如何提高Sklearn PCA的Logistic回归分数？

Question

I want to make a comparison between Lenet and PCA on regards to image recognition, so I used the German Traffic Signals Benchmark and the Sklearn PCA module, but when I tested it using Logistic Regression, the score didn't get higher than 6%, no matter what i tried.

I tried modifying the number of interations and the number of preprocesses (using normalizations and equalization), but it still didn't work

The files are loaded by Pickle by three archives:

train.p, with shape of (34799, 32, 32, 3)
test.p, with shape of (12630, 32, 32, 3)
valid.p, with shape of (4410, 32, 32, 3)

each of them with its labels, as written in y_train, y_test and y_valid. and this is the relevant part of the code:

def gray_scale(image):
    """
    Convert images to gray scale.
        Parameters:
            image: An np.array compatible with plt.imshow.
    """
    return cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)

def preprocess2(data):

    n_training = data.shape
    gray_images = np.zeros((n_training[0], n_training[1], n_training[2]))
    for i, img in enumerate(data):
        gray_images[i] = gray_scale(img)
    gray_images = gray_images[..., None]
    return gray_images

from sklearn.decomposition import PCA
from sklearn.preprocessing import StandardScaler

pca = PCA(0.95)

X_train_preprocess = preprocess2(X_train)
#Removing one dimension (34799,32,32,1) to (34799,32,32)
X_train_preprocess = X_train_preprocess.reshape(34799,32,32)
nsamples, nx, ny = X_train_preprocess.shape
X_train_preprocess = X_train_preprocess.reshape((nsamples,nx*ny))

X_test_preprocess = preprocess2(X_test)
#Removing one dimension (34799,32,32,1) to (12630,32,32)
X_test_preprocess = X_test_preprocess.reshape(12630,32,32) 
n2samples, n2x, n2y = X_test_preprocess.shape
X_test_preprocess = X_test_preprocess.reshape((n2samples,n2x*n2y))

print(X_train_preprocess.shape)
pca.fit(X_train_preprocess)
print(pca.n_components_)
scaler = StandardScaler()
scaler.fit(X_train_preprocess)
X_t_train = scaler.transform(X_train_preprocess)
X_t_test = scaler.transform(X_test_preprocess)

X_t_train = pca.transform(X_t_train)
X_t_test = pca.transform(X_t_test)

from sklearn.linear_model import LogisticRegression
logisticRegr = LogisticRegression(solver = 'lbfgs', max_iter = 5000)
logisticRegr.fit(X_t_train, y_train)
print('score', logisticRegr.predict(X_t_test[0:10]))
print('score', logisticRegr.score(X_t_test, y_test))

The results were these:

(34799, 1024)
62
/usr/local/lib/python3.6/dist-packages/sklearn/linear_model/logistic.py:469: FutureWarning: Default multi_class will be changed to 'auto' in 0.22. Specify the multi_class option to silence this warning.
  "this warning.", FutureWarning)
score [ 1  2 10 10 13 10 25  1  1  4]
score 0.028820269200316707

So I want to see if you guys can enlighten me on what am I doing wrong and what can I do to make this work properly

Answer 1

you got 2d data in image recognition ， it's best to use cnn network to present the relation in high dimension

related link : Training CNN with images in sklearn neural net