def run_svm(inputs, targets, nRuns=5):
total_time, total_percent = 0, 0
for i in range(nRuns):
x, y = shuffle_data(inputs, targets, True, i)
x_in, x_out, y_in, y_out = separate_data(x, y, False)
for i in range(x_out.shape[0]):
for j in range(x_out.shape[1]):
if x_out[i, j] == 0.0:
x_out[i, j] = -1.0
start = time()
predict = np.zeros((y_out.shape[0], 3))
classifier0 = svm(x_in, x_out[:, 0], kernel="rbf")
predict[:, 0] = classifier0(y_in, soft=True).T
classifier1 = svm(x_in, x_out[:, 1], kernel="rbf")
predict[:, 1] = classifier1(y_in, soft=True).T
classifier2 = svm(x_in, x_out[:, 2], kernel="rbf")
predict[:, 2] = classifier2(y_in, soft=True).T
total_time += time() - start
cm = conf_mat(predict, y_out)
total_percent += np.trace(cm) / np.sum(cm) * 100
return total_time / nRuns, total_percent / nRuns
def run_knn(inputs, targets, nRuns=5):
total_time, total_percent = 0, 0
for i in range(nRuns):
x, y = shuffle_data(inputs, targets, True, i)
x_in, x_out, y_in, y_out = separate_data(x, y, False)
start = time()
predict = knn(3, x_in, x_out, y_in)
total_time += time() - start
cm = conf_mat(predict, y_out)
total_percent += np.trace(cm) / np.sum(cm) * 100
return total_time / nRuns, total_percent / nRuns
def run_boost(inputs, targets, features, nRuns=5):
total_time, total_percent = 0, 0
#classes = [None] * len(targets)
classes = np.empty((len(targets), 1), dtype=int)
for i in range(len(targets)):
if targets[i][0] == 1:
classes[i] = 0 #"Setosa"
elif targets[i][1] == 1:
classes[i] = 1 #"Versicolour"
else:
classes[i] = 2 #"Virginica"
#classes = np.array(classes).reshape(len(targets), 1)
for i in range(nRuns):
x, y = shuffle_data(inputs, classes, True, i)
x_in, x_out, y_in, y_out = separate_data(x, y, False)
x_out, y_out = x_out[:, 0], y_out[:, 0]
start = time()
classify = boosting(x_in, x_out, features)
predict = np.empty((len(y_out), 1), dtype=int)
for idx, val in enumerate(y_in):
predict[idx] = classify(val)
total_time += time() - start
oneHotPredict = np.empty((len(y_out), 3), dtype=int)
oneHotYout = np.empty((len(y_out), 3), dtype=int)
for i in range(len(predict)):
if predict[i] == 0:
oneHotPredict[i] = [1, 0, 0]
elif predict[i] == 1:
oneHotPredict[i] = [0, 1, 0]
else:
oneHotPredict[i] = [0, 0, 1]
if y_out[i] == 0:
oneHotYout[i] = [1, 0, 0]
elif y_out[i] == 1:
oneHotYout[i] = [0, 1, 0]
else:
oneHotYout[i] = [0, 0, 1]
cm = conf_mat(oneHotPredict, oneHotYout)
total_percent += np.trace(cm) / np.sum(cm) * 100
return total_time / nRuns, total_percent / nRuns
def run_rbf_wtm(inputs, targets, nRuns=5):
total_time, total_percent = 0, 0
for i in range(nRuns):
x, y = shuffle_data(inputs, targets, True, i)
x_in, x_out, v_in, v_out, y_in, y_out = separate_data(x, y, True)
start = time()
forward = rbf(x_in,
x_out,
outtype="weight_matrix",
nRbf=5,
normalize=True)
predict = forward(y_in)
total_time += time() - start
cm = conf_mat(predict, y_out)
total_percent += np.trace(cm) / np.sum(cm) * 100
return total_time / nRuns, total_percent / nRuns
def run_mlp(inputs, targets, nRuns=5):
total_time, total_percent = 0, 0
for i in range(nRuns):
x, y = shuffle_data(inputs, targets, True, i)
x_in, x_out, v_in, v_out, y_in, y_out = separate_data(x, y, True)
start = time()
forward = mlp(x_in,
x_out,
valid_inputs=v_in,
valid_targets=v_out,
validate=True,
outtype="softmax",
eta=0.1,
nHidden=5,
valid_iterations=100)
predict = forward(y_in)
total_time += time() - start
cm = conf_mat(predict, y_out)
total_percent += np.trace(cm) / np.sum(cm) * 100
return total_time / nRuns, total_percent / nRuns
import os
import numpy as np
import pandas as pd
from Learning.ensemble_learning import randomForest
from Learning.scoring import conf_mat
from Preprocessing.reformat import one_hot_encode
if __name__ == "__main__":
df = pd.read_csv(os.path.join(os.getcwd(), "Data/car.data"))
car = df.to_numpy()
x_in = car[::2, :-1]
x_out = car[::2, -1]
y_in = car[1::2, :-1]
y_out = car[1::2, -1]
features = ["buying", "maint", "doors", "persons", "lug_boot", "safety"]
classify = randomForest(x_in, x_out, features, 5)
predict = classify(y_in)
options = ["unacc", "acc", "good", "vgood"]
predict_check, y_out_check = one_hot_encode(predict,
options), one_hot_encode(
y_out, options)
breakpoint()
conf_mat(predict_check, y_out_check)