def cv_performance(x, y, num_folds, k):
"""This function evaluates average accuracy in cross validation."""
length = len(y)
splits = split_cv(length, num_folds)
accuracy_array = []
for split in splits:
print(split)
test_x, test_y, train_x, train_y = [], [], [], []
# Finish this function to use the training instances
# indexed by `split.train` to train the classifier,
# and then store the accuracy
# on the testing instances indexed by `split.test`
# accuracy = knn.accuracy()
for i in range(len(split[1])):
test_x.append(x[split[1][i]])
test_y.append(y[split[1][i]])
for i in range(len(split[0])):
train_x.append(x[split[0][i]])
train_y.append(y[split[0][i]])
knn = Knearest(train_x, train_y, k)
confusion = knn.confusion_matrix(test_x, test_y)
accuracy = knn.accuracy(confusion)
accuracy_array.append(accuracy)
return np.mean(accuracy_array)
def cv_performance(x, y, num_folds, k):
"""This function evaluates average accuracy in cross validation."""
length = len(y)
splits = split_cv(length, num_folds)
accuracy_array = []
for split in splits:
# Finish this function to use the training instances
# indexed by `splits.train` to train the classifier,
# and then store the accuracy
# on the testing instances indexed by `splits.test`
### Begin Evan's code ###
test_data_x = [x[i] for i in split.test]
test_data_y = [y[i] for i in split.test]
train_data_x = [x[i] for i in split.train]
train_data_y = [y[i] for i in split.train]
knn = Knearest(train_data_x, train_data_y,k)
confusion = knn.confusion_matrix(test_data_x, test_data_y)
accuracy = knn.accuracy(confusion)
### End Evan's code ###
accuracy_array.append(accuracy)
return np.mean(accuracy_array)
def cv_performance(x, y, num_folds, k):
"""This function evaluates average accuracy in cross validation."""
length = len(y)
splits = split_cv(length, num_folds)
accuracy_array = []
for split in splits:
# Finish this function to use the training instances
# indexed by `split.train` to train the classifier,
# and then store the accuracy
# on the testing instances indexed by `split.test
train_x = []
train_y = []
test_x = []
test_y = []
for trainIndex in split.train:
train_x.append(x[trainIndex])
train_y.append(y[trainIndex])
for testIndex in split.test:
test_x.append(x[testIndex])
test_y.append(y[testIndex])
train_x = np.asarray(train_x)
test_x = np.asarray(test_x)
train_y = np.asarray(train_y)
test_y = np.asarray(test_y)
#print(train_y)
#print(test_x)
knn = Knearest(train_x[0], train_y[0], k)
confusion = knn.confusion_matrix(test_x[0], test_y[0])
accuracy = knn.accuracy(confusion)
accuracy_array.append(accuracy)
return np.mean(accuracy_array)
def cv_performance(x, y, num_folds, k):
length = len(y)
splits = split_cv(length, num_folds)
accuracy_array = []
for split in splits:
train_set_x = [x[i] for i in split.train]
train_set_y = [y[i] for i in split.train]
test_set_x = [x[i] for i in split.test]
test_set_y = [y[i] for i in split.test]
knn = Knearest(train_set_x, train_set_y, k)
confusion_mtr = knn.confusion_matrix(test_set_x, test_set_y)
accuracy = knn.accuracy(confusion_mtr)
accuracy_array.append(accuracy)
return np.mean(accuracy_array)
def cv_performance(x, y, num_folds, k):
"""This function evaluates average accuracy in cross validation."""
length = len(y)
splits = split_cv(length, num_folds)
accuracy_array = []
for split in splits:
# Finish this function to use the training instances
# indexed by `split.train` to train the classifier,
# and then store the accuracy
# on the testing instances indexed by `split.test`
knn = Knearest(x[split.train], y[split.train], k)
confusion = knn.confusion_matrix(x[split.test], y[split.test])
accuracy = knn.accuracy(confusion)
accuracy_array.append(accuracy)
return np.mean(accuracy_array)
def limit(lim, tr_x, tr_y, ts_x, ts_y, K):
accuracy = {}
plt.ion()
for k in K:
accr = []
for l in lim:
knn = Knearest(tr_x[:l], tr_y[:l], k)
conf = knn.confusion_matrix(ts_x[:l], ts_y[:l])
ac = knn.accuracy(conf)
accr.append(ac)
accuracy[k] = accr
plt.ion()
for vals in accuracy:
plt.plot(lim, accuracy[vals], label="K= " + str(vals))
plt.xlabel("Numbers of Training")
plt.ylabel("Accuracy")
plt.title("Figure 1 - Accuracies Against Numbers of Training ")
plt.legend(bbox_to_anchor=(0.5, 0.5), loc=2, borderaxespad=0.)
plt.savefig("question_1.png")
plt.show()
def cv_performance(x, y, num_folds, k):
"""This function evaluates average accuracy in cross validation."""
length = len(y)
splits = split_cv(length, num_folds)
accuracy_array = []
for split in splits:
# Finish this function to use the training instances
# indexed by `split.train` to train the classifier,
# and then store the accuracy
# on the testing instances indexed by `split.test`
X_train = x[:int(split.train[0])]
X_test = x[-int(split.test[0]):]
Y_train = y[:int(split.train[0])]
Y_test = y[-int(split.test[0]):]
knn1 = Knearest(X_train, Y_train, 3)
conf = knn1.confusion_matrix(X_test, Y_test)
accuracy1 = knn1.accuracy(conf)
accuracy_array.append(accuracy1)
return np.mean(accuracy_array)
knn = Knearest(train_x, train_y, k)
confusion = knn.confusion_matrix(test_x, test_y)
accuracy = knn.accuracy(confusion)
accuracy_array.append(accuracy)
return np.mean(accuracy_array)
if __name__ == "__main__":
parser = argparse.ArgumentParser(description='KNN classifier options')
parser.add_argument('--limit', type=int, default=-1,
help="Restrict training to this many examples")
args = parser.parse_args()
data = Numbers("../data/mnist.pkl.gz")
x, y = data.train_x, data.train_y
if args.limit > 0:
x, y = x[:args.limit], y[:args.limit]
best_k, best_accuracy = -1, 0
for k in [1, 3, 5, 7, 9]:
accuracy = cv_performance(x, y, 5, k)
print("%d-nearest neighber accuracy: %f" % (k, accuracy))
if accuracy > best_accuracy:
best_accuracy, best_k = accuracy, k
knn = Knearest(x, y, best_k)
confusion = knn.confusion_matrix(data.test_x, data.test_y)
accuracy = knn.accuracy(confusion)
print("Accuracy for chosen best k= %d: %f" % (best_k, accuracy))