我正在尝试制作一个LSTM模型，该模型能够在每个时间段计算三个类的概率。
我以以下方式定义模型：
from tensorflow import keras
from tensorflow.keras import layers
model = keras.Sequential()
model.add(layers.LSTM(128,input_shape=(None,input_size)))
model.add(layers.Dense(output_size))
model.add(layers.Softmax())
model.compile(optimizer="sgd",loss=keras.losses.CategoricalCrossentropy())
model.fit(Generator())
生成器提供由一个训练示例组成的批处理(我计划使用多个示例的批处理，但因为我的训练示例的长度可变，所以我现在尽可能简单(。更准确地说，生成器返回具有shape(1，T，input_size(的x，返回具有shake(1，T，3(的y。在处理了几个批次后，keras失败，出现以下错误
InvalidArgumentError Traceback (most recent call last)
<ipython-input-51-ad3e71307f94> in <module>
1 model.compile(optimizer="sgd",loss=keras.losses.CategoricalCrossentropy())
----> 2 model.fit(Generator())
3
c:theprojectvenvlibsite-packagestensorflowpythonkerasenginetraining.py in _method_wrapper(self, *args, **kwargs)
106 def _method_wrapper(self, *args, **kwargs):
107 if not self._in_multi_worker_mode(): # pylint: disable=protected-access
--> 108 return method(self, *args, **kwargs)
109
110 # Running inside `run_distribute_coordinator` already.
c:theprojectvenvlibsite-packagestensorflowpythonkerasenginetraining.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, validation_batch_size, validation_freq, max_queue_size, workers, use_multiprocessing)
1096 batch_size=batch_size):
1097 callbacks.on_train_batch_begin(step)
-> 1098 tmp_logs = train_function(iterator)
1099 if data_handler.should_sync:
1100 context.async_wait()
c:theprojectvenvlibsite-packagestensorflowpythoneagerdef_function.py in __call__(self, *args, **kwds)
778 else:
779 compiler = "nonXla"
--> 780 result = self._call(*args, **kwds)
781
782 new_tracing_count = self._get_tracing_count()
c:theprojectvenvlibsite-packagestensorflowpythoneagerdef_function.py in _call(self, *args, **kwds)
838 # Lifting succeeded, so variables are initialized and we can run the
839 # stateless function.
--> 840 return self._stateless_fn(*args, **kwds)
841 else:
842 canon_args, canon_kwds =
c:theprojectvenvlibsite-packagestensorflowpythoneagerfunction.py in __call__(self, *args, **kwargs)
2827 with self._lock:
2828 graph_function, args, kwargs = self._maybe_define_function(args, kwargs)
-> 2829 return graph_function._filtered_call(args, kwargs) # pylint: disable=protected-access
2830
2831 @property
c:theprojectvenvlibsite-packagestensorflowpythoneagerfunction.py in _filtered_call(self, args, kwargs, cancellation_manager)
1841 `args` and `kwargs`.
1842 """
-> 1843 return self._call_flat(
1844 [t for t in nest.flatten((args, kwargs), expand_composites=True)
1845 if isinstance(t, (ops.Tensor,
c:theprojectvenvlibsite-packagestensorflowpythoneagerfunction.py in _call_flat(self, args, captured_inputs, cancellation_manager)
1921 and executing_eagerly):
1922 # No tape is watching; skip to running the function.
-> 1923 return self._build_call_outputs(self._inference_function.call(
1924 ctx, args, cancellation_manager=cancellation_manager))
1925 forward_backward = self._select_forward_and_backward_functions(
c:theprojectvenvlibsite-packagestensorflowpythoneagerfunction.py in call(self, ctx, args, cancellation_manager)
543 with _InterpolateFunctionError(self):
544 if cancellation_manager is None:
--> 545 outputs = execute.execute(
546 str(self.signature.name),
547 num_outputs=self._num_outputs,
c:theprojectvenvlibsite-packagestensorflowpythoneagerexecute.py in quick_execute(op_name, num_outputs, inputs, attrs, ctx, name)
57 try:
58 ctx.ensure_initialized()
---> 59 tensors = pywrap_tfe.TFE_Py_Execute(ctx._handle, device_name, op_name,
60 inputs, attrs, num_outputs)
61 except core._NotOkStatusException as e:
InvalidArgumentError: Can not squeeze dim[2], expected a dimension of 1, got 3
[[node categorical_crossentropy/remove_squeezable_dimensions/Squeeze (defined at <ipython-input-51-ad3e71307f94>:2) ]] [Op:__inference_train_function_16213]
Function call stack:
train_function
这是一个样本(不是唯一一个(的x和y，它失败了：
0 1 2 3 4 5 6 7 8 9 ... 169 170
0 0.585963 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 ... 0.0 0.0
1 0.831822 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 ... 0.0 0.0
2 0.831822 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 ... 0.0 0.0
3 0.831822 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 ... 0.0 0.0
4 0.831822 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 ... 0.0 0.0
.. ... ... ... ... ... ... ... ... ... ... ... ... ...
984 0.989131 1.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 ... 0.0 0.0
985 0.989927 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 ... 0.0 0.0
986 0.990885 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 ... 0.0 0.0
987 0.990911 0.0 1.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 ... 0.0 0.0
988 0.991843 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 ... 0.0 0.0
171 172 173 174 175 176 177 178
0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
1 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
2 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
3 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
4 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
.. ... ... ... ... ... ... ... ...
984 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
985 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
986 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
987 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
988 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
事实证明，对于这种问题，LSTM之后的层应该用TimeDistributed层包裹，LSTM应该采用参数return_sequences=True