python - keras model.fit ValueError：indices.shape=[1,11,1] 的外部 2 维必须与 updates.shape=[2] 的外部 2 维匹配

Question

我正在训练具有自定义损失和评估指标的 keras 模型。它在没有度量的情况下进行训练。但是当我尝试训练时它给出了以下错误：

model.compile(optimizer= keras.optimizers.Adam(learning_rate = 1e-3), loss = inner_product, metrics=dice_index_metric)
 model.fit([X_train], [y_train], epochs=50, batch_size = 1, validation_split=0.2, 
                 callbacks = keras.callbacks.EarlyStopping(monitor="val_loss", min_delta=0, patience=5), verbose=2)

错误：

---

ValueError                                Traceback (most recent call last)
<ipython-input-38-270dbe25d468> in <module>
----> 1 hist = model.fit([X_train], [y_train1], epochs=50, batch_size = 1, validation_split=0.2, 
      2                  callbacks = keras.callbacks.EarlyStopping(monitor="val_loss", min_delta=0, patience=5), verbose=2)

~\AppData\Roaming\Python\Python38\site-packages\tensorflow\python\keras\engine\training.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.

~\AppData\Roaming\Python\Python38\site-packages\tensorflow\python\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, 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()

~\AppData\Roaming\Python\Python38\site-packages\tensorflow\python\eager\def_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()

~\AppData\Roaming\Python\Python38\site-packages\tensorflow\python\eager\def_function.py in _call(self, *args, **kwds)
    821       # This is the first call of __call__, so we have to initialize.
    822       initializers = []
--> 823       self._initialize(args, kwds, add_initializers_to=initializers)
    824     finally:
    825       # At this point we know that the initialization is complete (or less

~\AppData\Roaming\Python\Python38\site-packages\tensorflow\python\eager\def_function.py in _initialize(self, args, kwds, add_initializers_to)
    694     self._graph_deleter = FunctionDeleter(self._lifted_initializer_graph)
    695     self._concrete_stateful_fn = (
--> 696         self._stateful_fn._get_concrete_function_internal_garbage_collected(  # pylint: disable=protected-access
    697             *args, **kwds))
    698 

~\AppData\Roaming\Python\Python38\site-packages\tensorflow\python\eager\function.py in _get_concrete_function_internal_garbage_collected(self, *args, **kwargs)
   2853       args, kwargs = None, None
   2854     with self._lock:
-> 2855       graph_function, _, _ = self._maybe_define_function(args, kwargs)
   2856     return graph_function
   2857 

~\AppData\Roaming\Python\Python38\site-packages\tensorflow\python\eager\function.py in _maybe_define_function(self, args, kwargs)
   3211 
   3212       self._function_cache.missed.add(call_context_key)
-> 3213       graph_function = self._create_graph_function(args, kwargs)
   3214       self._function_cache.primary[cache_key] = graph_function
   3215       return graph_function, args, kwargs

~\AppData\Roaming\Python\Python38\site-packages\tensorflow\python\eager\function.py in _create_graph_function(self, args, kwargs, override_flat_arg_shapes)
   3063     arg_names = base_arg_names + missing_arg_names
   3064     graph_function = ConcreteFunction(
-> 3065         func_graph_module.func_graph_from_py_func(
   3066             self._name,
   3067             self._python_function,

~\AppData\Roaming\Python\Python38\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)
    984         _, original_func = tf_decorator.unwrap(python_func)
    985 
--> 986       func_outputs = python_func(*func_args, **func_kwargs)
    987 
    988       # invariant: `func_outputs` contains only Tensors, CompositeTensors,

~\AppData\Roaming\Python\Python38\site-packages\tensorflow\python\eager\def_function.py in wrapped_fn(*args, **kwds)
    598         # __wrapped__ allows AutoGraph to swap in a converted function. We give
    599         # the function a weak reference to itself to avoid a reference cycle.
--> 600         return weak_wrapped_fn().__wrapped__(*args, **kwds)
    601     weak_wrapped_fn = weakref.ref(wrapped_fn)
    602 

~\AppData\Roaming\Python\Python38\site-packages\tensorflow\python\framework\func_graph.py in wrapper(*args, **kwargs)
    971           except Exception as e:  # pylint:disable=broad-except
    972             if hasattr(e, "ag_error_metadata"):
--> 973               raise e.ag_error_metadata.to_exception(e)
    974             else:
    975               raise

ValueError: in user code:

    C:\Users\haluk\AppData\Roaming\Python\Python38\site-packages\tensorflow\python\keras\engine\training.py:806 train_function  *
        return step_function(self, iterator)
    <ipython-input-36-2c54f0983574>:5 dice_index_metric  *
        y_pred1 = tf.scatter_nd(ind, updates, tf.shape(y_pred))
    C:\Users\haluk\AppData\Roaming\Python\Python38\site-packages\tensorflow\python\ops\gen_array_ops.py:8855 scatter_nd  **
        _, _, _op, _outputs = _op_def_library._apply_op_helper(
    C:\Users\haluk\AppData\Roaming\Python\Python38\site-packages\tensorflow\python\framework\op_def_library.py:742 _apply_op_helper
        op = g._create_op_internal(op_type_name, inputs, dtypes=None,
    C:\Users\haluk\AppData\Roaming\Python\Python38\site-packages\tensorflow\python\framework\func_graph.py:591 _create_op_internal
        return super(FuncGraph, self)._create_op_internal(  # pylint: disable=protected-access
    C:\Users\haluk\AppData\Roaming\Python\Python38\site-packages\tensorflow\python\framework\ops.py:3477 _create_op_internal
        ret = Operation(
    C:\Users\haluk\AppData\Roaming\Python\Python38\site-packages\tensorflow\python\framework\ops.py:1974 __init__
        self._c_op = _create_c_op(self._graph, node_def, inputs,
    C:\Users\haluk\AppData\Roaming\Python\Python38\site-packages\tensorflow\python\framework\ops.py:1815 _create_c_op
        raise ValueError(str(e))

    ValueError: The outer 2 dimensions of indices.shape=[1,11,1] must match the outer 2 dimensions of updates.shape=[2]: Shapes must be equal rank, but are 2 and 1 for '{{node ScatterNd}} = ScatterNd[T=DT_INT32, Tindices=DT_INT32](strided_slice_2, Const_3, Shape_1)' with input shapes: [1,11,1], [2], [2].

自定义指标如下：

def dice_index_metric(y_true, y_pred):
    ind = tf.argsort(y_pred,axis=-1,direction='ASCENDING',stable=False,name=None)[-2:]
    ind = ind[..., tf.newaxis]
    updates = tf.constant([1, 1])
    y_pred1 = tf.scatter_nd(ind, updates, tf.shape(y_pred))
    innerproduct = tf.minimum(y_true, y_pred1)
    innerproduct = tf.reduce_sum(innerproduct)
    union= tf.maximum(y_true, y_pred1)
    union = tf.reduce_sum(union)
    return innerproduct/union

自定义度量将预测向量转换为一个向量，其中预测中的前 2 个元素为 1，其他元素为 0，然后将其与真值进行比较并计算它们的 (# of intersection)/(# of union) 所以让我们说来自模型的预测是：

pred = [0.01, 0.3, 0,4 0.01, 0.01, 0.2, 0.02, 0.05],

前 2 个值是 0.3 和 0.4，索引为 1、2。那么我应该推荐这个：

推荐 = [0, 1, 1, 0, 0, 0, 0, 0],

如果真值如下，则它们的交集仅为索引 2，并且它们的并集为 [1,2,3]，则 i 应返回 1/3。

真相 = [0, 0, 1, 1, 0, 0, 0, 0]

模型：

inputs = keras.Input(shape =(None,23))

features = layers.LSTM(100)(inputs)

next = layers.Dense(11, activation=activations.sigmoid)(features)

next = layers.Softmax()(next)

model = keras.Model(inputs=[inputs] , outputs=next, name="LSTMmodel2")

标签： pythontensorflowmachine-learningkerasrecommendation-engine

Answer 1

首先，从你的代码

ind = tf.argsort(y_pred,axis=-1,direction='ASCENDING',stable=False,name=None)[-2:]

您忘记了 y_pred 是成批的。这意味着 y_pred 的形状不是 [11,] 而是 [N,11]，并且从您的错误消息中假设批量大小为 1。因此，上面的线在批量轴（即轴 0）中分割。这就是为什么错误说

indices.shape=[1,11,1]

其次， scatter_nd 不能那样工作。与gather_nd 不同，它不支持batch_dims。'indices' 的最后一个轴值是唯一可以影响 'update' 元素可以进入的位置的东西。

例如，

tf.scatter_nd([[1,2],[1,3]],[1,1],shape=(2,10))
#<tf.Tensor: shape=(2, 10), dtype=int32, 
#numpy=array([[0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
#             [0, 0, 1, 1, 0, 0, 0, 0, 0, 0]])>

所以在你的情况下，

N = tf.shape(y_pred)[0]
# Shape: [N,2]
ind = tf.argsort(y_pred,axis=-1,direction='ASCENDING',stable=False,name=None)[:,-2:]

# Shape: [N,2,1]
ind = ind[...,tf.newaxis]

# Dummy range to add index for batch axis
# Shape : [N,1,1]
r = tf.range(N)[:,tf.newaxis,tf.newaxis]

# Shape : [N,2,1]
r = tf.repeat(r,2,axis=1)

# Shape : [N,2,2]
ind = tf.concat([r,ind],axis=-1)

# Shape : [N,2]
updates = tf.ones((N,2))

y_pred1 = tf.scatter_nd(ind, updates, tf.shape(y_pred))

是使用 tf.scatter_nd 的正确方法

然而，它看起来很脏。我宁愿推荐使用这种方式：

second_max = tf.sort(y_pred,axis=-1,direction='ASCENDING')[:,-2,tf.newaxis]
y_pred1 = tf.cast(y_pred>=second_max,tf.int32)