Question

鉴于以下代码，我正在尝试实时预测，因为数据库上有更多信息。我想知道是否有人建议提高结果的准确性，或者是否可以让模型继续学习，因为有更多信息可用，同时，它同时进行预测。

#inclusiones
import numpy
import matplotlib.pyplot as plt
import math
import sys
import time
import MySQLdb

from pandas import read_csv
from pandas import read_sql_query
from pandas import DataFrame
from keras.models import Sequential
from keras.layers import Dense
from keras.layers import LSTM
from keras.callbacks import TensorBoard
from sklearn.preprocessing import MinMaxScaler
from sklearn.metrics import mean_squared_error
from pytz import timezone
from datetime import datetime

#definimos zona horaria
est = timezone('UTC')

#definiciones db
sql_hn      = "********"
sql_p       = 3306
sql_uid     = "********"
sql_pwd     = "********"
sql_db      = "********"

#conexión db
conn = MySQLdb.connect(
            host    = sql_hn,
            port    = sql_p,
            user    = sql_uid,
            passwd  = sql_pwd,
            db      = sql_db
            )

#acomodamos la informacion
def acomodar_informacion(informacion, pasado=1):
    informacion_x, informacion_y = [], []
    for i in range(len(informacion)-pasado-1):
        a = informacion[i:(i+pasado), 0]
        informacion_x.append(a)
        informacion_y.append(informacion[i + pasado, 0])
    return numpy.array(informacion_x), numpy.array(informacion_y)


# fix random seed
numpy.random.seed(7)

# pedimos la información a db
#archivo                 = read_csv('dice_amplified/primeros_10_mil.csv', usecols=[0], engine='python')
archivo                 = read_sql_query("SELECT value FROM ( SELECT * FROM crawler ORDER BY round DESC LIMIT 30000) sub ORDER BY round ASC", conn)
informacion             = archivo.values
informacion             = informacion.astype('float32')

#normalización de la información
scaler                  = MinMaxScaler(feature_range=(0, 1))

#variables
verbose                 = 2
epochs                  = 200
ronda                   = 256
pasado                  = 1

#dividimos la información
aprender_size           = int(len(informacion) * 0.75)
evaluar_size            = len(informacion) - aprender_size
aprender, evaluar       = informacion[0:aprender_size,:], informacion[aprender_size:len(informacion),:]


#acomodamos la informacion
aprender_x, aprender_y  = acomodar_informacion(aprender, pasado)
evaluar_x, evaluar_y    = acomodar_informacion(evaluar, pasado)

#reshape
aprender_x              = numpy.reshape(aprender_x, (aprender_x.shape[0], 1, aprender_x.shape[1]))
evaluar_x               = numpy.reshape(evaluar_x, (evaluar_x.shape[0], 1, evaluar_x.shape[1]))

#creacion de la red con LSTM
model = Sequential()
model.add(LSTM(100, input_shape=(1, pasado)))
model.add(Dense(1))
model.compile(loss='mean_squared_error', optimizer='adam')
model.fit(  aprender_x,
            aprender_y,
            epochs=epochs,
            batch_size=ronda,
            validation_data=(evaluar_x, evaluar_y),
            verbose=verbose
            )

#volcar la informacion            
for i in range(len(aprender_x)):
    print("causa=%s, consecuencia=%s" % (aprender_x[i], aprender_y[i]))

print("*******************  PREDICCIONES  *******************")

while True:

    #pedimos la informacion
    archivo_xnew         = read_sql_query("SELECT value FROM crawler ORDER BY round DESC LIMIT 0,1", conn)
    informacion_xnew     = archivo_xnew.values
    informacion_xnew     = informacion_xnew.astype('float32')

    #limpiamos la cache
    conn.commit()

    #reshape
    informacion_xnew     = numpy.reshape(informacion_xnew, (informacion_xnew.shape[0], 1, informacion_xnew.shape[1]))

    #realizar prediccion
    ynew = model.predict(informacion_xnew)

    #volcar la informacion
    for i in range(len(informacion_xnew)):
        print("causa=%s, prediccion=%s" % (informacion_xnew[i], ynew[i]))

（1）有谁知道我如何才能进一步提高这个比率？

（2）最重要的是：有没有办法让它在进行预测时进一步学习？可以在最后一行代码中看到。

提前致谢！克里斯，

Keras + TensorFlow，在跑步时训练/学习

0 个答案: