计算 numpy 数组中的圆形和方形星星

Question

练习说明如下：

• 给定数据“stars_train_X_y.npz”。我需要找到一个模型来预测圆形星星是否多于方形星星，如果圆形星星更多则返回 1，否则返回 0。
• 我必须使用线性或逻辑回归模型来拟合火车数据。
• accuracy_score 必须高于 0.9

我的做法： 我预测了圆形和方形星星的数量并比较了两个列表，但我只得到 accuracy_score 为 0.715。

import sklearn
import sklearn.linear_model
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_squared_error
from sklearn.metrics import accuracy_score


import numpy as np

data = np.load('stars_train_X_y.npz')
X = data['X_train'] #   np.ndarray of size (8000, 4096),  flattened image (64x64) of the night sky.
y = data['y_train'] #   np.ndarray of size (8000, 2), number of circular stars and number square stars

#Split dataset into 80% trian, 10% test, 10# val
(X_train, X_temp, y_train, y_temp) = train_test_split(data['X_train'], data['y_train'], test_size=0.2, random_state=0, shuffle = True)
(X_test, X_val, y_test, y_val) = train_test_split(X_temp, y_temp, test_size=0.5, random_state=0, shuffle = True)

#Getting list of either number of circular stars and square stars
def shapeList(inp, elm):
    n = [(t[elm]) for t in inp]
    return n 

cLstTrain = shapeList(y_train, 0) # y_train: only circular stars count
cLstTest = shapeList(y_test, 0)   # y_test: only circular stars count
sLstTrain = shapeList(y_train, 1) # y_train: only square stars count
sLstTest = shapeList(y_test, 1)   # y_test: only square stars count

def predict(trainX_, trainY_, testX_, testY_ ):
    model = sklearn.linear_model.LogisticRegression(random_state=0, max_iter=1000)
    model.fit(trainX_, trainY_)
    pred = model.predict(testX_)
    MSE = mean_squared_error(testY_, pred)
    ACC = accuracy_score(testY_, pred)
    print("MSE : " + str(MSE))
    print("ACC : " + str(ACC) )
    return pred

circularPredictions = (predict(X_train, cLstTrain, X_test, cLstTest))
squarePredictions = (predict(X_train, sLstTrain, X_test, sLstTest))

'''1 for pictures with more stars, 0 for pictures with less stars. (compared with square stars)'''
def compare(CircPred, SqrePred):
    lst = []
    for i in range (0, len(CircPred)):
        if CircPred[i] > SqrePred[i]: 
            n = 1
            lst.append([n])
        else:
            n = 0
            lst.append([n])
    return lst

predictions = (compare(circularPredictions, squarePredictions))
testValues = (compare(cLstTest, sLstTest))
print(accuracy_score(testValues, predictions))

为了提高准确性，我尝试了以下方法：

网格搜索

我用过：

paramGrid = {
    'penalty': ['l1', 'l2'],
    'C': [0.1, 1, 10, 100]
}
model_search = GridSearchCV(model, paramGrid, cv=5)

这给了我 0.72 的稍微好一点的准确度，但运行时间从 2 分钟增加到 18 分钟。

我很确定解决方案是为此进行预处理我已经尝试了以下技术：

• MinMaxScaler：导致准确性降低
• 二值化器：导致精度降低
• PowerTransformer：导致精度降低

我对这类事情很陌生，你能建议什么样的预处理技术，或者有没有其他方法可以改进？

提前致谢：D