def day_chart_datas():
"""
chart how more data lowers error
note its diff each time
"""
df = pd.read_csv('amd.csv', index_col='timestamp', parse_dates=True)
# 1949 / 16 = 120
r = range(0, len(df), 120)[1:]
train_errors = []
test_errors = []
for amount in r:
d = DayTrader(df=df[:amount])
clf, train_acc = d.decision_tree(debug=False)
test_acc = day_simulations((clf, train_acc), d)
train_errors.append(1.0 - train_acc)
test_errors.append(1.0 - test_acc)
print train_errors
print test_errors
plt.xticks(r)
plt.title('Decision Trees: Amount of Data and Accuracy')
plt.plot(r, train_errors, label="Training Error")
plt.plot(r, test_errors, label="Testing Error")
plt.legend()
plt.show()
def lda():
s = DayTrader()
X_trans = s.lda()
fig, ax = plt.subplots()
plt.title('Random Projection Reduction to Two Days on Swing Trading')
ax.set_xlabel('First Price')
ax.set_ylabel('Second Price')
print X_trans
plt.plot(X_trans)
# plt.scatter(X_trans[:, 0], X_trans[:, 1])
plt.show()
def ideal_prune():
""" around 3 to 5"""
accs = []
d = DayTrader()
r = range(0, 35)
for pruneval in r:
clf, train_acc = d.decision_tree(debug=False, prune=pruneval)
test_acc = day_simulations((clf, train_acc), d)
accs.append(1.0 - test_acc)
plt.xticks(r)
plt.title('Decision Trees: Ideal Prune Value')
plt.plot(r, accs, label="Error")
plt.legend()
plt.show()
def neural_with_pca():
"""
old results
in paper: an average accuracy of 46.28% and daily returns of 1.318%.
takes about 15 sec to run
>>> analysis.day_trials()
average accuracies:
0.4574418604651163
avg daily returns:
1.3216907197464711 %
in paper: 53.7% accuracy and 1.23% in daily returns on avg
takes about 2 min to run
>>> analysis.swing_several_sims()
# Avg accuracies and money:
0.4830140946873871
3.0788267587245786
"""
# analysis.day_trials()
# analysis.swing_several_sims()
s = DayTrader()
r = range(1, 21)
accuracies = []
for i in r:
acc = 0
for _ in range(2):
X = s.pca(n_components=i, X=s.train_xs)
testX = s.pca(n_components=i, X=s.test_xs)
acc += analysis.day_trials(X=X, testX=testX)[0]
accuracies.append(acc / 2.)
title = 'Neural Net After PCA for Day Trading'
fig, ax = plt.subplots()
plt.title(title)
ax.set_xlabel('Num of Components')
ax.set_ylabel('Accuracy')
plt.plot(r, accuracies)
plt.show()
def day_trials(outer_trials=2,
inner_trials=5,
kernel='poly',
degree=5,
neighbors=5,
prune_val=5,
X=None,
testX=None):
allmoneys = []
allboosts = []
for _ in range(outer_trials):
d = DayTrader()
boosts, moneys = [], []
for _ in range(0, inner_trials):
# print "*******BOOSTING"
# acc, money = day_simulations(d.svm(debug=False, kernel=kernel, degree=degree), d, debug=False)
# acc, money = day_simulations(d.knn(debug=False, neighbors=neighbors), d, debug=False)
acc, money = day_simulations(d.neural_network(debug=False,
X=X,
testX=testX),
d,
debug=False)
boosts.append(acc)
moneys.append(money)
# print "*************\n\naverage accuracies for :"
# print np.mean(boosts)
# print "avg money:", np.mean(moneys)
# print "avg daily returns:"
dr = (abs(np.mean(moneys))**(1.0 / len(d.test_xs)) - 1) * 100
if np.mean(moneys) < 0: dr *= -1
allmoneys.append(dr)
allboosts.append(np.mean(boosts))
# print dr,"%"
print "*************\n\naverage accuracies for :"
print np.mean(allboosts)
print "avg daily returns:"
print np.mean(allmoneys), "%"
return np.mean(allboosts), np.mean(allmoneys)
def export_data(swing=True):
if swing:
s = SwingTrader()
else:
s = DayTrader()
lines = []
for i in range(len(s.xs)):
lines.append(s.xs[i] + [s.ys[i]])
print "len is ", len(s.xs)
with open('amd-out.txt', 'w') as f:
for line in lines:
for item in line:
f.write("%s " % item)
f.write("\n")
def export_day_trader():
s = DayTrader()
# limit to last 7, not last 20, data points
trimmed_xs, trimmed_ys = [], []
lines = []
for i in range(len(s.xs)):
trimmed_xs = s.xs[i][-5:]
y = [s.ys[i]]
lines.append(trimmed_xs + y)
print trimmed_xs + y
# print lines
print "len is ", len(s.xs)
with open('amd-out-trimmed.txt', 'w') as f:
for line in lines:
for item in line:
f.write("%s " % item)
f.write("\n")
def test_em():
SwingTrader().expectation_max()
DayTrader().expectation_max()
def test_k_means():
# s = SwingTrader()
# s.k_means_clustering(n_clusters=2)
d = DayTrader()
d.k_means_clustering(n_clusters=2)
def kmeans_rca():
s = DayTrader()
X_trans = s.rca(n_components=2)
plot_kmeans(s, X_trans,True, 2)
def kmeans_ica():
s = DayTrader()
X_trans = s.ica(n_components=4)
plot_kmeans(s, X_trans, False, 4)
def kmeans_nodimred():
s = DayTrader()
# X_trans = s.pca(n_components=3)
fig, ax = plt.subplots()
plot_kmeans(s, s.xs, True, 3)