如何将数据集中存在的单个数据训练到已经训练好的神经网络？

Question

我已经有一个经过训练的神经网络，

clf = neural_network.MLPClassifier(
        activation='relu',
        hidden_layer_sizes=(40,40,40,40),
        learning_rate='adaptive',
        learning_rate_init=0.01,
        solver='sgd',
        alpha=1e-6,
        max_iter=20000,
        warm_start=True,
)

我已经训练过了，

clf.fit(X,Y)

这个分类器有两个类 [0,1].

现在，当我一次使用数据集中的一个数据进一步训练它时。 clf.fit([features_1],[1]) 弹出错误

Traceback (most recent call last):
  File "<pyshell#20>", line 1, in <module>
    fav_clf.fit(X_, Y)
  File "D:\Python3.5\lib\site-packages\sklearn\neural_network\multilayer_perceptron.py", line 973, in fit    
hasattr(self, "classes_")))
  File "D:\Python3.5\lib\site-packages\sklearn\neural_network\multilayer_perceptron.py", line 331, in _fit
X, y = self._validate_input(X, y, incremental)
  File "D:\Python3.5\lib\site-packages\sklearn\neural_network\multilayer_perceptron.py", line 924, in _validate_input
(self.classes_, classes))
ValueError: warm_start can only be used where `y` has the same classes as in the previous call to fit. Previously got [0 1], `y` has [1]

Answer 1

暂时忘掉warm_start，努力去理解usage of fit()。

fit(X, y):
    Fit the model to data matrix X and target(s) y.

因此，如果您只提供单个 class 的数据，模型将无法学习任何东西。 warm_start 在 MLPClassifier 中用于：

reuse the solution of the previous call to fit as initialization

因此，训练可以更快，以前训练的权重将用作初始权重，但仍需要访问多个 classes 数据才能区分它们。

现在看到你的问题，我想你想将模型用作 incremental classifier:-

Actually, the ability to learn incrementally from a mini-batch of instances (sometimes called “online learning”). All estimators implementing the partial_fit API are candidates.

因此，如果您只是想让模型逐步学习较新的数据，则需要执行 partial_fit()。但请注意，您必须转 warm_start = False，才能使用 partial_fit()。

# First call to partial_fit
clf.partial_fit(X, Y, classes=[0, 1])

# All next calls
clf.partial_fit(X, Y)    #<==  Here you can pass a single sample.

如何将数据集中存在的单个数据训练到已经训练好的神经网络？

How to train single data present in a data set to a already trained neural network?

python

neural-network

python-3.x

scikit-learn

python-3.5