def main():
N = 124 # sincd we only use 80% for training (~100)
noiseVariance = .2
dataFunction = sinc
X_train, X_test, T_train, T_test = initData(N, sinc, noiseVariance)
clf = RVR(X_train,
T_train,
'linearSplineKernel',
useFast=False,
betaFixed=False)
clf.fit()
print("The relevance vectors (%d):" % len(clf.relevanceVectors))
print(clf.relevanceVectors)
T_pred, _ = clf.predict(X_test)
# Plot training data
X = np.linspace(-10, 10, 250)
plt.plot(X, np.sin(X) / X, label='Orig. func')
plt.scatter(X_train, T_train, s=20, label='Training samples', zorder=2)
# Plot predictions
predictedMu, _ = clf.predict(X)
plt.plot(X, predictedMu, label='Pred. func (mean)', dashes=[2, 2])
# Plot relevance vectors
plt.scatter(clf.relevanceVectors,
clf.relevanceTargets,
label="Relevance vectors",
s=50,
facecolors="none",
color="k",
zorder=10)
print("Re-estimated sigma:")
print(np.sqrt(1 / clf.beta))
plt.ylim(-0.3, 1.1)
plt.xlabel("x")
plt.ylabel("t")
plt.legend()
#plt.savefig("../plots/sincdataplotnoisy.png", bbox_inches="tight")
plt.show()
def main():
N = 100
noiseSpread = 0.001
X_train, X_test, T_train, T_test = initData(N, cos, noiseSpread)
# Very sensible to initial value of beta (as described in the paper)
clf = RVR(X_train,
T_train,
'RBFKernel',
beta=0.001**-2,
useFast=True,
convergenceThresh=10**-2,
maxIter=500)
clf.fit()
print("The relevance vectors:")
print(clf.relevanceVectors)
print("Number of relevance vectors", clf.relevanceVectors.shape[0])
print("Spread:", np.sqrt(clf.beta**-1))
print("beta:", clf.beta)
T_pred, _ = clf.predict(X_test)
# Plot training data
plt.scatter(X_train, T_train, s=20, label='Training data')
# Plot predictions
plt.scatter(X_test, T_pred, s=20, color='r', label='Predictions')
# Plot relevance vectors
plt.scatter(clf.relevanceVectors,
clf.relevanceTargets,
label="Relevance vectors",
s=50,
facecolors="none",
color="k",
zorder=1)
plt.xlabel("x")
plt.ylabel("t")
plt.legend()
# plt.savefig("../plots/sincdataplot.png", bbox_inches="tight")
plt.show()
def main():
N = None
dataFunction = airfoil
X_train, X_test, T_train, T_test = initData(N, dataFunction)
clf = RVR(X_train, T_train, 'RBFKernel')
clf.fit()
print("Number of relevance vectors:")
print(len(clf.relevanceVectors))
T_pred, _ = clf.predict(X_test)
relError = np.mean(
[abs(true - pred) / true for true, pred in zip(T_test, T_pred)])
absError = np.mean(
[abs(true - pred) for true, pred in zip(T_test, T_pred)])
stdRelError = np.std(
[abs(true - pred) / true for true, pred in zip(T_test, T_pred)])
stdAbsError = np.std(
[abs(true - pred) for true, pred in zip(T_test, T_pred)])
print("Mean absolute error: ", absError, " / std: ", stdAbsError)
print("Mean relative error: ", relError, " / std: ", stdRelError)
# number of features - plot every feature/data dimension
for i in range(5):
# Plot training data
plt.scatter(X_train[:, i], T_train, s=20, label='Training data')
# Plot predictions
plt.scatter(X_test[:, i], T_pred, s=20, color='r', label='Predictions')
# Plot relevance vectors
plt.scatter(clf.relevanceVectors[:, i],
clf.relevanceTargets,
label="Relevance vectors",
s=50,
facecolors="none",
color="k",
zorder=1)
plt.xlabel("x")
plt.ylabel("t")
plt.legend()
# plt.savefig("../plots/sincdataplot.png", bbox_inches="tight")
plt.show()
from data import initData
from flask_graphql import GraphQLView
from schema import schema
from mongoengine import connect
from model import Task
# from flask_pymongo import PyMongo
app = Flask(__name__)
default_query = """{
allTodo{
edges {
node {
id,
title,
description,
done
}
}
}
}"""
app.add_url_rule('/graphql',
view_func=GraphQLView.as_view('graphql',
schema=schema,
graphiql=True))
if __name__ == "__main__":
initData()
app.run(debug=True)
#!/usr/bin/env python3 # -*- coding: UTF-8 -*- """Main loop of UltraViolet.""" import config import mailer import data import time time.sleep(5) # wait 5 secs for datebase to come online data.initData()