自己做了个斗地主的预测程序，几个月之前做的，记不太清楚了，在这里分享一下，内容简单有不懂的加Q：271051631交流：

1、代码

下面是ANN部分的核心代码：

# -*- coding: UTF-8 -*-
import numpy as np
import tensorflow as tf
from GameData import getdata
from keras import Sequential
from keras.layers import Dense,Convolution2D,MaxPooling2D,Flatten
from keras.optimizers import Adam
a,b=getdata('GameData0-21.txt')#默认的序号是0，我这里使用了21万条斗地主游戏记录
aa=np.array(a)
bb=np.array(b)
model=Sequential()
model.add(Dense(100,activation="relu"))
model.add(Dense(100,activation="relu"))
model.add(Dense(100,activation="relu"))
model.add(Dense(100,activation="relu"))
model.add(Dense(100,activation="relu"))
model.add(Dense(100,activation="relu"))
model.add(Dense(100,activation="relu"))
model.add(Dense(100,activation="relu"))
model.add(Dense(45,activation="relu"))adam=Adam(lr=1e-4)
model.compile(optimizer=adam,loss="mse",metrics=["accuracy"])
model.fit(aa,bb,batch_size=64,verbose=2,epochs=5)model.summary()
model.save("cnnddz0-21.h5")
aa,bb=getdata("test01.dat")###获取预测数据a=np.array(aa)
b=np.array(bb)
py=[]
for i in range(len(a)):x=a[i].reshape(1,48)y=b[i].reshape(1,45)py.append(model.predict(x))loss,acc=model.evaluate(x,y)print(loss,acc)

2、预测结果

计算时间有点长啊，训练时候的精度达到0.34，预测我写的不好，好久了不爱改了。

Using TensorFlow backend.
Epoch 1/5- 110s - loss: 0.3342 - accuracy: 0.3313
Epoch 2/5- 115s - loss: 0.3055 - accuracy: 0.3444
Epoch 3/5- 117s - loss: 0.3026 - accuracy: 0.3452
Epoch 4/5- 121s - loss: 0.3013 - accuracy: 0.3473
Epoch 5/5- 115s - loss: 0.3005 - accuracy: 0.3481
Model: "sequential_1"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
dense_1 (Dense)              (None, 100)               4900      
_________________________________________________________________
dense_2 (Dense)              (None, 100)               10100     
_________________________________________________________________
dense_3 (Dense)              (None, 100)               10100     
_________________________________________________________________
dense_4 (Dense)              (None, 100)               10100     
_________________________________________________________________
dense_5 (Dense)              (None, 100)               10100     
_________________________________________________________________
dense_6 (Dense)              (None, 100)               10100     
_________________________________________________________________
dense_7 (Dense)              (None, 100)               10100     
_________________________________________________________________
dense_8 (Dense)              (None, 100)               10100     
_________________________________________________________________
dense_9 (Dense)              (None, 45)                4545      
=================================================================
Total params: 80,145
Trainable params: 80,145
Non-trainable params: 0
_________________________________________________________________1/1 [==============================] - 0s 34ms/step
0.2668555676937103 0.01/1 [==============================] - 0s 1ms/step
0.2698018252849579 0.01/1 [==============================] - 0s 1ms/step
0.28025326132774353 0.01/1 [==============================] - 0s 1000us/step
0.28360310196876526 0.01/1 [==============================] - 0s 1ms/step
0.2639198303222656 0.01/1 [==============================] - 0s 0us/step
0.24844811856746674 0.01/1 [==============================] - 0s 1000us/step
0.24210020899772644 0.01/1 [==============================] - 0s 0us/step
0.24342185258865356 0.01/1 [==============================] - 0s 1000us/step
0.2314874678850174 0.01/1 [==============================] - 0s 1ms/step
0.2442777007818222 0.01/1 [==============================] - 0s 0us/step
0.07064052671194077 0.0

3、游戏数据说明

训练数据说明：

每条记录都是START开始 END结束

4、存在的问题

使用21万条记录训练5次，ANN的精度在0.3，精度还不是很高，可能是程序写的不是很好，CNN我也试过，感觉与ANN差不多。

5、对数据的利用

可以做个预测的模块放到经常玩的游戏里面，我在疫情期间分析过波克斗地主，数据基本都已经拿到了，只是没有动力去做了。