_reference和y_reference的形状为(10,6)，(10,5)。类型float32。

有人能指出我如何在Keras中使用ScipyOptimizers吗？

在Keras中使用时，tensorflow ScipyOptimizerInterface中的TypeError

0 个答案: