python - GridSearchCV - 跨测试访问预测值？

Question

有没有办法访问在 GridSearchCV 过程中计算的预测值？

我希望能够根据实际值(来自测试/验证集)绘制预测的 y 值。

网格搜索完成后，我可以使用

将其与其他一些数据相匹配

 ypred = grid.predict(xv)

但我希望能够绘制在网格搜索期间计算的值。也许有一种方法可以将点保存为 Pandas 数据框？

from sklearn.preprocessing import StandardScaler
from sklearn.model_selection import GridSearchCV, KFold, 
cross_val_score, train_test_split
from sklearn.pipeline import Pipeline
from sklearn.svm import SVR

scaler = StandardScaler()
svr_rbf = SVR(kernel='rbf')
pipe = Pipeline(steps=[('scaler', scaler), ('svr_rbf', svr_rbf)])
grid = GridSearchCV(pipe, param_grid=parameters, cv=splits, refit=True, verbose=3, scoring=msescorer, n_jobs=4)
grid.fit(xt, yt)

Answer 1

一个解决方案是创建一个自定义记分器并将它接收到的参数保存到一个全局变量中:

from sklearn.grid_search import GridSearchCV
from sklearn.svm import SVR
from sklearn.metrics import mean_squared_error,make_scorer

X, y = np.random.rand(2,200)
clf = SVR()

ys = []

def MSE(y_true,y_pred):
    global ys
    ys.append(y_pred)
    mse = mean_squared_error(y_true, y_pred)
    return mse

def scorer():
    return make_scorer(MSE, greater_is_better=False)

n_splits = 3 
cv = GridSearchCV(clf, {'degree':[1,2,3]}, scoring=scorer(), cv=n_splits)
cv.fit(X.reshape(-1, 1), y)

然后我们需要将每个拆分收集到一个完整的数组中:

idxs = range(0, len(ys)+1, n_splits)
#e.g. [0, 3, 6, 9]
#collect every n_split elements into a single list
new = [ys[j[0]+1:j[1]] for j in zip(idxs,idxs[1:])]
#summing every such list
ys = [reduce(lambda x,y:np.concatenate((x,y), axis=0), i) for i in new]