如何在凱拉斯獲得sklearns子集多標籤精度？

Question

Sklearn針對多標籤問題計算accuracy_score（docs）作爲子集準確性：如果每個預測標籤都正確，則每個預測僅計爲正確。 Keras的默認'accuracy'是二進制精度：

def accuracy(y_true, y_pred): 
    return K.mean(K.equal(y_true, K.round(y_pred)), axis=-1) 

值太高。我認爲K.equal計算每個元素上的相等性，對嗎？我如何使它計算每個矢量的相等性？

Answer 1

我認爲你需要爲此寫一個callback。這是一個例子。

import numpy as np 

from keras.models import Sequential 
from keras.layers import Dense 
from keras.callbacks import Callback 

from sklearn.metrics.classification import accuracy_score 

n_labels = 3 
n_samples = 20 
n_feats = 10 

x = np.random.normal(size=(n_samples, n_feats)) 
y = np.random.randint(2, size=(n_samples, n_labels)) 

model = Sequential() 
model.add(Dense(50, input_shape=(n_feats,))) 
model.add(Dense(n_labels, activation='sigmoid')) 

model.compile(
    loss='binary_crossentropy', 
    optimizer='rmsprop', 
    metrics=['accuracy'] 
) 

class SkLearnHistory(Callback): 
    def __init__(self, x, y): 
     self.x = x 
     self.y = y 

    def on_train_begin(self, logs={}): 
     self.accuracy = [] 

    def on_epoch_end(self, epoch, logs={}): 
     y_pred = np.round(self.model.predict(self.x)) 
     y_true = self.y 

     score = accuracy_score(y_true, y_pred) 
     self.accuracy.append(score) 

     return 

sklearn_history = SkLearnHistory(x, y) 
keras_history = model.fit(x, y, epochs=5, verbose=0, callbacks=[sklearn_history]) 

keras_history.history['acc'] 

# [0.46666663885116577, 
# 0.48333334922790527, 
# 0.51666665077209473, 
# 0.58333337306976318, 
# 0.60000002384185791] 

sklearn_history.accuracy 

# [0.14999999999999999, 0.20000000000000001, 0.25, 0.25, 0.25]