Function benchmark_pca

ann_class2/util.py:255–303

()

def benchmark_pca():
    Xtrain, Xtest, Ytrain, Ytest = get_transformed_data()
    print("Performing logistic regression...")

    N, D = Xtrain.shape
    Ytrain_ind = np.zeros((N, 10))
    for i in range(N):
        Ytrain_ind[i, Ytrain[i]] = 1

    Ntest = len(Ytest)
    Ytest_ind = np.zeros((Ntest, 10))
    for i in range(Ntest):
        Ytest_ind[i, Ytest[i]] = 1

    W = np.random.randn(D, 10) / np.sqrt(D)
    b = np.zeros(10)
    LL = []
    LLtest = []
    CRtest = []

    # D = 300 -> error = 0.07
    lr = 0.0001
    reg = 0.01
    for i in range(200):
        p_y = forward(Xtrain, W, b)
        # print "p_y:", p_y
        ll = cost(p_y, Ytrain_ind)
        LL.append(ll)

        p_y_test = forward(Xtest, W, b)
        lltest = cost(p_y_test, Ytest_ind)
        LLtest.append(lltest)

        err = error_rate(p_y_test, Ytest)
        CRtest.append(err)

        W += lr*(gradW(Ytrain_ind, p_y, Xtrain) - reg*W)
        b += lr*(gradb(Ytrain_ind, p_y) - reg*b)
        if i % 10 == 0:
            print("Cost at iteration %d: %.6f" % (i, ll))
            print("Error rate:", err)

    p_y = forward(Xtest, W, b)
    print("Final error rate:", error_rate(p_y, Ytest))
    iters = range(len(LL))
    plt.plot(iters, LL, iters, LLtest)
    plt.show()
    plt.plot(CRtest)
    plt.show()


if __name__ == '__main__':