def testPolicy(self, symbol = "IBM", \
sd=dt.datetime(2009,1,1), \
ed=dt.datetime(2010,1,1), \
sv = 10000):
syms = [symbol]
dates = pd.date_range(sd, ed)
# Call indicators function to receive indicators with dates before starting range
normalAverage, bollBand, MACDInd, momentum, prices = plotIndicators(
sd - dt.timedelta(days=80), ed, syms)
# Get indicators for the given dates
bollBand = bollBand.loc[prices.index >= sd]
normalAverage = normalAverage.loc[prices.index >= sd]
MACDInd = MACDInd.loc[prices.index >= sd]
momentum = momentum.loc[prices.index >= sd]
prices = prices.loc[prices.index >= sd]
# Discretize values in the dataframs
bollBand = self.dfDiscretize(bollBand, steps=self.bins)
normalAverage = self.dfDiscretize(normalAverage, steps=self.bins)
MACDInd = self.dfDiscretize(MACDInd, steps=self.bins)
momentum = self.dfDiscretize(momentum, steps=self.bins)
firstState = (bollBand.iloc[0][symbol], normalAverage.iloc[0][symbol],
MACDInd.iloc[0][symbol], momentum.iloc[0][symbol])
dims = (self.bins, self.bins, self.bins, self.bins)
firstState = np.ravel_multi_index(firstState, dims=dims)
self.learner = ql.QLearner(num_states=self.bins**4,
num_actions=5,
rar=0.98,
verbose=False)
action = self.learner.querysetstate(firstState)
shares = 0
orders = pd.DataFrame(columns=["Orders"], index=prices.index)
for j in range(len(prices.index)):
# Buy allowed
if action == 0 and shares in [0, -500]:
shares += 500
orders.iloc[j]["Orders"] = 500
elif action == 1 and shares == -500:
shares += 1000
orders.iloc[j]["Orders"] = 1000
# sell allowed
elif action == 2 and shares in [0, 500]:
shares -= 500
orders.iloc[j]["Orders"] = -500
elif action == 3 and shares == 500:
shares -= 1000
orders.iloc[j]["Orders"] = -1000
else:
orders.iloc[j]["Orders"] = 0
nextState = np.ravel_multi_index(
(bollBand.iloc[j][symbol], normalAverage.iloc[j][symbol],
MACDInd.iloc[j][symbol], momentum.iloc[j][symbol]),
dims=dims)
action = self.learner.querysetstate(nextState)
return orders
def main(gen_plot=False):
of = "benchmark.csv"
sv = 100000
#Get Technical Values from indicators file
normalAverage, bollBand, MACDInd,momentum,prices = indicators.plotIndicators(gen_plot=gen_plot)
orders = ordersFile(normalAverage=normalAverage, bollBand=bollBand, MACDInd=MACDInd, momentum=momentum)
# Process orders
portvals = marketsim.compute_portvals(orders_file=of, start_val=sv)
#Print statements to determine start and end value
print portvals.head(1)
print portvals.tail(1)
of = "orders.csv"
#default name of file is orders.csv
portvals1 = marketsim.compute_portvals(orders_file=of, start_val=sv)
print portvals1.head(1)
print portvals1.tail(1)
if gen_plot:
df = pd.concat([portvals/portvals.iloc[0], portvals1/portvals1.iloc[0]], keys=['Benchmark', 'Rule-Based'], axis=1)
ax = df.plot(title="Rule Based Trader", fontsize=12,color=["Black","Blue"])
ax.set_xlabel("Date")
ax.set_ylabel("Price Normalized")
stock = 0
#Loop through orders to add vertical lines
for date,order in orders.iterrows():
if order.loc["Order"]=="SELL":
stock-=500
if stock==0:
ax.axvline(date.date(),color="Black")
else:
ax.axvline(date.date(), color="Red")
else:
stock+=500
if stock==0:
ax.axvline(date.date(),color="Black")
else:
ax.axvline(date.date(), color="Green")
plt.show()
def main(gen_plot=False):
normalAverage, bollBand, MACDInd, momentum, prices = indicators.plotIndicators(
gen_plot=gen_plot)
normalArray = normalAverage.as_matrix()
bollBandArray = bollBand.as_matrix()
MACDIndArray = MACDInd.as_matrix()
momentumArray = momentum.as_matrix()
indArrayTrain = np.concatenate(
(normalArray, bollBandArray, MACDIndArray, momentumArray), axis=1)
Y = prices.shift(-10) / prices - 1
YBUY = .03
YSELL = -.04
for i in range(Y.shape[0]):
#print Y.iloc[i]["IBM"]
if Y.iloc[i]["IBM"] > YBUY:
Y.iloc[i]["IBM"] = 1
elif Y.iloc[i]["IBM"] < YSELL:
Y.iloc[i]["IBM"] = -1
else:
Y.iloc[i]["IBM"] = 0
YArray = Y['IBM'].values
#Build Test Data
normalAverage, bollBand, MACDInd, momentum, prices = indicators.plotIndicators(
sd=dt.datetime(2010, 1, 4),
ed=dt.datetime(2010, 12, 31),
gen_plot=gen_plot)
normalArray = normalAverage.as_matrix()
bollBandArray = bollBand.as_matrix()
MACDIndArray = MACDInd.as_matrix()
momentumArray = momentum.as_matrix()
indArrayTest = np.concatenate(
(normalArray, bollBandArray, MACDIndArray, momentumArray), axis=1)
orderInfo = mlTrade(indArrayTrain, YArray, indArrayTest)
orders = generateFutureOrder(orderInfo, prices.index, name="Test.csv")
sv = 100000
of = "Test.csv"
portvals = marketsim.compute_portvals(orders_file=of, start_val=sv)
print portvals.head(1)
print portvals.tail(1)
of = "TestOrders.csv"
rule_based.ordersFile(sd = dt.datetime(2010,1,4), ed = dt.datetime(2010,12,31),normalAverage=normalAverage,\
bollBand=bollBand,MACDInd=MACDInd,momentum=momentum,filename=of)
portvals1 = marketsim.compute_portvals(orders_file=of, start_val=sv)
print portvals1.head(1)
print portvals1.tail(1)
of = "benchmarkTest.csv"
portvals2 = marketsim.compute_portvals(orders_file=of, start_val=sv)
print portvals2.head(1)
print portvals2.tail(1)
if gen_plot:
df = pd.concat([
portvals2 / portvals2.iloc[0], portvals1 / portvals1.iloc[0],
portvals / portvals.iloc[0]
],
keys=['Benchmark', 'Rule-Based', "ML-Based"],
axis=1)
ax = df.plot(title="ML Based Trader",
fontsize=12,
color=["Black", "Blue", "Green"])
ax.set_xlabel("Date")
ax.set_ylabel("Price Normalized")
plt.show()
def addEvidence(self, symbol = "IBM", \
sd=dt.datetime(2008,1,1), \
ed=dt.datetime(2009,1,1), \
sv = 10000):
syms = [symbol]
dates = pd.date_range(sd, ed)
#Call indicators function to receive indicators with dates before starting range
normalAverage, bollBand, MACDInd, momentum, prices = plotIndicators(
sd - dt.timedelta(days=80), ed, syms)
#Get indicators for the given dates
bollBand = bollBand.loc[prices.index >= sd]
normalAverage = normalAverage.loc[prices.index >= sd]
MACDInd = MACDInd.loc[prices.index >= sd]
momentum = momentum.loc[prices.index >= sd]
prices = prices.loc[prices.index >= sd]
#Discretize values in the dataframs
bins = self.bins
bollBand = self.dfDiscretize(bollBand, steps=self.bins)
normalAverage = self.dfDiscretize(normalAverage, steps=self.bins)
MACDInd = self.dfDiscretize(MACDInd, steps=self.bins)
momentum = self.dfDiscretize(momentum, steps=self.bins)
#loop varialbles
iterationMin = 3
iterationMax = 50
rewardPos = 100
rewardNeg = -2000
firstState = (bollBand.iloc[0][symbol], normalAverage.iloc[0][symbol],
MACDInd.iloc[0][symbol], momentum.iloc[0][symbol])
dims = (self.bins, self.bins, self.bins, self.bins)
firstState = np.ravel_multi_index(firstState, dims=dims)
returnArr = []
self.learner = ql.QLearner(num_states=self.bins**4,
num_actions=5,
verbose=False)
orders = pd.DataFrame(columns=["Cash", "Shares"], index=prices.index)
for i in range(iterationMax):
action = self.learner.querysetstate(firstState)
shares = 0
for j in range(len(prices.index)):
lastRet = shares * prices.iloc[j - 1][symbol]
#Buy allowed
if action == 0 and shares in [0, -500]:
shares += 500
elif action == 1 and shares == -500:
shares += 1000
#sell allowed
elif action == 2 and shares in [0, 500]:
shares -= 500
elif action == 3 and shares == 500:
shares -= 1000
ret = shares * prices.iloc[j][symbol]
daily = ret / lastRet - 1
if j == 0:
cash = sv
else:
cash = orders.iloc[j - 1]["Cash"] + ret
reward = rewardPos if daily > 0 else rewardNeg
orders.iloc[j] = pd.Series({"Cash": cash, "Shares": shares})
nextState = np.ravel_multi_index(
(bollBand.iloc[j][symbol], normalAverage.iloc[j][symbol],
MACDInd.iloc[j][symbol], momentum.iloc[j][symbol]),
dims=dims)
action = self.learner.query(nextState, reward)
returnArr.append(cash)
if i > iterationMin - 1:
converge = 0.1 * returnArr[i - 1]
if (returnArr[i] - converge) < returnArr[i - 1] < (
returnArr[i] + converge):
break
def main(gen_plot=False):
#Get values for Indicators
normalAverage, bollBand, MACDInd, momentum,prices = indicators.plotIndicators(gen_plot=gen_plot)
normalArray = normalAverage.as_matrix()
bollBandArray = bollBand.as_matrix()
MACDIndArray = MACDInd.as_matrix()
momentumArray = momentum.as_matrix()
indArray = np.concatenate((normalArray,bollBandArray,MACDIndArray,momentumArray),axis=1)
#Create future Y predictions
Y = prices.shift(-10)/prices - 1
YBUY =.03
YSELL=-.02
#Convert predictions to 1, 0, -1 for classifier
for i in range(Y.shape[0]):
#print Y.iloc[i]["IBM"]
if Y.iloc[i]["IBM"]>YBUY:
Y.iloc[i]["IBM"]=1
elif Y.iloc[i]["IBM"]<YSELL:
Y.iloc[i]["IBM"]=-1
else:
Y.iloc[i]["IBM"] = 0
YArray = Y['IBM'].values
orderInfo = mlTrade(indArray,YArray)
#create orders file
orders = generateOrder(orderInfo,prices.index)
sv =100000
of = "MLBased.csv"
portvals = marketsim.compute_portvals(orders_file=of, start_val=sv)
print portvals.head(1)
print portvals.tail(1)
#orders.csv contains data for the rule based trader
of = "orders.csv"
portvals1 = marketsim.compute_portvals(orders_file=of, start_val=sv)
print portvals1.head(1)
print portvals1.tail(1)
of = "benchmark.csv"
portvals2 = marketsim.compute_portvals(orders_file=of, start_val=sv)
print portvals2.head(1)
print portvals2.tail(1)
#generate the plot
if gen_plot:
df = pd.concat([portvals2/portvals2.iloc[0], portvals1/portvals1.iloc[0],portvals/portvals.iloc[0]], keys=['Benchmark', 'Rule-Based',"ML-Based"], axis=1)
ax = df.plot(title="ML Based Trader", fontsize=12,color=["Black","Blue","Green"])
ax.set_xlabel("Date")
ax.set_ylabel("Price Normalized")
stock = 0
for date,order in orders.iterrows():
if order.loc["Order"]=="SELL":
stock-=500
if stock==0:
ax.axvline(date.date(),color="Black")
else:
ax.axvline(date.date(), color="Red")
else:
stock+=500
if stock==0:
ax.axvline(date.date(),color="Black")
else:
ax.axvline(date.date(), color="Green")
plt.show()