DFN.zip_dfn

# -*- coding: utf-8 -*- """ Created on Tue Sep 4 17:15:10 2018 @author: lenovo """ import matplotlib.pyplot as plt import numpy as np import pandas as pd import datetime import math from datetime import timedelta close = pd.read_csv('./close.csv') #收盘价 ev = pd.read_csv('./ev.csv') #市值 pb_lf = pd.read_csv('./pb_lf.csv') #市净率 risk_beta120 = pd.read_csv('./risk_beta120.csv') #beta tech_revs120 = pd.read_csv('./tech_revs120.csv') #6月动量 hs300 = pd.read_csv('./hs300.csv') #沪深300 def regular(data): '''数据初步标准化，索引改为日期''' data.index = data.iloc[:, 0] data.drop('Unnamed: 0',axis=1, inplace=True) data.index = pd.to_datetime(data.index) return data close = regular(close) ev = regular(ev) pb_lf = regular(pb_lf) risk_beta120 = regular(risk_beta120) tech_revs120 = regular(tech_revs120) hs300 = regular(hs300) def end_m(data): return data.groupby(lambda x: x.year*100 + x.month).tail(1) '''计算日对数收益率''' yield_day = close/close.shift(1) yield_day = yield_day.fillna(1) yield_rate_day = np.log(yield_day) yield_rate_month = yield_rate_day.resample('BM', how='sum') #月对数收益率 #yield_rate_day.index = pd.to_datetime(yield_rate_day.index) vol_year = yield_rate_day.resample('BM', how='std') * math.sqrt(252) #年波动率 hs300_yield = hs300/hs300.shift(1) hs300_day = hs300_yield.fillna(1) hs300_day = np.log(hs300_day) #hs300_day.index = pd.to_datetime(hs300_day.index) hs300_month = hs300_day.resample('BM', how='sum') sharpe_m = yield_rate_month.sub(hs300_month.iloc[:,0], axis=0) * 12 / vol_year sharpe_m = sharpe_m.replace([float('inf'), float('-inf')], 0) #夏普率 sharpe_m[np.abs(sharpe_m)>50]=0 """市值处理 initial_day = datetime.datetime(2010, 2, 1) #数据从2010年1月开始 def add_month(date_time): '''让时间的月份加一''' if date_time.month==12: date = datetime.datetime(date_time.year+1, 1, date_time.day) else: date = datetime.datetime(date_time.year, date_time.month+1, date_time.day) return date def month(data): '''取出月末数据''' data_1 = pd.DataFrame() compare_time = initial_day for i in range(data.shape[0]): # date_time = datetime.datetime.strptime(data.index[i],'%Y-%m-%d') #将str转换为时间格式 # if i%50 == 0: # print(i) if data.index[i]>= compare_time: #当这个月结束时 # print(i) compare_time = add_month(compare_time) #更新比较日期为下个月1号 data_1 = data_1.append(data.iloc[i-1]) # 将月末的数据添入表格当中 return data_1 """ ev_month = end_m(ev) ev_max = max(ev_month.max()) ev_month = ev_month / ev_max #归一化市值 #pb_lf.index = pd.to_datetime(pb_lf.index) #账面市值比 pb_lf = 1 / pb_lf pb_lf_month = end_m(pb_lf) #risk_beta120.index = pd.to_datetime(risk_beta120.index) #tech_revs120.index = pd.to_datetime(tech_revs120.index) risk_beta120_month = end_m(risk_beta120) #120日beta tech_revs120_month = end_m(tech_revs120) #六个月动量 dataset = pd.DataFrame(np.zeros(shape=[(ev_month.shape[0]-1)*300,8]), columns=('ev', 'pb_lf', 'beta', '6month', 'vol', '1month', 'sharpe', 'next_m')) for i in range(ev_month.shape[0]-1): for j in range(300): dataset.iloc[i*300+j][0] = ev_month.iloc[i][j] dataset.iloc[i*300+j][1] = pb_lf_month.iloc[i][j] dataset.iloc[i*300+j][2] = risk_beta120_month.iloc[i][j] dataset.iloc[i*300+j][3] = tech_revs120_month.iloc[i][j] dataset.iloc[i*300+j][4] = vol_year.iloc[i][j] dataset.iloc[i*300+j][5] = yield_rate_month.iloc[i][j] dataset.iloc[i*300+j][6] = sharpe_m.iloc[i][j] dataset.iloc[i*300+j][7] = yield_rate_month.iloc[i+1][j] if i%50 == 0: print(i) #dataset1 = dataset.dropna(axis=0) dataset.to_csv('dataset.csv') #用之前的数据做训练集，把空值数据直接去掉 data_train = dataset.iloc[:(ev_month.shape[0]-7)*300].dropna(axis=0) #用最后半年的数据做测试集，可以做6组，其中把空值附为0 data_test = dataset.iloc[(ev_month.shape[0]-7)*300:].fillna(0) data_train.to_csv('train.csv') data_test.to_csv('test.csv') ''' 模型预测 ''' import tensorflow as tf import tensorflow.contrib.eager as tfe tf.enable_eager_execution() def parse_csv(line): '''将数据一条一条读进来，前7项为x，最后1项为y''' example_defaults = [[0], [0.], [0.], [0.], [0.], [0.], [0.], [0.], [0.]] # sets field types parsed_line = tf.decode_csv(line, example_defaults) # First 4 fields are features, combine into single tensor features = tf.reshape(parsed_line[1:-1], shape=(7,)) #因为第一列为索引，不要了 # Last field is the label label = tf.reshape(parsed_line[-1], shape=()) return features, label train_dataset = tf.data.TextLineDataset('train.csv') train_dataset = train_dataset.skip(1) # skip the first header row train_dataset = train_dataset.map(parse_csv) # parse each row train_dataset = train_dataset.shuffle(buffer_size=24000) # randomize train_dataset = train_dataset.batch(128) # View a single example entry from a batch features, label = iter(train_dataset).next() #print("example features:", features[0]) #print("example label:", label[0]) #构建模型 '''这里可以设置模型参数，先暂定为两层隐藏层，每层10个神经元''' model = tf.keras.Sequential([ tf.keras.layers.Dense(10, activation="relu", input_shape=(7,)), # input shape required tf.keras.layers.Dense(10, activation="relu"), tf.keras.layers.Dense(1) ]) def loss(model, x, y): '''定义损失函数，用均方误差''' # print(x) # print(2) # print(y) y_ = model(x) # print(3) # print(y_) return tf.reduce_mean(tf.square(y_ - y)) # return tf.losses.sparse_softmax_cross_entropy(labels=y, logits=y_) def grad(model, inputs, targets): '''定义参数更新''' with tf.GradientTape() as tape: loss_value = loss(model, inputs, targets) return tape.gradient(loss_value, model.variables) #优化器 optimizer = tf.train.GradientDescentOptimizer(learning_rate=0.01) #放置loss train_loss_results = [] #train_accuracy_results = [] #训练代数 num_epochs = 201 for epoch in range(num_epochs): epoch_loss_avg = tfe.metrics.Mean() # epoch_accuracy = tfe.metrics.Accuracy() # Training loop - using batches of 32 for x, y in train_dataset: # Optimize the model grads = grad(model, x, y) optimizer.apply_gradients(zip(grads, model.variables), global_step=tf.train.get_or_create_global_step()) # Track progress epoch_loss_avg(loss(model, x, y)) # add current batch loss # compare predicted label to actual label # epoch_accuracy(tf.argmax(model(x), axis=1, output_type=tf.int32), y) # end epoch train_loss_results.append(epoch_loss_avg.result()) # train_accuracy_results.append(epoch_accuracy.result()) if epoch % 50 == 0: print("Epoch {:03d}: Loss: {:.3f}".format(epoch,epoch_loss_avg.result())) # epoch_accuracy.result())) #绘制损失函数图像 fig, axes = plt.subplots(1, sharex=True, figsize=(8, 4)) fig.suptitle('Training Metrics') axes.set_ylabel("Loss", fontsize=14) axes.plot(train_loss_results) #axes[1].set_ylabel("Accuracy", fontsize=14) #axes[1].set_xlabel("Epoch", fontsize=14) #axes[1].plot(train_accuracy_results) plt.show() '''构建测试集''' test_dataset = tf.data.TextLineDataset('test.csv') test_dataset = test_dataset.skip(1) # skip header row test_dataset = test_dataset.map(parse_csv)