ValueError: 層序的輸入 0 與層不兼容：預期 ndim=3，發現 ndim=2。 收到的完整形狀：（無，1）

Question

使用我在網上找到的一些代碼，我成功地訓練了一個 model（哇。）能夠判斷我插入的音頻文件中是否有笑聲。

這是 model：

from keras.models import Sequential
from keras.layers import Dense, BatchNormalization, Flatten

lr_model = Sequential()
# lr_model.add(keras.Input((None, 128)))
lr_model.add(BatchNormalization(input_shape=(10, 128)))
lr_model.add(Flatten())
lr_model.add(Dense(1, activation='sigmoid'))
print(lr_model.summary())

# try using different optimizers and different optimizer configs
lr_model.compile(loss='binary_crossentropy',
              optimizer='adam',
              metrics=['accuracy'])

batch_size=32

CV_frac = 0.1
train_gen = data_generator(batch_size,'.../Data/bal_laugh_speech_subset.tfrecord', 0, 1-CV_frac)
val_gen = data_generator(128,'.../Data/bal_laugh_speech_subset.tfrecord', 1-CV_frac, 1)

rec_len = 18768



lr_h = lr_model.fit(train_gen,steps_per_epoch=int(rec_len*(1-CV_frac))//batch_size, epochs=100, 
                       validation_data=val_gen, validation_steps=int(rec_len*CV_frac)//128,
                       verbose=1)

然后我寫了這段代碼來插入我的音頻文件。它類似於檢測音頻文件中的 bat 調用的一些代碼。

soundarray, sr = librosa.load("video-1609522619.wav")
maxseconds = int(len(soundarray)/sr)
for second in range(maxseconds-1):
    audiosample = np.array(soundarray[second*sr:(second+1)*sr])
    print(audiosample)
    prediction = lr_model.predict(audiosample)
    
    if np.argmax(prediction)==1:
        IPython.display.display(IPython.display.Audio(audiosample, rate = sr,autoplay=True))
        librosa.output.write_wav('first_test.wav', audiosample, sr)
        
        time.sleep(2)
        print("hello")
        print("Detected laugh at " + str(second) + " out of " + str(maxseconds) + " seconds")
        print(prediction)

然后我得到了這個錯誤：

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-28-0d974db166cf> in <module>
      4     audiosample = np.array(soundarray[second*sr:(second+1)*sr])
      5     print(audiosample)
----> 6     prediction = lr_model.predict(audiosample)
      7 
      8     if np.argmax(prediction)==1:

~/.local/lib/python3.8/site-packages/tensorflow/python/keras/engine/training.py in predict(self, x, batch_size, verbose, steps, callbacks, max_queue_size, workers, use_multiprocessing)
   1627           for step in data_handler.steps():
   1628             callbacks.on_predict_batch_begin(step)
-> 1629             tmp_batch_outputs = self.predict_function(iterator)
   1630             if data_handler.should_sync:
   1631               context.async_wait()

~/.local/lib/python3.8/site-packages/tensorflow/python/eager/def_function.py in __call__(self, *args, **kwds)
    826     tracing_count = self.experimental_get_tracing_count()
    827     with trace.Trace(self._name) as tm:
--> 828       result = self._call(*args, **kwds)
    829       compiler = "xla" if self._experimental_compile else "nonXla"
    830       new_tracing_count = self.experimental_get_tracing_count()

~/.local/lib/python3.8/site-packages/tensorflow/python/eager/def_function.py in _call(self, *args, **kwds)
    860       # In this case we have not created variables on the first call. So we can
    861       # run the first trace but we should fail if variables are created.
--> 862       results = self._stateful_fn(*args, **kwds)
    863       if self._created_variables:
    864         raise ValueError("Creating variables on a non-first call to a function"

~/.local/lib/python3.8/site-packages/tensorflow/python/eager/function.py in __call__(self, *args, **kwargs)
   2939     with self._lock:
   2940       (graph_function,
-> 2941        filtered_flat_args) = self._maybe_define_function(args, kwargs)
   2942     return graph_function._call_flat(
   2943         filtered_flat_args, captured_inputs=graph_function.captured_inputs)  # pylint: disable=protected-access

~/.local/lib/python3.8/site-packages/tensorflow/python/eager/function.py in _maybe_define_function(self, args, kwargs)
   3355               self.input_signature is None and
   3356               call_context_key in self._function_cache.missed):
-> 3357             return self._define_function_with_shape_relaxation(
   3358                 args, kwargs, flat_args, filtered_flat_args, cache_key_context)
   3359 

~/.local/lib/python3.8/site-packages/tensorflow/python/eager/function.py in _define_function_with_shape_relaxation(self, args, kwargs, flat_args, filtered_flat_args, cache_key_context)
   3277           expand_composites=True)
   3278 
-> 3279     graph_function = self._create_graph_function(
   3280         args, kwargs, override_flat_arg_shapes=relaxed_arg_shapes)
   3281     self._function_cache.arg_relaxed[rank_only_cache_key] = graph_function

~/.local/lib/python3.8/site-packages/tensorflow/python/eager/function.py in _create_graph_function(self, args, kwargs, override_flat_arg_shapes)
   3194     arg_names = base_arg_names + missing_arg_names
   3195     graph_function = ConcreteFunction(
-> 3196         func_graph_module.func_graph_from_py_func(
   3197             self._name,
   3198             self._python_function,

~/.local/lib/python3.8/site-packages/tensorflow/python/framework/func_graph.py in func_graph_from_py_func(name, python_func, args, kwargs, signature, func_graph, autograph, autograph_options, add_control_dependencies, arg_names, op_return_value, collections, capture_by_value, override_flat_arg_shapes)
    988         _, original_func = tf_decorator.unwrap(python_func)
    989 
--> 990       func_outputs = python_func(*func_args, **func_kwargs)
    991 
    992       # invariant: `func_outputs` contains only Tensors, CompositeTensors,

~/.local/lib/python3.8/site-packages/tensorflow/python/eager/def_function.py in wrapped_fn(*args, **kwds)
    632             xla_context.Exit()
    633         else:
--> 634           out = weak_wrapped_fn().__wrapped__(*args, **kwds)
    635         return out
    636 

~/.local/lib/python3.8/site-packages/tensorflow/python/framework/func_graph.py in wrapper(*args, **kwargs)
    975           except Exception as e:  # pylint:disable=broad-except
    976             if hasattr(e, "ag_error_metadata"):
--> 977               raise e.ag_error_metadata.to_exception(e)
    978             else:
    979               raise

ValueError: in user code:

    /home/pete/.local/lib/python3.8/site-packages/tensorflow/python/keras/engine/training.py:1478 predict_function  *
        return step_function(self, iterator)
    /home/pete/.local/lib/python3.8/site-packages/tensorflow/python/keras/engine/training.py:1468 step_function  **
        outputs = model.distribute_strategy.run(run_step, args=(data,))
    /home/pete/.local/lib/python3.8/site-packages/tensorflow/python/distribute/distribute_lib.py:1259 run
        return self._extended.call_for_each_replica(fn, args=args, kwargs=kwargs)
    /home/pete/.local/lib/python3.8/site-packages/tensorflow/python/distribute/distribute_lib.py:2730 call_for_each_replica
        return self._call_for_each_replica(fn, args, kwargs)
    /home/pete/.local/lib/python3.8/site-packages/tensorflow/python/distribute/distribute_lib.py:3417 _call_for_each_replica
        return fn(*args, **kwargs)
    /home/pete/.local/lib/python3.8/site-packages/tensorflow/python/keras/engine/training.py:1461 run_step  **
        outputs = model.predict_step(data)
    /home/pete/.local/lib/python3.8/site-packages/tensorflow/python/keras/engine/training.py:1434 predict_step
        return self(x, training=False)
    /home/pete/.local/lib/python3.8/site-packages/tensorflow/python/keras/engine/base_layer.py:998 __call__
        input_spec.assert_input_compatibility(self.input_spec, inputs, self.name)
    /home/pete/.local/lib/python3.8/site-packages/tensorflow/python/keras/engine/input_spec.py:219 assert_input_compatibility
        raise ValueError('Input ' + str(input_index) + ' of layer ' +

    ValueError: Input 0 of layer sequential is incompatible with the layer: expected ndim=3, found ndim=2. Full shape received: (None, 1)

當我四處搜索時，這通常是 LSTM 模型的錯誤。 但事實並非如此。 無論如何，我的輸入與預期相比有些奇怪。

這是我的 model 總結：

Model: "sequential_2"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
batch_normalization_2 (Batch (None, 10, 128)           512       
_________________________________________________________________
flatten_2 (Flatten)          (None, 1280)              0         
_________________________________________________________________
dense_2 (Dense)              (None, 1)                 1281      
=================================================================
Total params: 1,793
Trainable params: 1,537
Non-trainable params: 256
_________________________________________________________________
None

我實際上不知道為什么 BatchNormalization 會有一個input_shape 。 因為這不是告訴 model 以某種方式批處理嗎？ 看起來更像是一個設置而不是一個層......這是問題的一部分嗎？ 無論如何，這部分很大程度上是從比我更有經驗的人那里復制而來的，並且最終它奏效了，所以我不願意把它弄得太多。

Answer 1

問題在於您的輸入形狀

工作示例代碼片段

from keras.models import Sequential
from keras.layers import Dense, BatchNormalization, Flatten
import tensorflow as tf

X_train = np.random.random((10,128))
y_train = np.random.random((10,1))

dataset = tf.data.Dataset.from_tensor_slices((X_train, y_train))
train_data = dataset.shuffle(len(X_train)).batch(32)
train_data = train_data.prefetch(
        buffer_size=tf.data.experimental.AUTOTUNE)

lr_model = Sequential()
#lr_model.add(tf.keras.Input((10, 128)))
lr_model.add(BatchNormalization(input_shape=(128,)))
lr_model.add(Flatten())
lr_model.add(Dense(1, activation='sigmoid'))
print(lr_model.summary())

lr_model.compile(optimizer='adam',
                loss=tf.keras.losses.BinaryCrossentropy(),
                metrics=['accuracy'])

lr_model.fit(train_data, epochs=5)

Output

Model: "sequential_10"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 batch_normalization_6 (Batc  (None, 128)              512       
 hNormalization)                                                 
                                                                 
 flatten_8 (Flatten)         (None, 128)               0         
                                                                 
 dense_18 (Dense)            (None, 1)                 129       
                                                                 
=================================================================
Total params: 641
Trainable params: 385
Non-trainable params: 256
_________________________________________________________________
None
Epoch 1/5
1/1 [==============================] - 1s 673ms/step - loss: 0.8737 - accuracy: 0.0000e+00
Epoch 2/5
1/1 [==============================] - 0s 11ms/step - loss: 0.8573 - accuracy: 0.0000e+00
Epoch 3/5
1/1 [==============================] - 0s 10ms/step - loss: 0.8413 - accuracy: 0.0000e+00
Epoch 4/5
1/1 [==============================] - 0s 8ms/step - loss: 0.8256 - accuracy: 0.0000e+00
Epoch 5/5
1/1 [==============================] - 0s 6ms/step - loss: 0.8104 - accuracy: 0.0000e+00