def main():
#read data
x, labels = readData("Data/iris.data", ",", scale=False)
#shuffle
p = np.random.permutation(len(x))
x = x[p]
labels = labels[p]
classes, y = np.unique(labels, return_inverse=True)
# #plot h vs f-measure
accuracies = list()
fmeasures = list()
for h in range(1, 8):
validation = kfoldCrossValidation(x,
labels,
10,
h=h,
maxIterations=500,
learning_rate=0.003,
learning_rate_v=0.003)
fmeasures.append(validation[4])
plt.bar(range(1, 8), fmeasures)
plt.xlabel("h")
plt.ylabel("F-Measure")
plt.show()
#plot effect of learning_rate
handles = np.empty([0, 1])
k = len(labels)
h = 4
n = x.shape[1]
iv = np.random.rand(k, h) / 10
iw = np.random.rand(h, n) / 10
for lr in [0.001, 0.002, 0.003, 0.004, 0.005]:
#perform gradient descent
v, w, iterations, errors = learnmultilayer.gradient_descent(
x,
y,
classes,
h=4,
maxIterations=100,
learning_rate=lr,
learning_rate_v=lr,
iv=iv,
iw=iw)
# print("Itearative Method Coefficients: {},Iterations:{}".format(thetas,iterations))
handle = plt.plot(errors[1:100, 0],
errors[1:100, 1],
linewidth=4,
label="Learning Rate = {:.3f}".format(lr))
handles = np.append(handles, handle)
plt.legend(
handles=[handles[0], handles[1], handles[2], handles[3], handles[4]])
plt.ticklabel_format(useOffset=False)
plt.xlabel("Iteration")
plt.ylabel("- log likelihood")
plt.show()
def main():
#read data
X, y = getDigitsData()
#get only first 3 classes
ind = [k for k in range(len(y)) if y[k] in [0, 1, 2]]
X = X[ind]
y = y[ind]
#shuffle
p = np.random.permutation(len(X))
X = X[p]
y = y[p]
#use first 500 examples
X = X[:500]
y = y[:500]
# perfrom 10 fold cross validation
validation = kfoldCrossValidation(X,
y,
5,
h=784,
maxIterations=100,
learning_rate=0.005,
learning_rate_v=0.00001)
#plot effect of learning rate
classes, y = np.unique(y, return_inverse=True)
handles = np.empty([0, 1])
h = 784
n = X.shape[1]
k = len(y)
iv = np.random.rand(k, h) / 10
iw = np.random.rand(h, n) / 10
for lr in [0.05, 0.1, 0.15, 0.2]:
#perform gradient descent
v, w, iterations, errors = learnmultilayer.gradient_descent(
X,
y,
classes,
h=h,
maxIterations=50,
learning_rate=lr,
learning_rate_v=0.00001,
iv=iv,
iw=iw)
handle = plt.plot(errors[1:50, 0],
errors[1:50, 1],
linewidth=4,
label="Learning Rate = {:.3f}".format(lr))
handles = np.append(handles, handle)
plt.legend(handles=[handles[0], handles[1], handles[2], handles[3]])
plt.ticklabel_format(useOffset=False)
plt.xlabel("Iteration")
plt.ylabel("- log likelihood")
plt.show()
def train(x,
labels,
h,
threshold=0.01,
maxIterations=900,
learning_rate=0.001,
learning_rate_v=0.0001):
classes, y = np.unique(labels, return_inverse=True)
return learnmultilayer.gradient_descent(x,
y,
classes,
threshold=threshold,
h=h,
maxIterations=maxIterations,
learning_rate=learning_rate,
learning_rate_v=learning_rate_v)
def main():
#read data
X,y = getDigitsData()
#get only first 3 classes
ind = [ k for k in range(len(y)) if y[k] in [0,1,2] ]
X = X[ind]
y = y[ind]
#shuffle
p = np.random.permutation(len(X))
X = X[p]
y = y[p]
#use first 500 examples
X = X[:500]
y = y[:500]
# perfrom 10 fold cross validation
validation = kfoldCrossValidation(X, y, 5, h=784, maxIterations=100, learning_rate=0.005, learning_rate_v=0.00001)
#plot effect of learning rate
classes, y = np.unique(y, return_inverse=True)
handles = np.empty([0,1])
h=784
n = X.shape[1]
k = len(y)
iv = np.random.rand(k,h)/10
iw = np.random.rand(h,n)/10
for lr in [0.05,0.1, 0.15, 0.2]:
#perform gradient descent
v,w, iterations, errors = learnmultilayer.gradient_descent(X,y, classes, h=h, maxIterations=50,learning_rate=lr, learning_rate_v=0.00001,iv=iv,iw=iw)
handle = plt.plot(errors[1:50,0],errors[1:50,1], linewidth = 4, label="Learning Rate = {:.3f}".format(lr))
handles = np.append(handles,handle)
plt.legend(handles=[handles[0],handles[1],handles[2],handles[3]])
plt.ticklabel_format(useOffset=False)
plt.xlabel("Iteration")
plt.ylabel("- log likelihood")
plt.show()
def main():
#read data
x, labels = readData("Data/iris.data",",",scale=False)
#shuffle
p = np.random.permutation(len(x))
x = x[p]
labels = labels[p]
classes, y = np.unique(labels, return_inverse=True)
# #plot h vs f-measure
accuracies = list()
fmeasures = list()
for h in range(1,8):
validation = kfoldCrossValidation(x,labels, 10, h=h, maxIterations=500,learning_rate=0.003, learning_rate_v=0.003)
fmeasures.append(validation[4])
plt.bar(range(1,8),fmeasures)
plt.xlabel("h")
plt.ylabel("F-Measure")
plt.show()
#plot effect of learning_rate
handles = np.empty([0,1])
k = len(labels)
h=4
n = x.shape[1]
iv = np.random.rand(k,h)/10
iw = np.random.rand(h,n)/10
for lr in [0.001,0.002, 0.003, 0.004, 0.005]:
#perform gradient descent
v,w, iterations, errors = learnmultilayer.gradient_descent(x,y, classes, h=4, maxIterations=100,learning_rate=lr, learning_rate_v=lr,iv=iv,iw=iw)
# print("Itearative Method Coefficients: {},Iterations:{}".format(thetas,iterations))
handle = plt.plot(errors[1:100,0],errors[1:100,1], linewidth = 4, label="Learning Rate = {:.3f}".format(lr))
handles = np.append(handles,handle)
plt.legend(handles=[handles[0],handles[1],handles[2],handles[3],handles[4]])
plt.ticklabel_format(useOffset=False)
plt.xlabel("Iteration")
plt.ylabel("- log likelihood")
plt.show()
def train(x,labels, h, threshold=0.01, maxIterations = 900, learning_rate=0.001, learning_rate_v=0.0001):
classes, y = np.unique(labels, return_inverse=True)
return learnmultilayer.gradient_descent(x, y, classes, threshold=threshold, h=h, maxIterations=maxIterations, learning_rate=learning_rate, learning_rate_v=learning_rate_v);
