我正在尝试训练一个模仿特定玩家下棋风格的神经网络。 问题是我无法训练模型，因为它的形状。也许我需要制作 y.shape 1D 因为 y_train 的形状应该与输出层的形状相匹配。但后来我认为该模型无法跟踪每场比赛和每一步。此外，245 和 14 维度似乎也不适合训练模型。我如何实现模型以便它可以学习，或者我需要如何修改数据才能训练模型并获得移动作为输出？我期待模型进入棋盘状态并返回 1 个移动/标签
X 形 (24242, 245, 14, 8, 8) 包含 24242 个游戏的 numpy 数组，以数字表示， 每场比赛被填充到245步， 每一步由 14 个 8x8 棋盘组成（每种类型的棋子（兵、车、骑士、主教、皇后、国王）1 个棋盘）+ 1 个棋盘，以数字表示从白棋的角度显示棋盘上被攻击的每个方块 + 同样的事情黑色的碎片）
y 形 (24242, 245) 每个游戏的 1 个列表（总共 24242 个）每个包含 245 个元素（用“0”填充）标签编码 SAN，但只有我想模仿的玩家在游戏中玩过的动作。前 3 场比赛是这样的：
(24242, 245)
[[3830 3956 0 0 0 0 0 0 0 0 0 0 0 0...
0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0]
[4026 3961 0 0 0 0 0 0 0 0 0 0 0 0...
0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0]
[3769 3711 715 3830 76 3958 3772 4024 4026 4051 3986 3873 58 225
976 1065 1643 1702 3893 79 3917 1318 890 0 0 0 0 0 ...
0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0]]
<class 'numpy.ndarray'>
型号：
def build_model(conv_size, conv_depth):
board3d = layers.Input(shape=(max_len, 14, 8, 8))
x = board3d
for _ in range(conv_depth):
x = layers.Conv2D(filters=conv_size, kernel_size=3, padding="same", activation="relu")(x)
x = layers.Flatten()(x)
x = layers.Dense(64, "relu")(x)
x = layers.Dense(len(d), "sigmoid")(x) #len(d) = 4066 distinct labels
return models.Model(inputs=board3d, outputs=x)
model = build_model(32, 4)
model.compile(optimizer=optimizers.Adam(5e-4), loss="categorical_crossentropy")
model.summary()
model.fit(X_train, y_train,
batch_size=5,
epochs=1000,
verbose=1,
validation_split=0.1,
callbacks=[callbacks.ReduceLROnPlateau(monitor="loss", patience=10),
callbacks.EarlyStopping(monitor="loss", patience=15, min_delta=0.0001)])
输出
Model: "model_2"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
input_9 (InputLayer) [(None, 248, 14, 8, 8)] 0
conv2d_8 (Conv2D) (None, 248, 14, 8, 32) 2336
conv2d_9 (Conv2D) (None, 248, 14, 8, 32) 9248
conv2d_10 (Conv2D) (None, 248, 14, 8, 32) 9248
conv2d_11 (Conv2D) (None, 248, 14, 8, 32) 9248
flatten_2 (Flatten) (None, 888832) 0
dense_6 (Dense) (None, 64) 56885312
dense_7 (Dense) (None, 4063) 264095
=================================================================
Total params: 57,179,487
Trainable params: 57,179,487
Non-trainable params: 0
_________________________________________________________________
Epoch 1/1000
InvalidArgumentError Traceback (most recent call last)
Cell In[151], line 23
20 # Reshape y_train to match the output shape of the model
21 y_train = y_train.reshape(y_train.shape[0], -1)
---> 23 model.fit(X_train, y_train,
24 batch_size=5,
25 epochs=1000,
26 verbose=1,
27 validation_split=0.1,
28 callbacks=[callbacks.ReduceLROnPlateau(monitor="loss", patience=10),
29 callbacks.EarlyStopping(monitor="loss", patience=15, min_delta=0.0001)])
File ~\AppData\Roaming\Python\Python310\site-packages\keras\utils\traceback_utils.py:70, in filter_traceback..error_handler(*args, **kwargs)
67 filtered_tb = _process_traceback_frames(e.__traceback__)
68 # To get the full stack trace, call:
69 # `tf.debugging.disable_traceback_filtering()`
---> 70 raise e.with_traceback(filtered_tb) from None
71 finally:
72 del filtered_tb
File ~\AppData\Roaming\Python\Python310\site-packages\tensorflow\python\eager\execute.py:52, in quick_execute(op_name, num_outputs, inputs, attrs, ctx, name)
50 try:
51 ctx.ensure_initialized()
---> 52 tensors = pywrap_tfe.TFE_Py_Execute(ctx._handle, device_name, op_name,
...
File "C:\Users\ffgd\AppData\Roaming\Python\Python310\site-packages\keras\backend.py", line 5633, in sparse_categorical_crossentropy
res = tf.nn.sparse_softmax_cross_entropy_with_logits(
Node: 'sparse_categorical_crossentropy/SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits'
logits and labels must have the same first dimension, got logits shape [5,4066] and labels shape [1225]
[[{{node sparse_categorical_crossentropy/SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits}}]] [Op:__inference_train_function_4291]
有时我在尝试更改模型时也会收到与此错误消息类似的信息：
#ValueError：形状（无，245）和（无，4066）不兼容