'''
机器学习系列策略
K-Means聚类择时
根据K-Means在历史一段时间的聚类结果，对当天的属性进行分类
并根据分类结果所属的类别判断买入/卖出
'''
from sklearn.externals import joblib
from features import gen_attrs#初始化账户
def initialize(account):account.security = '000300.SH'def handle_data(account,data):time = get_last_datetime().strftime("%Y%m%d")log.info(time)#获取证券过去20日的收盘价数据history = get_price(account.security,'20050101',time,'1d',\['open','close','low','high','avg_price','prev_close','volume'],False,'pre')# history = data.attribute_history(account.security, ['open','close','low','high','avg_price','prev_close','volume'], 34, '1d')
    log.info(history)# 计算当天属性值attrs, logreturn = gen_attrs(history, 33)# state = 1model = joblib.load('kmeans/kmeans.m')min_max_scaler = joblib.load('kmeans/std.m')attrs = min_max_scaler.transform(attrs)state = model.predict(attrs)[-1]if state in [0,1,7]:# if state in [1, 5, 10] and g.is_pos == 0:# order_value('510300.OF',account.cash)order_value('159919.OF',account.cash)# order_value('510050.OF',account.cash)log.info("买入 %s" % (account.security))# g.is_pos = 1elif state not in [0,1,7]:# elif state not in [1, 5, 10] and g.is_pos == 1:# order_target('510300.OF',0)order_target('159919.OF',0)# order_target('510050.OF',0)log.info("卖出 %s" % (account.security))# g.is_pos = 0

from sklearn.externals import joblib
from sklearn.cluster import KMeans from sklearn.preprocessing import MinMaxScaler import numpy as np import talib from scipy.stats import boxcox import features # history_price = get_price('000300.SH', 'None', '2013-12-31', '1d',['open','close','low','high','avg_price','prev_close','volume'], False, None, 2500,is_panel=1) history_price = get_price('000300.SH', '2005-01-04', '2013-12-31', '1d',['open','close','low','high','avg_price','prev_close','volume'], False, None) close = history_price.loc[:, 'close'] prev_close = history_price.loc[:, 'prev_close'] high = history_price.loc[:, 'high'] low = history_price.loc[:, 'low'] train_X, train_logreturn = features.gen_attrs(history_price, 33) min_max_scaler = MinMaxScaler() train_X = min_max_scaler.fit_transform(train_X) joblib.dump(min_max_scaler, 'std.m') kmeans = KMeans(n_clusters=8, random_state=0, n_init=50, max_iter=1000).fit(train_X) joblib.dump(kmeans, 'kmeans.m')