def save(filename, title, prices, returns, decisions):
padding = zeros(len(prices) - len(returns))
returns = append(padding, returns)
decisions = append(padding, decisions)
x_axis = range(len(prices))
# Two subplots, the axes array is 1-d
nplots = 4
f, axarr = plt.subplots(nplots, sharex=True)
axarr[0].plot(x_axis, array(prices) / prices[0])
axarr[0].set_ylabel('Prices')
axarr[0].set_title(title)
axarr[1].plot(x_axis, wealth(returns))
axarr[1].set_ylabel('Wealth')
#axarr[2].plot(x_axis, sharpe(returns))
#axarr[2].set_ylabel('Sharpe')
axarr[2].plot(x_axis, returns)
axarr[2].set_ylabel('Returns')
axarr[3].plot(x_axis, decisions)
axarr[3].set_ylabel('Decisions')
for i in xrange(nplots):
axarr[i].locator_params(axis='y', nbins=6)
axarr[nplots - 1].set_xlabel('Time step (t)')
plt.savefig(filename, bbox_inches='tight')
def run(self, trX, trY, tsX, tsY, maxiter, validating=False):
self.scale(trX, tsX)
weights = self.model.weights(trX, seed=1)
gradients = zeros(weights.shape)
trR = zeros(maxiter + 1)
tsR = zeros(maxiter + 1)
costs = []
for i in range(maxiter):
#rs = RandomState(i)
#r = rs.randint(0, trX.shape[0]/2)
#tempX = trX[r:, :]
#tempY = trY[r:]
gradient = self.model.grad(weights, trX, trY)
gradients += power(gradient, 2)
rate = 1. / sqrt(gradients)
weights = weights - rate * gradient
if validating:
costs.append(self.model.cost(weights, tsX, tsY))
# Calculate returns and decisions on test set.
returns, decisions = self.model.returns(weights, tsX, tsY)
s = sharpe(returns)
w = wealth(returns)
return returns, decisions
def run(self, trX, trY, tsX, tsY, maxiter, validating=False):
self.scale(trX, tsX)
weights = self.model.weights(trX, seed=1)
gradients = zeros(weights.shape)
trR = zeros(maxiter+1)
tsR = zeros(maxiter+1)
costs = []
for i in range(maxiter):
#rs = RandomState(i)
#r = rs.randint(0, trX.shape[0]/2)
#tempX = trX[r:, :]
#tempY = trY[r:]
gradient = self.model.grad(weights, trX, trY)
gradients += power(gradient, 2)
rate = 1. / sqrt(gradients)
weights = weights - rate*gradient
if validating:
costs.append(self.model.cost(weights, tsX, tsY))
# Calculate returns and decisions on test set.
returns, decisions = self.model.returns(weights, tsX, tsY)
s = sharpe(returns)
w = wealth(returns)
return returns, decisions
def save(filename, title, prices, returns, decisions):
padding = zeros(len(prices) - len(returns))
returns = append(padding, returns)
decisions = append(padding, decisions)
x_axis = range(len(prices))
# Two subplots, the axes array is 1-d
nplots = 4
f, axarr = plt.subplots(nplots, sharex=True)
axarr[0].plot(x_axis, array(prices) / prices[0])
axarr[0].set_ylabel("Prices")
axarr[0].set_title(title)
axarr[1].plot(x_axis, wealth(returns))
axarr[1].set_ylabel("Wealth")
# axarr[2].plot(x_axis, sharpe(returns))
# axarr[2].set_ylabel('Sharpe')
axarr[2].plot(x_axis, returns)
axarr[2].set_ylabel("Returns")
axarr[3].plot(x_axis, decisions)
axarr[3].set_ylabel("Decisions")
for i in xrange(nplots):
axarr[i].locator_params(axis="y", nbins=6)
axarr[nplots - 1].set_xlabel("Time step (t)")
plt.savefig(filename, bbox_inches="tight")
def train(self, window=100, lookback=5, slide=50, maxiter=100):
returns = array([])
decisions = array([])
self.dataset.reset(window=window, lookback=lookback, slide=slide)
trX, trY, vsX, vsY = self.dataset.gen()
while(self.dataset.can_gen()):
score = float('-inf')
best = None
for model in self.models:
self.model = model
r, _ = self.run(trX, trY, vsX, vsY, maxiter, validating=True)
w = wealth(r)[-1]
if w > score:
score = w
best = model
next_trX, next_trY, tsX, tsY = self.dataset.gen()
self.model = best
r, d = self.run(next_trX, next_trY, tsX, tsY, maxiter)
returns = append(returns, r)
decisions = append(decisions, d)
print "Wealth: %f\tModel: %s" % (wealth(r)[-1], self.model)
# Done with loop carry over to next iteration.
trX = next_trX
trY = next_trY
vsX = tsX
vsY = tsY
return returns, decisions
def train(self, window=100, lookback=5, slide=50, maxiter=100):
returns = array([])
decisions = array([])
self.dataset.reset(window=window, lookback=lookback, slide=slide)
trX, trY, vsX, vsY = self.dataset.gen()
while (self.dataset.can_gen()):
score = float('-inf')
best = None
for model in self.models:
self.model = model
r, _ = self.run(trX, trY, vsX, vsY, maxiter, validating=True)
w = wealth(r)[-1]
if w > score:
score = w
best = model
next_trX, next_trY, tsX, tsY = self.dataset.gen()
self.model = best
r, d = self.run(next_trX, next_trY, tsX, tsY, maxiter)
returns = append(returns, r)
decisions = append(decisions, d)
print "Wealth: %f\tModel: %s" % (wealth(r)[-1], self.model)
# Done with loop carry over to next iteration.
trX = next_trX
trY = next_trY
vsX = tsX
vsY = tsY
return returns, decisions
hidden = 10
model = Linear(delta=delta, lmb=lmb)
#model = Nonlinear(delta=delta, lmb=lmb, hidden=hidden)
data = Dataset(cov, [jnj, nvs, pfe])
trainer = AdaGradTrainer(data, model)
returns, decisions = trainer.train(
window=window, lookback=lookback, slide=slide, maxiter=40)
padding = zeros(len(jnj)-len(returns))
returns = append(padding, returns)
decisions = append(padding, decisions)
print "Wealth: ", wealth(returns)[-1]
x_axis = range(len(jnj))
# Two subplots, the axes array is 1-d
f, axarr = plt.subplots(2, sharex=True)
axarr[0].plot(x_axis, jnj / jnj[0], 'r', label='Prices')
axarr[0].plot(x_axis, wealth(returns), 'b', label='Wealth')
axarr[0].set_title('Performance')
axarr[0].legend(loc="upper left")
#axarr[1].plot(x_axis, sharpe(returns))
#axarr[1].set_title('Sharpe')
#axarr[3].plot(x_axis, returns)
#axarr[3].set_title('Returns')
axarr[1].plot(x_axis, decisions)
# filename = 'figures/synthetic_noside_single_w_%i_s_%i_l_%i_d_%f.pdf' % (
# window, slide, lookback, delta)
# title = 'Single Layer - Synthetic (No Side Information)'
# print "%s\tWealth: %f\tSharpe: %f" % (filename,
# wealth(returns)[-1], sharpe(returns)[-1])
# plotter.save(filename, title, series, returns, decisions)
for window in [200]:
for slide in [5]:
for lookback in [10]:
for delta in [0.001]:
models = []
for hidden in [4, 6, 8]:
for lmb in [0.0, 0.0001, 0.001, 0.01]:
models.append(
Nonlinear(delta=delta, lmb=lmb, hidden=hidden))
trainer = ValidatingTrainer(data, models)
returns, decisions = trainer.train(window=window,
lookback=lookback,
slide=slide,
maxiter=20)
filename = 'figures/synthetic_noside_multi_w_%i_s_%i_l_%i_d_%f.pdf' % (
window, slide, lookback, delta)
title = 'Multiple Layer - Synthetic (No Side Information)'
print "%s\tWealth: %f\tSharpe: %f" % (
filename, wealth(returns)[-1], sharpe(returns)[-1])
plotter.save(filename, title, series, returns, decisions)
model = Linear(delta=delta, lmb=lmb)
#model = Nonlinear(delta=delta, lmb=lmb, hidden=hidden)
data = Dataset(cov, [jnj, nvs, pfe])
trainer = AdaGradTrainer(data, model)
returns, decisions = trainer.train(window=window,
lookback=lookback,
slide=slide,
maxiter=40)
padding = zeros(len(jnj) - len(returns))
returns = append(padding, returns)
decisions = append(padding, decisions)
print "Wealth: ", wealth(returns)[-1]
x_axis = range(len(jnj))
# Two subplots, the axes array is 1-d
f, axarr = plt.subplots(2, sharex=True)
axarr[0].plot(x_axis, jnj / jnj[0], 'r', label='Prices')
axarr[0].plot(x_axis, wealth(returns), 'b', label='Wealth')
axarr[0].set_title('Performance')
axarr[0].legend(loc="upper left")
#axarr[1].plot(x_axis, sharpe(returns))
#axarr[1].set_title('Sharpe')
#axarr[3].plot(x_axis, returns)
#axarr[3].set_title('Returns')
axarr[1].plot(x_axis, decisions)
padding = zeros(len(mxe)-len(returns))
returns = append(padding, returns)
decisions = append(padding, decisions)
x_axis = range(len(mxe))
# Two subplots, the axes array is 1-d
nplots = 4
f, axarr = plt.subplots(nplots, sharex=True)
axarr[0].plot(x_axis, array(mxf)/mxf[0], 'r', label='MXF')
axarr[0].plot(x_axis, array(mxe)/mxe[0], 'b', label='MXE')
axarr[0].legend(loc=2)
axarr[0].set_ylabel('Prices')
axarr[0].set_title('Trading MXE & MXF (Side) with Cost %f' % (delta))
axarr[1].plot(x_axis, wealth(returns))
axarr[1].set_ylabel('Wealth')
#axarr[2].plot(x_axis, sharpe(returns))
#axarr[2].set_ylabel('Sharpe')
axarr[2].plot(x_axis, returns)
axarr[2].set_ylabel('Returns')
axarr[3].plot(x_axis, decisions)
axarr[3].set_ylabel('Decisions')
for i in xrange(nplots):
axarr[i].locator_params(axis='y', nbins=6)
axarr[nplots-1].set_xlabel('Time step (t)')
plt.draw()
fn = 'figures/mexico_cost_%f.pdf' % (delta)
returns = append(padding, returns)
decisions = append(padding, decisions)
x_axis = range(len(mxe))
# Two subplots, the axes array is 1-d
nplots = 4
f, axarr = plt.subplots(nplots, sharex=True)
axarr[0].plot(x_axis, array(mxf) / mxf[0], 'r', label='MXF')
axarr[0].plot(x_axis, array(mxe) / mxe[0], 'b', label='MXE')
axarr[0].legend(loc=2)
axarr[0].set_ylabel('Prices')
axarr[0].set_title('Trading MXE & MXF (Side) with Cost %f' %
(delta))
axarr[1].plot(x_axis, wealth(returns))
axarr[1].set_ylabel('Wealth')
#axarr[2].plot(x_axis, sharpe(returns))
#axarr[2].set_ylabel('Sharpe')
axarr[2].plot(x_axis, returns)
axarr[2].set_ylabel('Returns')
axarr[3].plot(x_axis, decisions)
axarr[3].set_ylabel('Decisions')
for i in xrange(nplots):
axarr[i].locator_params(axis='y', nbins=6)
axarr[nplots - 1].set_xlabel('Time step (t)')
plt.draw()
fn = 'figures/mexico_cost_%f.pdf' % (delta)
plt.savefig(fn, bbox_inches='tight')
# slide=slide, maxiter=20)
#
# filename = 'figures/synthetic_noside_single_w_%i_s_%i_l_%i_d_%f.pdf' % (
# window, slide, lookback, delta)
# title = 'Single Layer - Synthetic (No Side Information)'
# print "%s\tWealth: %f\tSharpe: %f" % (filename,
# wealth(returns)[-1], sharpe(returns)[-1])
# plotter.save(filename, title, series, returns, decisions)
for window in [200]:
for slide in [5]:
for lookback in [10]:
for delta in [0.001]:
models = []
for hidden in [4, 6, 8]:
for lmb in [0.0, 0.0001, 0.001, 0.01]:
models.append(Nonlinear(delta=delta, lmb=lmb,
hidden=hidden))
trainer = ValidatingTrainer(data, models)
returns, decisions = trainer.train(window=window, lookback=lookback,
slide=slide, maxiter=20)
filename = 'figures/synthetic_noside_multi_w_%i_s_%i_l_%i_d_%f.pdf' % (
window, slide, lookback, delta)
title = 'Multiple Layer - Synthetic (No Side Information)'
print "%s\tWealth: %f\tSharpe: %f" % (filename,
wealth(returns)[-1], sharpe(returns)[-1])
plotter.save(filename, title, series, returns, decisions)
trainer = AdaGradTrainer(data, model)
returns, decisions = trainer.train(
window=window, lookback=lookback, slide=slide, maxiter=40)
padding = zeros(len(series)-len(returns))
returns = append(padding, returns)
decisions = append(padding, decisions)
# Linear : 2,618,987,047.18
# Nonlin (15): 2,272,781,214.71
# Nonlin (10): 2,884,104,773.60
# Nonlin (5) : 1,232,632,146.82
# Linear:
print "Wealth: ", wealth(returns)[-1]
x_axis = range(len(series))
# Two subplots, the axes array is 1-d
f, axarr = plt.subplots(5, sharex=True)
axarr[0].plot(x_axis, series)
axarr[0].set_title('Prices')
axarr[1].plot(x_axis, wealth(returns))
axarr[1].set_title('Wealth')
axarr[2].plot(x_axis, sharpe(returns))
axarr[2].set_title('Sharpe')
axarr[3].plot(x_axis, returns)
axarr[3].set_title('Returns')
axarr[4].plot(x_axis, decisions)
axarr[4].set_title('Decisions')
