java调用保存好的神经网络模型

1.训练模型并保存

import pandas as pd
from sklearn.preprocessing import LabelEncoder
import numpy as np
from tensorflow.python.keras.utils.np_utils import to_categorical
from tensorflow.keras.models import Sequential
from tensorflow.keras.models import Model as KerasModel
from tensorflow.keras.layers import Input, Dense, Activation, Reshape,Dropout
from tensorflow.keras.layers import Concatenate
from tensorflow.keras.layers import Embedding
from tensorflow.keras.callbacks import ModelCheckpoint
from tensorflow. keras.optimizers import Adam
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import OneHotEncoder
import pickle 
df = pd.read_csv(r'C:\Users\张家豪\Desktop\keras\diamonds1.csv')
df = df[['carat','cut','color','clarity','depth','table','price','x','y','z']]
df.head()
allcols = list(df)
continuouscols = ['carat','depth','table','price','x','y','z']
collist = ['color','clarity','cut']
df.info()
for col in collist: 
    df[col] = df[col].astype(str)
dict_f = {}             #得到字段
for col in collist:
    key = df[col].unique().tolist()
    m = dict(zip(key, range(len(key))))
    dict_f[col]=m
    print(col,len(dict_f[col]))
for col in collist:     #字典展示
    print(dict_f[col])
for col in collist:       #字典转换
    df[col] = df[col].map(dict_f[col])
f_save = open(r'C:\Users\张家豪\Desktop\keras\dicts.pkl','wb')
pickle.dump(dict_f,f_save)
enc = OneHotEncoder()
enc.fit(df['cut'].values.reshape(-1,1))
train, valid = train_test_split(df, test_size = 0.2)
xtrain = train[['carat','color','clarity','depth','table','price','x','y','z']]
ytrain = train[['cut']]
xvalid = valid[['carat','color','clarity','depth','table','price','x','y','z']]
yvalid = valid[['cut']]
print(xtrain.shape)
print(xvalid.shape)
xtrain = xtrain.values
xvalid = xvalid.values
ytrain = enc.transform(ytrain.values.reshape(-1,1)).toarray()
yvalid = enc.transform(yvalid.values.reshape(-1,1)).toarray()
def split_features(X):
    X_list = []
    
    carat = X[..., [0]]     #取第一列 ,得到arry数组 （n，1）
    X_list.append(carat)
    
    color = X[..., [1]]
    X_list.append(color)

    clarity = X[..., [2]]
    X_list.append(clarity)

    depth = X[..., [3]]
    X_list.append(depth)

    table = X[..., [4]]
    X_list.append(table)

    price = X[..., [5]]
    X_list.append(price)

    x = X[..., [6]]
    X_list.append(x)

    y = X[..., [7]]
    X_list.append(y)

    z = X[..., [8]]
    X_list.append(z)

    return X_list
class NN_with_EntityEmbedding():
    def __init__(self, X_train, y_train, X_val, y_val):
        self.epochs = 20
        self.batch_size = 256
        self.lr = 1e-3
#         self.dropout_rate = 0.1
        self.__build_keras_model()   #初始化会调用这这个内部函数
        self.fit(X_train, y_train, X_val, y_val)#初始化会调用这个内部函数
        self.full_output = []#存放全连接层输入的特征
    
    def preprocessing(self, X):
        X_list = split_features(X)
        return X_list
    def __build_keras_model(self):
        input_carat = Input(shape=(1,))
        output_carat = Dense(1)(input_carat)

        input_color = Input(shape=(1,))
        output_color = Embedding(8, 6, name='color_embedding')(input_color)  #7是总共几个分类，6是向量的长度
        output_color = Reshape(target_shape=(6,))(output_color)

        input_clarity = Input(shape=(1,))
        output_clarity = Embedding(9, 6, name='clarity_embedding')(input_clarity)
        output_clarity = Reshape(target_shape=(6,))(output_clarity)

        input_depth = Input(shape=(1,))
        output_depth = Dense(1)(input_depth)

        input_table = Input(shape=(1,))
        output_table = Dense(1)(input_table)

        input_price = Input(shape=(1,))
        output_price = Dense(1)(input_price)

        input_x = Input(shape=(1,))
        output_x = Dense(1)(input_x)

        input_y = Input(shape=(1,))
        output_y = Dense(1)(input_y)

        input_z = Input(shape=(1,))
        output_z = Dense(1)(input_z)

        input_model = [input_carat, input_color, input_clarity,
                       input_depth, input_table, input_price,
                       input_x, input_y, input_z]

        output_embeddings = [output_carat, output_color, output_clarity,
                             output_depth, output_table, output_price,
                             output_x, output_y, output_z]

        output_model = Concatenate()(output_embeddings)
#         output_model = Dropout(self.dropout_rate)(Dense(1000, kernel_initializer="uniform")(output_model))
        output_model = Dense(1000, kernel_initializer="uniform")(output_model)
        output_model = Activation('relu')(output_model)
#         output_model = Dropout(self.dropout_rate)(Dense(500, kernel_initializer="uniform")(output_model))
        output_model =Dense(500, kernel_initializer="uniform")(output_model)
        output_model = Activation('relu')(output_model)
        output_model = Dense(6,activation='softmax')(output_model)

        self.model = KerasModel(inputs=input_model, outputs=output_model)    #注意这里的model成员
        optimizer = Adam(lr=self.lr)
        self.model.compile(loss='categorical_crossentropy', optimizer=optimizer)
    def fit(self, X_train, y_train, X_val, y_val):
        self.model.fit(self.preprocessing(X_train), y_train,
                       validation_data=(self.preprocessing(X_val), y_val),
                       epochs=self.epochs, batch_size=self.batch_size,
                       # callbacks=[self.checkpointer],
                       )
model = NN_with_EntityEmbedding(xtrain, ytrain, xvalid, yvalid)
xvalid = model.preprocessing(xvalid)
from sklearn.metrics import accuracy_score
y_hat = model.model.predict(xvalid)
acc = accuracy_score(yvalid.argmax(1),y_hat.argmax(1))
print(acc)
model.model.save(r'C:\Users\张家豪\Desktop\keras\sf.h5')

2.读取模型和字典

import pandas as pd
from sklearn.preprocessing import LabelEncoder
import numpy as np
from tensorflow.python.keras.utils.np_utils import to_categorical
from tensorflow.keras.models import Sequential
from tensorflow.keras.models import Model as KerasModel
from tensorflow.keras.layers import Input, Dense, Activation, Reshape,Dropout
from tensorflow.keras.layers import Concatenate
from tensorflow.keras.layers import Embedding
from tensorflow.keras.callbacks import ModelCheckpoint
from tensorflow. keras.optimizers import Adam
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import OneHotEncoder
import pickle
from tensorflow.keras.models import load_model
from sklearn.metrics import accuracy_score
import sys
print(sys.argv)
df = pd.read_csv(r'C:\Users\张家豪\Desktop\keras\diamonds2.csv')
df = df[['carat','cut','color','clarity','depth','table','price','x','y','z']]
df.head()
allcols = list(df)
continuouscols = ['carat','depth','table','price','x','y','z']
collist = ['color','clarity','cut']
df.info()
for col in collist: 
    df[col] = df[col].astype(str)
f_open = open(r'C:\Users\张家豪\Desktop\keras\dicts.pkl','rb')
dict_f = pickle.load(f_open)
print(dict_f)
for col in collist:       #字典转换
    df[col] = df[col].map(dict_f[col])
def p_color(df):
    line = df['color']
    if not pd.isnull(line):
        return line
    return '_'
def p_clarity(df):
    line = df['clarity']
    if not pd.isnull(line):
        return line
    return '_' 
df['color'] = df.apply(p_color,axis='columns')
df['clarity'] = df.apply(p_clarity,axis='columns')
df.color.replace('_',dict_f['color']['_'],inplace=True)
df.clarity.replace('_',dict_f['clarity']['_'],inplace=True)
def split_features(X):
    X_list = []
    
    carat = X[..., [0]]     #取第一列 ,得到arry数组 （n，1）
    X_list.append(carat)
    
    color = X[..., [1]]
    X_list.append(color)

    clarity = X[..., [2]]
    X_list.append(clarity)

    depth = X[..., [3]]
    X_list.append(depth)

    table = X[..., [4]]
    X_list.append(table)

    price = X[..., [5]]
    X_list.append(price)

    x = X[..., [6]]
    X_list.append(x)

    y = X[..., [7]]
    X_list.append(y)

    z = X[..., [8]]
    X_list.append(z)

    return X_list
class NN_with_EntityEmbedding():
    def __init__(self):
        self.epochs = 20
        self.batch_size = 256
        self.lr = 1e-3
#         self.dropout_rate = 0.1
        self.__build_keras_model()   #初始化会调用这这个内部函数
#         self.fit(X_train, y_train, X_val, y_val)#初始化会调用这个内部函数
#         self.full_output = []#存放全连接层输入的特征
    
    def preprocessing(self, X):
        X_list = split_features(X)
        return X_list
    def __build_keras_model(self):

        input_carat = Input(shape=(1,))
        output_carat = Dense(1)(input_carat)
        
        input_color = Input(shape=(1,))
        output_color = Embedding(8, 6, name='color_embedding')(input_color)  #7是总共几个分类，6是向量的长度
        output_color = Reshape(target_shape=(6,))(output_color)

        input_clarity = Input(shape=(1,))
        output_clarity = Embedding(9, 6, name='clarity_embedding')(input_clarity)
        output_clarity = Reshape(target_shape=(6,))(output_clarity)

        input_depth = Input(shape=(1,))
        output_depth = Dense(1)(input_depth)

        input_table = Input(shape=(1,))
        output_table = Dense(1)(input_table)

        input_price = Input(shape=(1,))
        output_price = Dense(1)(input_price)

        input_x = Input(shape=(1,))
        output_x = Dense(1)(input_x)

        input_y = Input(shape=(1,))
        output_y = Dense(1)(input_y)

        input_z = Input(shape=(1,))
        output_z = Dense(1)(input_z)

        input_model = [input_carat, input_color, input_clarity,
                       input_depth, input_table, input_price,
                       input_x, input_y, input_z]

        output_embeddings = [output_carat, output_color, output_clarity,
                             output_depth, output_table, output_price,
                             output_x, output_y, output_z]

        output_model = Concatenate()(output_embeddings)
#         output_model = Dropout(self.dropout_rate)(Dense(1000, kernel_initializer="uniform")(output_model))
        output_model = Dense(1000, kernel_initializer="uniform")(output_model)
        output_model = Activation('relu')(output_model)
#         output_model = Dropout(self.dropout_rate)(Dense(500, kernel_initializer="uniform")(output_model))
        output_model =Dense(500, kernel_initializer="uniform")(output_model)
        output_model = Activation('relu')(output_model)
        output_model = Dense(6,activation='softmax')(output_model)

        self.model = KerasModel(inputs=input_model, outputs=output_model)    #注意这里的model成员
        optimizer = Adam(lr=self.lr)
        self.model.compile(loss='categorical_crossentropy', optimizer=optimizer)
#     def fit(self, X_train, y_train, X_val, y_val):
#         self.model.fit(self.preprocessing(X_train), y_train,
#                        validation_data=(self.preprocessing(X_val), y_val),
#                        epochs=self.epochs, batch_size=self.batch_size,
#                        # callbacks=[self.checkpointer],
#                        )
model = NN_with_EntityEmbedding()
l = 'C:\\Users\\张家豪\\Desktop\\keras\\sf.h5'
model.model = load_model(l)
x=df[['carat','color','clarity','depth','table','price','x','y','z']]
print(x)
x=np.array(x)   #北京那个是x.values
print(x)
y=df.cut
x=model.preprocessing(x)
y_hat=model.model.predict(x)
acc = accuracy_score(y,y_hat.argmax(1))
print(acc)

3.java调用h5模型

package python;
import java.io.BufferedInputStream;
import java.io.BufferedReader;
import java.io.IOException;
import java.io.InputStreamReader;
import java.util.Scanner;

import org.python.antlr.PythonParser.test_return;
import org.python.util.PythonInterpreter;

import jnr.ffi.Struct.int16_t;

public class java_python {
	

	 public static void main(String[] args) {

         Scanner input = new Scanner(System.in);
         System.out.println("请输入需要传递给python的参数");
         String url = input.nextLine();
         String a = "jia";
//         System.out.println(integers);
//         String url="http://blog.csdn.net/thorny_v/article/details/61417386";
         String[] cmds = new String[]{"python", "C:\\Users\\张家豪\\Desktop\\keras\\test1.py",url,a};
//         System.out.println(cmds);
         System.out.println("调用python程序");
         Process pcs;
         try {
             pcs = Runtime.getRuntime().exec(cmds);
             pcs.waitFor();
             // 定义Python脚本的返回值
             String result = null;
             // 获取CMD的返回流
             BufferedInputStream in = new BufferedInputStream(pcs.getInputStream());
             // 字符流转换字节流
             BufferedReader br = new BufferedReader(new InputStreamReader(in));
             // 这里也可以输出文本日志
             String lineStr = null;
//             System.out.println(br.readLine());
             while ((lineStr = br.readLine()) != null) {
                 result = lineStr;
//                 System.out.println(br.readLine());
                 System.out.println(result);
             }
//
             // 关闭输入流
             br.close();
             in.close();


         } catch (IOException | InterruptedException e) {
             // TODO Auto-generated catch block
             e.printStackTrace();
         }



 }

}