def train_fmeasure(self, patterns,iteration_no,learning_rate,momentum_factor):
# N: learning rate
accl = []
for i in range(iteration_no):
error = 0.0
E=[]
y_true = []
y_pred = []
num=1
for p in patterns:
p=p.split(',')
targets=float(p.pop())
inputs = []
for x in p:
if len(x)>0:
inputs.append(float(x))
y_true_val = self.update(inputs).pop()
error = error + self.backPropagate(targets, learning_rate, momentum_factor)
if abs(round(error,2)) <= 0.01:
y_true.append([targets])
else:
y_true.append([y_true_val])
y_pred.append([targets])
num = num+1
acc = mean_f1(y_true, y_pred)
accl.append(acc*100)
return accl
def train_fmeasure(self, iterations=1000, train_vector=[[[0.0],[0.0]]]):
y_true=[]
y_pred=[]
accl=[]
time_constant=iterations/log(self.radius)
radius_decaying=0.0
learning_rate_decaying=0.0
influence=0.0
stack=[] #Stack for storing best matching unit's index and updated FV and PV
temp_FV=[0.0]*self.FV_size
temp_PV=[0.0]*self.PV_size
for i in range(1,iterations+1):
radius_decaying=self.radius*exp(-1.0*i/time_constant)
learning_rate_decaying=self.learning_rate*exp(-1.0*i/time_constant)
for j in range(len(train_vector)):
input_FV=train_vector[j][0]
input_PV=train_vector[j][1]
best=self.best_match(input_FV)
stack=[]
for k in range(self.total):
dist=self.distance(self.nodes[best],self.nodes[k])
if dist < radius_decaying:
temp_FV=[0.0]*self.FV_size
temp_PV=[0.0]*self.PV_size
influence=exp((-1.0*(dist**2))/(2*radius_decaying*i))
for l in range(self.FV_size):
#Learning
temp_FV[l]=self.nodes[k].FV[l]+influence*learning_rate_decaying*(input_FV[l]-self.nodes[k].FV[l])
for l in range(self.PV_size):
#Learning
temp_PV[l]=self.nodes[k].PV[l]+influence*learning_rate_decaying*(input_PV[l]-self.nodes[k].PV[l])
#Push the unit onto stack to update in next interval
stack[0:0]=[[[k],temp_FV,temp_PV]]
for l in range(len(stack)):
self.nodes[stack[l][0][0]].FV[:]=stack[l][1][:]
self.nodes[stack[l][0][0]].PV[:]=stack[l][2][:]
y_true.append(input_PV)
y_pred.append([int(round(self.predict(input_FV)[0],0))])
acc = mean_f1(y_true, y_pred)
accl.append(acc*100)
return accl
def kmeans_fmeasure(file_name,iteration_no,learning_rate):
dataSet,target = loadDataset(file_name)
k = 3
accl =[]
data_length = len(dataSet)
for x in range(iteration_no):
labels, centroids = kmeans(dataSet,k)
y_true = []
y_pred = []
for i in range(data_length):
y_true.append([target[i]])
y_pred.append([labels[i]])
acc = mean_f1(y_true, y_pred)
accl.append(acc*100)
if file_name=="iris.csv":
clusters2 = [[] for _ in range(k)]
for i, p in enumerate(dataSet):
clusters2[labels[i]].append(p)
colors = cycle("rgbcmyk")
clusters2.sort()
try:
from pylab import plot,show,grid,xlabel,ylabel
except ImportError:
pass
else:
for group, (ca,cb,cc,cd), color in zip(clusters2,centroids,colors):
pa,pb,pc,pd = zip(*group)
x_cords=[]
y_cords=[]
for i in range(len(pa)):
x_cords.append(pa[i]+pb[i])
for i in range(len(pc)):
y_cords.append(pc[i]+pd[i])
plot(x_cords,y_cords, "o" + color)
plot([ca+cb],[cc+cd], '^' + 'y')
xlabel('Sepal Length + Sepal Width')
ylabel('Petal Length + Petal Width')
grid(True)
show()
return accl
print "Number of test examples", num_test
'''
predictor = learn_one_vs_all(X_train,Y_train,classifier)
raw_predictions = predictor.decision_function(X_test)
boolean_predictions = make_class_labels(raw_predictions,topk=3)
label_predictions = class_inverse_transform(boolean_predictions,uniq_label_mapping)
#label_predictions = uniq_label_mapping.inverse_transform(numpy.array(boolean_predictions))
label_predictions = [list(item) for item in label_predictions]
assert(len(label_predictions)==len(true_labels))
'''
#new custom classifier code, will give n tag classifiers. sparsified
class_label_preds = train_custom_one_vs_all(X_train,
X_test,
Y_train,
topk=3)
label_predictions = []
for pred in class_label_preds:
t = [uniq_label_mapping[i] for i in pred]
label_predictions += [t]
# print the predictions mapping
for orig, lp in zip(true_labels, label_predictions):
print orig, "----", lp, "\n"
#calculate the mean-fscore
print "The mean f-score is ", mean_f1.mean_f1(true_labels,
label_predictions)
'''
label_predictions = run_multiclass(X_train,X_test,Y_train_labels,fp_set_list)
'''
#new custom classifier code, will give n tag classifiers. sparsified
class_label_preds = train_custom_one_vs_all(X_train,X_test,Y_train,topk=3)
label_predictions = []
for pred in class_label_preds:
t = [uniq_label_mapping[i] for i in pred]
label_predictions += [t]
'''
# print the predictions mapping
for orig,lp in zip(true_labels,label_predictions):
print orig, "----" ,lp,"\n"
#calculate the mean-fscore
print "The mean f-score is ", mean_f1.mean_f1(true_labels,label_predictions)
def lms_fmeasure(file_name,iteration_no,learning_rate):
SSEL=[]
with open(file_name,'r') as calc:
l=calc.read().split('\n')
MCN=0
x = l[0].split(',')
w=numpy.random.random(len(x)-1)
b=random.random()
w=numpy.matrix(w)
b=numpy.matrix(b)
while MCN<iteration_no:
y_true = []
y_pred = []
count = 0
for i in l:
j=i.split(',')
target=float(j.pop())
p=[]
for x in j:
if len(x)>0:
p.append(float(x))
p=numpy.matrix(p)
net=summation(p,w,b)
a=act_func(net)
e=target-a
if abs(round(e,2)) <= 0.01:
y_true.append([target])
else:
w=modify_wb(w,learning_rate,target,p)
y_true.append([a])
y_pred.append([target])
acc = mean_f1(y_true, y_pred)
SSEL.append(acc*100)
MCN = MCN+1
if file_name=="iris.csv":
with open(file_name,'r') as calc:
pat = calc.read().split('\n')
print(len(pat))
x_cord1=[]
y_cord1=[]
x_cord2=[]
y_cord2=[]
x_cord3=[]
y_cord3=[]
for i in pat:
p = i.split(',')
targets=float(p.pop())
inputs = []
for x in p:
if len(x)>0:
inputs.append(float(x))
if targets==0.0:
x_cord1.append(inputs[0]+inputs[1])
y_cord1.append(inputs[2]+inputs[3])
if targets==1.0:
x_cord2.append(inputs[0]+inputs[1])
y_cord2.append(inputs[2]+inputs[3])
if targets==2.0:
x_cord3.append(inputs[0]+inputs[1])
y_cord3.append(inputs[2]+inputs[3])
fig = plt.figure()
ax = fig.add_subplot(111)
type1 = ax.scatter(x_cord1, y_cord1, s=50, c='red')
type2 = ax.scatter(x_cord2, y_cord2, s=50, c='green')
type3 = ax.scatter(x_cord3, y_cord3, s=50, c='blue')
ax.set_title('Petal size vs Sepal size', fontsize=14)
ax.set_xlabel('Petal size (cm)')
ax.set_ylabel('Sepal size (cm)')
ax.legend([type1, type2, type3], ["Iris Setosa", "Iris Versicolor", "Iris Virginica"], loc=2)
ax.grid(True,linestyle='-',color='0.75')
plt.show()
return SSEL
def knn_fmeasure(file_name,iteration_no,learning_rate):
# prepare data
trainingSet=[]
testSet=[]
split = 0.67
loadDataset(file_name, split, trainingSet, testSet)
print('Train set: ' + repr(len(trainingSet)))
print('Test set: ' + repr(len(testSet)))
# generate predictions
predictions=[]
accl = []
k = 3
for i in range(iteration_no):
y_true = []
y_pred = []
for x in range(len(testSet)):
neighbors = getNeighbors(trainingSet, testSet[x], k)
result = getResponse(neighbors)
predictions.append(result)
y_true.append([float(testSet[x][-1])])
y_pred.append([float(result)])
acc = mean_f1(y_true, y_pred)
accl.append(acc*100)
if file_name=="iris.csv":
with open(file_name,'r') as calc:
pat = calc.read().split('\n')
print(len(pat))
x_cord1=[]
y_cord1=[]
x_cord2=[]
y_cord2=[]
x_cord3=[]
y_cord3=[]
for i in pat:
p = i.split(',')
targets=float(p.pop())
inputs = []
for x in p:
if len(x)>0:
inputs.append(float(x))
if targets==0.0:
x_cord1.append(inputs[0]+inputs[1])
y_cord1.append(inputs[2]+inputs[3])
if targets==1.0:
x_cord2.append(inputs[0]+inputs[1])
y_cord2.append(inputs[2]+inputs[3])
if targets==2.0:
x_cord3.append(inputs[0]+inputs[1])
y_cord3.append(inputs[2]+inputs[3])
fig = plt.figure()
ax = fig.add_subplot(111)
type1 = ax.scatter(x_cord1, y_cord1, s=50, c='red')
type2 = ax.scatter(x_cord2, y_cord2, s=50, c='green')
type3 = ax.scatter(x_cord3, y_cord3, s=50, c='blue')
ax.set_title('Petal size vs Sepal size', fontsize=14)
ax.set_xlabel('Petal size (cm)')
ax.set_ylabel('Sepal size (cm)')
ax.legend([type1, type2, type3], ["Iris Setosa", "Iris Versicolor", "Iris Virginica"], loc=2)
ax.grid(True,linestyle='-',color='0.75')
plt.show()
return accl
