我的准确性保持不变，我该怎么做才能解决这个问题

Question

在最终修复了这段代码给我的所有错误之后，我偶然发现了一个新问题。这次是工作模型为我提供了它。这是我创建的代码，这是我编写的第三个深度学习代码，我在编写它时玩得很开心，但是，因为我是 Python 的初学者，所以有些想法很难掌握。

import pandas as pd
from sklearn.model_selection import train_test_split
import keras as kr
from keras import layers
from keras import Sequential
from keras.layers import Dense
from sklearn.preprocessing import MinMaxScaler
from sklearn.model_selection import train_test_split
import tensorflow as tf
from sklearn.preprocessing import LabelEncoder
from pandas import DataFrame


config = tf.compat.v1.ConfigProto()
config.gpu_options.allow_growth = True
session = tf.compat.v1.InteractiveSession(config=config)

pd.set_option('display.max_columns', None)

headers = ['id', 'rated', 'created_at', 'last_move_at', 'turns', 'victory_status', 'winner', 'increment_code',
           'white_id', 'white_rating', 'black_id', 'black_rating', 'moves', 'opening_eco', 'opening_name',
           'opening_ply']
data = pd.read_csv(r'C:\games.csv', header=None, names=headers)
dataset = DataFrame(data)
dd = dataset.drop([0])
df = dd.drop(columns=['id', 'rated', 'opening_name', 'created_at', 'last_move_at', 'increment_code', 'white_id',
                      'black_id', 'opening_ply', 'opening_name', 'turns', 'victory_status', 'moves', 'opening_eco'],
             axis=1)

df['winner'] = df['winner'].map({'black': 0, 'white': 1})
y = df['winner']
encoder = LabelEncoder()
encoder.fit(y)
encoded_y = encoder.transform(y)
X = df.drop('winner', axis=1)
X = X.astype("float32")
X_train, X_test, y_train, y_test = train_test_split(X, encoded_y, test_size=0.2)
sc = MinMaxScaler()
scaled_X_train = sc.fit_transform(X_train)
scaled_X_test = sc.fit_transform(X_test)

model = Sequential()
model.add(Dense(2, input_dim=2, activation='relu'))
model.add(tf.keras.Input(shape=(12, 2)))
model.add(Dense(4, activation='relu'))
model.add(Dense(8, activation='relu'))
model.add(Dense(16, activation='relu'))
model.add(Dense(8, activation='relu'))
model.add(Dense(1, activation='sigmoid'))
model.compile(loss=kr.losses.binary_crossentropy, optimizer='adam',
              metrics=['accuracy'])
history = model.fit(scaled_X_train, y_train, batch_size=50, epochs=100, verbose=1, validation_data=(scaled_X_test,
                                                                                                    y_test))
print(history.history)
score = model.evaluate(scaled_X_train, y_train, verbose=1)

我的代码似乎在前几个时期运行良好，准确性提高了。然而，在那之后，准确度似乎不再有任何进步，并且落在了 0.610 左右的适度准确度上，或者具体如下所示。不知道如何让它更高，我来找你问你一个问题：'我该如何解决这个问题？'

Epoch 1/100
321/321 [==============================] - 0s 2ms/step - loss: 0.6386 - accuracy: 0.5463 - val_loss: 0.6208 - val_accuracy: 0.5783
Epoch 2/100
321/321 [==============================] - 0s 925us/step - loss: 0.6098 - accuracy: 0.6091 - val_loss: 0.6078 - val_accuracy: 0.5960
Epoch 3/100
321/321 [==============================] - 0s 973us/step - loss: 0.6055 - accuracy: 0.6102 - val_loss: 0.6177 - val_accuracy: 0.5833
Epoch 4/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6042 - accuracy: 0.6129 - val_loss: 0.6138 - val_accuracy: 0.5850
Epoch 5/100
321/321 [==============================] - 0s 973us/step - loss: 0.6041 - accuracy: 0.6106 - val_loss: 0.6233 - val_accuracy: 0.5763
Epoch 6/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6046 - accuracy: 0.6097 - val_loss: 0.6276 - val_accuracy: 0.5733
Epoch 7/100
321/321 [==============================] - 0s 973us/step - loss: 0.6033 - accuracy: 0.6086 - val_loss: 0.6238 - val_accuracy: 0.5733
Epoch 8/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6023 - accuracy: 0.6116 - val_loss: 0.6202 - val_accuracy: 0.5770
Epoch 9/100
321/321 [==============================] - 0s 973us/step - loss: 0.6030 - accuracy: 0.6091 - val_loss: 0.6210 - val_accuracy: 0.5738
Epoch 10/100
321/321 [==============================] - 0s 973us/step - loss: 0.6028 - accuracy: 0.6098 - val_loss: 0.6033 - val_accuracy: 0.5932
Epoch 11/100
321/321 [==============================] - 0s 973us/step - loss: 0.6022 - accuracy: 0.6094 - val_loss: 0.6166 - val_accuracy: 0.5780
Epoch 12/100
321/321 [==============================] - 0s 925us/step - loss: 0.6025 - accuracy: 0.6104 - val_loss: 0.6026 - val_accuracy: 0.5947
Epoch 13/100
321/321 [==============================] - 0s 925us/step - loss: 0.6021 - accuracy: 0.6099 - val_loss: 0.6243 - val_accuracy: 0.5733
Epoch 14/100
321/321 [==============================] - 0s 876us/step - loss: 0.6027 - accuracy: 0.6098 - val_loss: 0.6176 - val_accuracy: 0.5775
Epoch 15/100
321/321 [==============================] - 0s 925us/step - loss: 0.6029 - accuracy: 0.6091 - val_loss: 0.6286 - val_accuracy: 0.5690
Epoch 16/100
321/321 [==============================] - 0s 876us/step - loss: 0.6025 - accuracy: 0.6083 - val_loss: 0.6104 - val_accuracy: 0.5840
Epoch 17/100
321/321 [==============================] - 0s 876us/step - loss: 0.6021 - accuracy: 0.6102 - val_loss: 0.6039 - val_accuracy: 0.5897
Epoch 18/100
321/321 [==============================] - 0s 973us/step - loss: 0.6021 - accuracy: 0.6113 - val_loss: 0.6046 - val_accuracy: 0.5887
Epoch 19/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6019 - accuracy: 0.6083 - val_loss: 0.6074 - val_accuracy: 0.5860
Epoch 20/100
321/321 [==============================] - 0s 971us/step - loss: 0.6021 - accuracy: 0.6089 - val_loss: 0.6194 - val_accuracy: 0.5738
Epoch 21/100
321/321 [==============================] - 0s 876us/step - loss: 0.6025 - accuracy: 0.6099 - val_loss: 0.6093 - val_accuracy: 0.5857
Epoch 22/100
321/321 [==============================] - 0s 925us/step - loss: 0.6020 - accuracy: 0.6097 - val_loss: 0.6154 - val_accuracy: 0.5773
Epoch 23/100
321/321 [==============================] - 0s 973us/step - loss: 0.6027 - accuracy: 0.6104 - val_loss: 0.6044 - val_accuracy: 0.5895
Epoch 24/100
321/321 [==============================] - 0s 973us/step - loss: 0.6015 - accuracy: 0.6112 - val_loss: 0.6305 - val_accuracy: 0.5710
Epoch 25/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6016 - accuracy: 0.6114 - val_loss: 0.6067 - val_accuracy: 0.5867
Epoch 26/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6017 - accuracy: 0.6102 - val_loss: 0.6140 - val_accuracy: 0.5800
Epoch 27/100
321/321 [==============================] - 0s 973us/step - loss: 0.6025 - accuracy: 0.6075 - val_loss: 0.6190 - val_accuracy: 0.5755
Epoch 28/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6029 - accuracy: 0.6087 - val_loss: 0.6337 - val_accuracy: 0.5666
Epoch 29/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6021 - accuracy: 0.6095 - val_loss: 0.6089 - val_accuracy: 0.5840
Epoch 30/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6026 - accuracy: 0.6106 - val_loss: 0.6273 - val_accuracy: 0.5690
Epoch 31/100
321/321 [==============================] - 0s 925us/step - loss: 0.6020 - accuracy: 0.6083 - val_loss: 0.6146 - val_accuracy: 0.5785
Epoch 32/100
321/321 [==============================] - 0s 973us/step - loss: 0.6017 - accuracy: 0.6116 - val_loss: 0.6093 - val_accuracy: 0.5837
Epoch 33/100
321/321 [==============================] - 0s 973us/step - loss: 0.6025 - accuracy: 0.6096 - val_loss: 0.6139 - val_accuracy: 0.5780
Epoch 34/100
321/321 [==============================] - 0s 973us/step - loss: 0.6022 - accuracy: 0.6087 - val_loss: 0.6090 - val_accuracy: 0.5850
Epoch 35/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6018 - accuracy: 0.6096 - val_loss: 0.6127 - val_accuracy: 0.5810
Epoch 36/100
321/321 [==============================] - 0s 876us/step - loss: 0.6024 - accuracy: 0.6091 - val_loss: 0.6001 - val_accuracy: 0.5975
Epoch 37/100
321/321 [==============================] - 0s 973us/step - loss: 0.6027 - accuracy: 0.6104 - val_loss: 0.6083 - val_accuracy: 0.5862
Epoch 38/100
321/321 [==============================] - 0s 973us/step - loss: 0.6020 - accuracy: 0.6090 - val_loss: 0.6073 - val_accuracy: 0.5875
Epoch 39/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6023 - accuracy: 0.6109 - val_loss: 0.6149 - val_accuracy: 0.5785
Epoch 40/100
321/321 [==============================] - 0s 973us/step - loss: 0.6022 - accuracy: 0.6085 - val_loss: 0.6175 - val_accuracy: 0.5758
Epoch 41/100
321/321 [==============================] - 0s 973us/step - loss: 0.6017 - accuracy: 0.6079 - val_loss: 0.6062 - val_accuracy: 0.5865
Epoch 42/100
321/321 [==============================] - 0s 973us/step - loss: 0.6018 - accuracy: 0.6097 - val_loss: 0.6060 - val_accuracy: 0.5867
Epoch 43/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6018 - accuracy: 0.6082 - val_loss: 0.6074 - val_accuracy: 0.5862
Epoch 44/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6019 - accuracy: 0.6096 - val_loss: 0.6150 - val_accuracy: 0.5785
Epoch 45/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6014 - accuracy: 0.6112 - val_loss: 0.6241 - val_accuracy: 0.5740
Epoch 46/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6023 - accuracy: 0.6111 - val_loss: 0.6118 - val_accuracy: 0.5815
Epoch 47/100
321/321 [==============================] - 0s 973us/step - loss: 0.6017 - accuracy: 0.6073 - val_loss: 0.6110 - val_accuracy: 0.5835
Epoch 48/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6021 - accuracy: 0.6074 - val_loss: 0.6107 - val_accuracy: 0.5835
Epoch 49/100
321/321 [==============================] - 0s 973us/step - loss: 0.6020 - accuracy: 0.6097 - val_loss: 0.6081 - val_accuracy: 0.5862
Epoch 50/100
321/321 [==============================] - 0s 973us/step - loss: 0.6014 - accuracy: 0.6078 - val_loss: 0.6214 - val_accuracy: 0.5770
Epoch 51/100
321/321 [==============================] - 0s 973us/step - loss: 0.6023 - accuracy: 0.6093 - val_loss: 0.6011 - val_accuracy: 0.5952
Epoch 52/100
321/321 [==============================] - 0s 973us/step - loss: 0.6028 - accuracy: 0.6094 - val_loss: 0.6013 - val_accuracy: 0.5950
Epoch 53/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6022 - accuracy: 0.6079 - val_loss: 0.6158 - val_accuracy: 0.5770
Epoch 54/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6019 - accuracy: 0.6103 - val_loss: 0.6080 - val_accuracy: 0.5862
Epoch 55/100
321/321 [==============================] - 0s 973us/step - loss: 0.6020 - accuracy: 0.6095 - val_loss: 0.6180 - val_accuracy: 0.5775
Epoch 56/100
321/321 [==============================] - 0s 973us/step - loss: 0.6018 - accuracy: 0.6099 - val_loss: 0.6106 - val_accuracy: 0.5842
Epoch 57/100
321/321 [==============================] - 0s 973us/step - loss: 0.6022 - accuracy: 0.6078 - val_loss: 0.6232 - val_accuracy: 0.5740
Epoch 58/100
321/321 [==============================] - 0s 973us/step - loss: 0.6017 - accuracy: 0.6099 - val_loss: 0.6155 - val_accuracy: 0.5788
Epoch 59/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6026 - accuracy: 0.6119 - val_loss: 0.6150 - val_accuracy: 0.5775
Epoch 60/100
321/321 [==============================] - 0s 973us/step - loss: 0.6014 - accuracy: 0.6092 - val_loss: 0.5982 - val_accuracy: 0.6012
Epoch 61/100
321/321 [==============================] - 0s 973us/step - loss: 0.6025 - accuracy: 0.6087 - val_loss: 0.6022 - val_accuracy: 0.5947
Epoch 62/100
321/321 [==============================] - 0s 973us/step - loss: 0.6017 - accuracy: 0.6099 - val_loss: 0.6265 - val_accuracy: 0.5735
Epoch 63/100
321/321 [==============================] - 0s 899us/step - loss: 0.6019 - accuracy: 0.6099 - val_loss: 0.6172 - val_accuracy: 0.5775
Epoch 64/100
321/321 [==============================] - 0s 982us/step - loss: 0.6018 - accuracy: 0.6099 - val_loss: 0.6116 - val_accuracy: 0.5815
Epoch 65/100
321/321 [==============================] - 0s 969us/step - loss: 0.6015 - accuracy: 0.6099 - val_loss: 0.6230 - val_accuracy: 0.5738
Epoch 66/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6019 - accuracy: 0.6094 - val_loss: 0.6058 - val_accuracy: 0.5870
Epoch 67/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6019 - accuracy: 0.6103 - val_loss: 0.6250 - val_accuracy: 0.5723
Epoch 68/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6015 - accuracy: 0.6109 - val_loss: 0.6129 - val_accuracy: 0.5790
Epoch 69/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6016 - accuracy: 0.6099 - val_loss: 0.6061 - val_accuracy: 0.5867
Epoch 70/100
321/321 [==============================] - 0s 2ms/step - loss: 0.6031 - accuracy: 0.6084 - val_loss: 0.5999 - val_accuracy: 0.5980
Epoch 71/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6020 - accuracy: 0.6080 - val_loss: 0.6065 - val_accuracy: 0.5862
Epoch 72/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6015 - accuracy: 0.6097 - val_loss: 0.6193 - val_accuracy: 0.5745
Epoch 73/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6024 - accuracy: 0.6081 - val_loss: 0.6183 - val_accuracy: 0.5753
Epoch 74/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6017 - accuracy: 0.6094 - val_loss: 0.6165 - val_accuracy: 0.5778
Epoch 75/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6016 - accuracy: 0.6091 - val_loss: 0.6008 - val_accuracy: 0.5955
Epoch 76/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6021 - accuracy: 0.6094 - val_loss: 0.6235 - val_accuracy: 0.5733
Epoch 77/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6020 - accuracy: 0.6083 - val_loss: 0.6178 - val_accuracy: 0.5773
Epoch 78/100
321/321 [==============================] - 0s 973us/step - loss: 0.6016 - accuracy: 0.6099 - val_loss: 0.6232 - val_accuracy: 0.5715
Epoch 79/100
321/321 [==============================] - 0s 973us/step - loss: 0.6024 - accuracy: 0.6052 - val_loss: 0.6262 - val_accuracy: 0.5705
Epoch 80/100
321/321 [==============================] - 0s 973us/step - loss: 0.6022 - accuracy: 0.6050 - val_loss: 0.6150 - val_accuracy: 0.5785
Epoch 81/100
321/321 [==============================] - 0s 973us/step - loss: 0.6011 - accuracy: 0.6111 - val_loss: 0.6177 - val_accuracy: 0.5755
Epoch 82/100
321/321 [==============================] - 0s 973us/step - loss: 0.6025 - accuracy: 0.6087 - val_loss: 0.6124 - val_accuracy: 0.5783
Epoch 83/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6018 - accuracy: 0.6090 - val_loss: 0.6107 - val_accuracy: 0.5833
Epoch 84/100
321/321 [==============================] - 0s 973us/step - loss: 0.6025 - accuracy: 0.6102 - val_loss: 0.6110 - val_accuracy: 0.5800
Epoch 85/100
321/321 [==============================] - 0s 973us/step - loss: 0.6018 - accuracy: 0.6094 - val_loss: 0.6077 - val_accuracy: 0.5845
Epoch 86/100
321/321 [==============================] - 0s 973us/step - loss: 0.6016 - accuracy: 0.6069 - val_loss: 0.6109 - val_accuracy: 0.5798
Epoch 87/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6020 - accuracy: 0.6092 - val_loss: 0.6117 - val_accuracy: 0.5798
Epoch 88/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6021 - accuracy: 0.6089 - val_loss: 0.6105 - val_accuracy: 0.5808
Epoch 89/100
321/321 [==============================] - 0s 973us/step - loss: 0.6020 - accuracy: 0.6063 - val_loss: 0.6190 - val_accuracy: 0.5753
Epoch 90/100
321/321 [==============================] - 0s 973us/step - loss: 0.6022 - accuracy: 0.6083 - val_loss: 0.6211 - val_accuracy: 0.5740
Epoch 91/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6023 - accuracy: 0.6058 - val_loss: 0.6117 - val_accuracy: 0.5785
Epoch 92/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6019 - accuracy: 0.6077 - val_loss: 0.6200 - val_accuracy: 0.5740
Epoch 93/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6014 - accuracy: 0.6078 - val_loss: 0.6230 - val_accuracy: 0.5735
Epoch 94/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6018 - accuracy: 0.6087 - val_loss: 0.6113 - val_accuracy: 0.5810
Epoch 95/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6019 - accuracy: 0.6086 - val_loss: 0.6203 - val_accuracy: 0.5755
Epoch 96/100
321/321 [==============================] - 0s 1ms/step - loss: 0.6013 - accuracy: 0.6088 - val_loss: 0.6273 - val_accuracy: 0.5693
Epoch 97/100
321/321 [==============================] - 0s 925us/step - loss: 0.6019 - accuracy: 0.6071 - val_loss: 0.6023 - val_accuracy: 0.5927
Epoch 98/100
321/321 [==============================] - 0s 973us/step - loss: 0.6023 - accuracy: 0.6072 - val_loss: 0.6093 - val_accuracy: 0.5810
Epoch 99/100
321/321 [==============================] - 0s 925us/step - loss: 0.6012 - accuracy: 0.6091 - val_loss: 0.6018 - val_accuracy: 0.5937
Epoch 100/100
321/321 [==============================] - 0s 973us/step - loss: 0.6015 - accuracy: 0.6092 - val_loss: 0.6255 - val_accuracy: 0.5710

Answer 1

要么你的训练数据有问题，要么模型太小。

通过查看损失 - 这显然根本没有改变 - 我会说问题在于模型的大小。尝试在密集层中添加更多神经元。

Answer 2

您的模型不够大，无法处理数据。
因此，请尝试增加您的模型大小。
但是，增加模型大小会使其更容易受到过度拟合的影响，但使用一些 Dropout 层可以解决问题。

model = Sequential()
model.add(Input(shape=(12, 2)))
model.add(Dense(48, activation='relu'))
model.add(Dropout(0.20))
model.add(Dense(32, activation='relu'))
model.add(Dense(24, activation='relu'))
model.add(Dropout(0.15))
model.add(Dense(16, activation='relu'))
model.add(Dropout(0.1))
model.add(Dense(8, activation='relu'))
model.add(Dense(1, activation='sigmoid'))

此外，较低的学习率可以让模型获得更好的准确性和损失。
您可以在编译时调整 Adam 优化器的学习率，

model.compile(keras.optimizers.Adam(lr=0.0002), loss=keras.losses.binary_crossentropy, metrics=['accuracy'])

Adam优化器的学习率默认值为0.001