/ml_test.py

import Blotter
import load_stock_data
import numpy
import pandas
import PortfolioAllocator as pa
import mlpy 

raw = load_stock_data.load_1sec_data(['MS'], '20120309')
tickers=raw.keys()
timeIndex=raw[tickers[0]].index
bid=load_stock_data.generate_indicator(raw,'average_bid_price')
offer=load_stock_data.generate_indicator(raw,'average_offer_price')
bid_size=load_stock_data.generate_indicator(raw, 'average_bid_size')
offer_size=load_stock_data.generate_indicator(raw,'average_offer_size')
long_order_flow=load_stock_data.generate_indicator(raw,'long_order_flow')
short_order_flow=load_stock_data.generate_indicator(raw,'short_order_flow')

training_data = pandas.DataFrame(index=timeIndex)
training_data['bid_size']=bid_size['MS']
training_data['offer_size']=offer_size['MS']
training_data['long_order_flow']=long_order_flow['MS']
training_data['short_order_flow']=short_order_flow['MS']

#one minute forward average return
mid = 0.5*(bid+offer)
lag = 60
fwd_return = numpy.log(mid.shift(-lag)/mid).fillna(0)*1000
fwd_return = fwd_return.values.flatten()
label = numpy.zeros(len(timeIndex))
label[fwd_return>0.5]=1
label[fwd_return<-0.5]=-1

svm = mlpy.libsvm.LibSvm(svm_type='epsilon_svr', kernel_type='rbf', gamma=0.001)

svm.learn(training_data[:11700].values, label[:11700])

output = pandas.DataFrame(timeIndex[:11700])
output['pred'] = svm.pred(training_data[11700:])
output['real'] = label[11700:]
output.to_csv('output.csv')