12 KiB
12 KiB
from load_data import get_dataset
import numpy as np
from tabulate import tabulate
X_train, y_train, X_test, y_test = get_dataset(new_size=64)
Zadanie 1 (2 pkt)
Rozwiń algorytm regresji logistycznej z lab. 1, wprowadzając do niego człon regularyzacyjny.
class LogisticRegression():
def __init__(self, l2=1):
self.l2 = l2
def mapY(self, y, cls):
m = len(y)
yBi = np.matrix(np.zeros(m)).reshape(m, 1)
yBi[y == cls] = 1.
return yBi
def indicatorMatrix(self, y):
classes = np.unique(y.tolist())
m, k = len(y), len(classes)
# k = len(classes)
Y = np.matrix(np.zeros((m, k)))
for i, cls in enumerate(classes):
Y[:, i] = self.mapY(y, cls)
return Y
# Zapis macierzowy funkcji softmax
def softmax(self, X):
return np.exp(X) / np.sum(np.exp(X))
# Funkcja regresji logistcznej
def h(self, theta, X):
return 1.0 /(1.0 + np.exp(-X * theta))
# Funkcja kosztu dla regresji logistycznej
def J(self, h, theta, X, y):
m = len(y)
h_val = h(theta, X)
s1 = np.multiply(y, np.log(h_val))
s2 = np.multiply((1 - y), np.log(1 - h_val))
s3 = np.sum(s1+s2, axis=0)/m
s4 = (self.l2 * np.sum(np.square(theta))) / 2*m
return -s3 + s4
# Gradient dla regresji logistycznej
def dJ(self, h, theta, X, y):
return 1.0 / (self.l2/len(y)) * (X.T * (h(theta, X) - y))
# Metoda gradientu prostego dla regresji logistycznej
def GD(self, h, fJ, fdJ, theta, X, y, alpha=0.01, eps=10**-3, maxSteps=1000):
errorCurr = fJ(h, theta, X, y)
errors = [[errorCurr, theta]]
while True:
# oblicz nowe theta
theta = theta - alpha * fdJ(h, theta, X, y)
# raportuj poziom błędu
errorCurr, errorPrev = fJ(h, theta, X, y), errorCurr
# kryteria stopu
if abs(errorPrev - errorCurr) <= eps:
break
if len(errors) > maxSteps:
break
errors.append([errorCurr, theta])
# return theta, errors
return theta
def trainMaxEnt(self, X, Y):
n = X.shape[1]
thetas = []
for c in range(Y.shape[1]):
YBi = Y[:,c]
theta = np.matrix(np.random.random(n)).reshape(n,1)
# Macierz parametrów theta obliczona dla każdej klasy osobno.
# thetaBest, errors = self.GD(self.h, self.J, self.dJ, theta,
# X, YBi, alpha=0.1, eps=10**-4)
# thetas.append(thetaBest)
thetas.append(self.GD(self.h, self.J, self.dJ, theta, X, YBi, alpha=0.1, eps=10**-4))
return thetas
def classify(self, thetas, X):
regs = np.array([(X*theta).item() for theta in thetas])
return np.argmax(self.softmax(regs))
# probs = self.softmax(regs)
# result = np.argmax(probs)
# return result
def class_score(self, expected, predicted):
# accuracy = TP + TN / FP + FN + TP + TN
accuracy = sum(1 for exp, pred in zip(expected, predicted) if exp == pred) / len(expected)
# precision = TP / FP + TP
precision = sum(
1 for exp, pred in zip(expected, predicted) if exp == 1.0 and pred == 1.0) / sum(
1 for exp, pred in zip(expected, predicted) if exp == 1.0)
# recall = TP / FN + TP
recall = sum(
1 for exp, pred in zip(expected, predicted) if exp == 1.0 and pred == 1.0) / sum(
1 for exp, pred in zip(expected, predicted) if pred == 1.0)
f1 = (2 * precision * recall) / (precision + recall)
return accuracy, precision, recall, f1
def fit(self, X_train, y_train):
# Y = self.indicatorMatrix(y_train)
# self.thetas = self.trainMaxEnt(X_train, Y)
self.thetas = self.trainMaxEnt(X_train, self.indicatorMatrix(y_train))
def predict(self, X_test):
return np.array([self.classify(self.thetas, x) for x in X_test])
def accuracy(self, expected, predicted):
return sum(1 for x, y in zip(expected, predicted) if x == y) / len(expected)
# 16x16, l2=0.01 -> 5m 11.4s
# 32x32, l2=0.1 -> 20m 31.3s
# 64x64, l2=0.1 -> 219m 10.8s
logreg = LogisticRegression(l2=0.1)
logreg.fit(X_train, y_train)
/var/folders/3r/c8tg1h051m18qhsdccdysrt40000gn/T/ipykernel_43367/551403934.py:33: RuntimeWarning: divide by zero encountered in log s2 = np.multiply((1 - y), np.log(1 - h_val)) /var/folders/3r/c8tg1h051m18qhsdccdysrt40000gn/T/ipykernel_43367/551403934.py:33: RuntimeWarning: invalid value encountered in multiply s2 = np.multiply((1 - y), np.log(1 - h_val)) /var/folders/3r/c8tg1h051m18qhsdccdysrt40000gn/T/ipykernel_43367/551403934.py:26: RuntimeWarning: overflow encountered in exp return 1.0 /(1.0 + np.exp(-X * theta)) /var/folders/3r/c8tg1h051m18qhsdccdysrt40000gn/T/ipykernel_43367/551403934.py:32: RuntimeWarning: divide by zero encountered in log s1 = np.multiply(y, np.log(h_val)) /var/folders/3r/c8tg1h051m18qhsdccdysrt40000gn/T/ipykernel_43367/551403934.py:32: RuntimeWarning: invalid value encountered in multiply s1 = np.multiply(y, np.log(h_val))
logreg.accuracy(y_test, logreg.predict(X_test))
/var/folders/3r/c8tg1h051m18qhsdccdysrt40000gn/T/ipykernel_43367/551403934.py:22: RuntimeWarning: invalid value encountered in true_divide return np.exp(X) / np.sum(np.exp(X)) /var/folders/3r/c8tg1h051m18qhsdccdysrt40000gn/T/ipykernel_43367/551403934.py:22: RuntimeWarning: overflow encountered in exp return np.exp(X) / np.sum(np.exp(X))
0.32432432432432434
Zadanie 2 (4 pkt)
Zaimplementuj algorytm SVM z miękkim marginesem (regularyzacją).
class SVM():
def __init__(self, lr=0.001, lambda_param=10**-6, n_iters=1000):
self.lr = lr
self.lambda_param = lambda_param
self.n_iters = n_iters
def _mapY(self, y, cls):
m = len(y)
yBi = np.matrix(np.zeros(m)).reshape(m, 1)
yBi[y == cls] = 1.
return yBi
def _indicatorMatrix(self, y):
classes = np.unique(y.tolist())
m, k = len(y), len(classes)
Y = np.matrix(np.zeros((m, k)))
for i, cls in enumerate(classes):
Y[:, i] = self._mapY(y, cls)
return Y
def fit(self, X, y):
n_classes = len(np.unique(y))
y = self._indicatorMatrix(y)
y = np.where(y == 0, -1, 1)
n_features = X.shape[1]
self.weights, self.biases = [], []
for cls in range(n_classes):
y_ = y[:,cls]
y_ = np.where(y_ <= 0, -1, 1)
w, b = np.zeros(n_features), 0
for _ in range(self.n_iters):
for idx, x_i in enumerate(X):
condition = y_[idx] * (np.dot(x_i, w) - b) >= 1
if condition:
w -= self.lr * (2 * self.lambda_param * w)
else:
w -= self.lr * (2 * self.lambda_param * w - np.dot(x_i, y_[idx]))
b -= self.lr * y_[idx]
self.weights.append(w)
self.biases.append(b)
def _classify(self, x):
cls = [np.sign(np.dot(x, self.weights[i]) - self.biases[i]) for i in range(len(self.biases))]
return cls.index(1.0) if 1.0 in cls else 0
def predict(self, X):
return list(map(lambda x: self._classify(x), X))
def accuracy(self, expected, predicted):
return sum(1 for x, y in zip(expected, predicted) if x == y) / len(expected)
# 16x16 -> 1m 28.8s
# 32x32 -> 2m 2.2s
# 64x64 -> 4m 55.3s
svm = SVM()
svm.fit(X_train, y_train)
svm.accuracy(y_test, svm.predict(X_test))
0.32432432432432434