plt.figure()
[numItems, numFeatures] = knn.data.shape
for i in range(0, numItems / 2):
itemClass = int(trainY[i])
currColor = colors[itemClass, :]
plt.scatter(trainX[i, 0],
trainX[i, 1],
facecolor=currColor,
edgecolor='black',
s=50,
lw=2)
correctCount = 0
for i in range(0, numItemsTrain / 2):
itemClass = int(testY[i])
currColor = colors[itemClass, :]
prediction = int(knn.Predict(testX[i, :]))
edgeColor = colors[prediction, :]
if (itemClass == prediction):
correctCount = correctCount + 1
plt.scatter(testX[i, 0],
testX[i, 1],
facecolor=currColor,
edgecolor=edgeColor,
s=50,
lw=2)
#print(correctCount)
acc = int((float(correctCount) / (numItemsTrain / 2)) * 100)
print("accuracy: " + str(acc) + "%")
#plt.scatter(trainX[:,0:1].flatten(),trainX[:,1:2].flatten(),c=trainY)
#plt.scatter(testX[:,0:1].flatten(),testX[:,1:2].flatten(),c=testY)
plt.show()
pickle.dump(var, pickle_out)
pickle_out.close()
SaveDataToPickle([trainX, trainY, testX, testY], "KNN_dataset")
knn = KNN()
knn.Use_K_Of(15)
knn.Fit(trainX,trainY)
print('fitted.')
import time
start_time = time.time()
if False: # using KNN_backup
great = 0
for row in range(0,2000):
prediction = knn.Predict(testX[row])
if prediction==testY[row]:
great += 1
else:
print('start.')
step_size = 1
for i in range(0, testX.shape[0], step_size):
predictions = knn.Predict(testX[i:i+step_size])
great = np.sum(np.equal(predictions,testY[i:i+step_size]))
print("accuracy:", (1.*great)/step_size)
if great!=step_size:
# Error occur
print(predictions)
print(testY[i:i+step_size])
print(testX[i].shape)
for bone in range(0, 2):
for joint in range(0, 3):
X[row, col] = set1[finger, bone, joint, row]
X[row + 1000, col] = set2[finger, bone, joint, row]
col = col + 1
return X, y
trainM = ReduceData(trainM)
trainN = ReduceData(trainN)
testM = ReduceData(testM)
testN = ReduceData(testN)
trainM = CenterData(trainM)
trainN = CenterData(trainN)
testM = CenterData(testM)
testN = CenterData(testN)
trainX, trainy = ReshapeData(trainM, trainN)
testX, testy = ReshapeData(testM, testN)
knn.Use_K_Of(15)
knn.Fit(trainX, trainy)
counter = 0
for row in range(2000):
prediction = knn.Predict(testX[row, :])
if prediction == testy[row]:
counter = counter + 1
print((counter / 2000) * 100)
trainX = knn.data[::2, 1:3]
trainy = knn.target[::2]
testX = knn.data[1::2, 1:3]
testy = knn.target[1::2]
colors = np.zeros((3, 3), dtype='f')
colors[0, :] = [1, 0.5, 0.5]
colors[1, :] = [0.5, 1, 0.5]
colors[2, :] = [0.5, 0.5, 1]
knn.Use_K_Of(15)
knn.Fit(trainX, trainy)
actualClass = testy[0]
prediction = knn.Predict(testX[0, 0:2])
#visualization
plt.figure()
# plt.scatter(trainX[:,0],trainX[:,1],c=trainy)
# plt.scatter(testX[:,0],testX[:,1],c=testy)
[numItems, numFeatures] = knn.data.shape
for i in range(0, numItems / 2):
itemClass = int(trainy[i])
currColor = colors[itemClass, :]
plt.scatter(trainX[i, 0],
trainX[i, 1],
facecolor=currColor,
edgeColor=[0, 0, 0],
s=50,
pickle_out.close()
SaveDataToPickle([trainX, trainY, testX, testY], "KNN_dataset")
knn = KNN()
knn.Use_K_Of(15)
knn.Fit(trainX, trainY)
import time
start_time = time.time()
if False: # using KNN_backup
great = 0
for row in range(0, 2000):
prediction = knn.Predict(testX[row])
if prediction == testY[row]:
great += 1
else:
predictions = knn.Predict(testX)
great = np.sum(np.equal(predictions, testY))
print("accuracy:", great / 10000.)
print("time: ", time.time() - start_time)
# First accuracy: 90.75%
# Second accuracy: 97.4%
# Third accuracy: 98.65%
pickle.dump(knn, open('userData/classifier.p', 'wb'))
X = np.zeros((2000,5*2*3),dtype='f')
y = np.zeros(2000,dtype='f')
for row in range(0,1000):
y[row] = 5
y[row+1000] = 6
col = 0
for finger in range(0,5):
for bone in range(0,2):
for tipAndBase in range(0,3):
X[row,col] = set1[finger,bone,tipAndBase,row]
X[row+1000,col] = set2[finger,bone,tipAndBase,row]
col = col + 1
return X,y
trainX,trainy= ReshapeData(train5,train6)
testX,testy= ReshapeData(test5,test6)
knn.Fit(trainX,trainy)
# for row in range(0,2000):
counter = 0
for row in range(0, 2000):
itemClass = int(testy[row])
# prediction
prediction = int(knn.Predict(testX[row]))
if itemClass == prediction:
counter += 1
# print counter
print counter
print counter/float(2000)*100
colors = np.zeros((3, 3), dtype='f')
colors[0, :] = [1, 0.5, 0.5]
colors[1, :] = [0.5, 1, 0.5]
colors[2, :] = [0.5, 0.5, 1]
#plt.scatter(trainX[:,0], trainX[:,1], c=trainy)
#plt.show()
#plt.scatter(testX[:,0], testX[:,1], c=testy)
#plt.show()
knn.Use_K_Of(15)
knn.Fit(trainX, trainy)
actualClass = testy[70]
prediction = knn.Predict(testX[70, 1:3])
print(actualClass, prediction)
[numItems, numFeatures] = knn.data.shape
for i in range(0, numItems / 2):
itemClass = int(trainy[i])
currColor = colors[itemClass, :]
plt.scatter(trainX[i, 0], trainX[i, 1], facecolor=currColor, s=50, lw=2)
counter = 0
for i in range(0, numItems / 2):
itemClass = int(testy[i])
currColor = colors[itemClass, :]
prediction = int(knn.Predict(testX[i, 1:3]))
if prediction == int(testy[i]):
test9 = CenterData(test9)
# add new one
trainX, trainy = ReshapeData(train0, train1, train2, train3, train4, train5,
train6, train7, train8, train9)
testX, testy = ReshapeData(test0, test1, test2, test3, test4, test5, test6,
test7, test8, test9)
knn = KNN()
knn.Use_K_Of(25)
trainX = trainX.astype(int)
trainy = trainy.astype(int)
knn.Fit(trainX, trainy[:, 0])
actualClass = testy[0]
prediction = knn.Predict(testX[0, :])
#print(actualClass, prediction)
#prediction = knn.Predict(testX[1])
#print(prediction)
counter = 0
# add 1000
for row in range(0, 10000):
prediction = int(knn.Predict(testX[row, :]))
if prediction == int(testy[row, 0]):
counter += 1
print("Number correct is %s" % str(counter))
pickle.dump(knn, open('userData/classifier.p', 'wb'))