我写了一个基本程序来了解MLP分类器中发生了什么？
from sklearn.neural_network import MLPClassifier
数据：标记为男性或女性的身体指标（身高，宽度和鞋码）的数据集：
X = [[181, 80, 44], [177, 70, 43], [160, 60, 38], [154, 54, 37], [166, 65, 40],
[190, 90, 47], [175, 64, 39],
[177, 70, 40], [159, 55, 37], [171, 75, 42], [181, 85, 43]]
y = ['male', 'male', 'female', 'female', 'male', 'male', 'female', 'female',
'female', 'male', 'male']
准备模型：
clf= MLPClassifier(hidden_layer_sizes=(3,), activation='logistic',
solver='adam', alpha=0.0001,learning_rate='constant',
learning_rate_init=0.001)
列车
clf= clf.fit(X, y)
学习分类器的属性：
print('current loss computed with the loss function: ',clf.loss_)
print('coefs: ', clf.coefs_)
print('intercepts: ',clf.intercepts_)
print(' number of iterations the solver: ', clf.n_iter_)
print('num of layers: ', clf.n_layers_)
print('Num of o/p: ', clf.n_outputs_)
测试
print('prediction: ', clf.predict([ [179, 69, 40],[175, 72, 45] ]))
计算值。精度
print( 'accuracy: ',clf.score( [ [179, 69, 40],[175, 72, 45] ], ['female','male'], sample_weight=None ))
current loss computed with the loss function: 0.617580287851
coefs: [array([[ 0.17222046, -0.02541928, 0.02743722],
[-0.19425909, 0.14586716, 0.17447281],
[-0.4063903 , 0.148889 , 0.02523247]]), array([[-0.66332919],
[ 0.04249613],
[-0.10474769]])]
intercepts: [array([-0.05611057, 0.32634023, 0.51251098]), array([ 0.17996649])]
number of iterations the solver: 200
num of layers: 3
Num of o/p: 1
prediction: ['female' 'male']
accuracy: 1.0
/home/anubhav/anaconda3/envs/mytf/lib/python3.6/site-packages/sklearn/neural_network/multilayer_perceptron.py:563: ConvergenceWarning: Stochastic Optimizer: Maximum iterations reached and the optimization hasn't converged yet.
% (), ConvergenceWarning)
current loss computed with the loss function: 0.639478303643
coefs: [array([[ 0.02300866, 0.21547873, -0.1272455 ],
[-0.2859666 , 0.40159542, 0.55881399],
[ 0.39902066, -0.02792529, -0.04498812]]), array([[-0.64446013],
[ 0.60580985],
[-0.22001532]])]
intercepts: [array([-0.10482234, 0.0281211 , -0.16791644]), array([-0.19614561])]
number of iterations the solver: 39
num of layers: 3
Num of o/p: 1
prediction: ['female' 'female']
accuracy: 0.5
current loss computed with the loss function: 0.691966937074
coefs: [array([[ 0.21882191, -0.48037975, -0.11774392],
[-0.15890357, 0.06887471, -0.03684797],
[-0.28321762, 0.48392007, 0.34104955]]), array([[ 0.08672174],
[ 0.1071615 ],
[-0.46085333]])]
intercepts: [array([-0.36606747, 0.21969636, 0.10138625]), array([-0.05670653])]
number of iterations the solver: 4
num of layers: 3
Num of o/p: 1
prediction: ['male' 'male']
accuracy: 0.5
current loss computed with the loss function: 0.697102567593
coefs: [array([[ 0.32489731, -0.18529689, -0.08712877],
[-0.35425908, 0.04214241, 0.41249622],
[-0.19993622, -0.38873908, -0.33057999]]), array([[ 0.43304555],
[ 0.37959392],
[ 0.55998979]])]
intercepts: [array([ 0.11555407, -0.3473817 , -0.16852093]), array([ 0.31326347])]
number of iterations the solver: 158
num of layers: 3
Num of o/p: 1
prediction: ['male' 'male']
accuracy: 0.5
我有以下问题：
1.Why in the RUN1 the optimizer did not converge?
2.Why in RUN3 the number of iteration were suddenly becomes so low and in the RUN4 so high?
3.What else can be done to increase the accuracy which I get in RUN1.?
答案 0 :(得分：12)
1：你的MLP没有收敛： 该算法通过逐步收敛到最小值进行优化，在运行1中找不到最小值。
2跑步的差异：
您的MLP有一些随机起始值，因此您不会得到与数据中相同的结果。似乎你在第四次跑步时开始非常接近最低限度。您可以将MLP的random_state参数更改为常量，例如random_state=0一遍又一遍地获得相同的结果。
3是最困难的一点。 您可以使用
优化参数from sklearn.model_selection import GridSearchCV
Gridsearch将您的测试集拆分为大小相同的部分，使用一个部分作为测试数据，其余部分作为训练数据。因此，它优化了将数据拆分为多个分类器。
你需要指定（你的数据很小，所以我建议2或3）你分割的部分数量，分类器（你的MLP），以及你想要优化的参数网格：
param_grid = [
{
'activation' : ['identity', 'logistic', 'tanh', 'relu'],
'solver' : ['lbfgs', 'sgd', 'adam'],
'hidden_layer_sizes': [
(1,),(2,),(3,),(4,),(5,),(6,),(7,),(8,),(9,),(10,),(11,), (12,),(13,),(14,),(15,),(16,),(17,),(18,),(19,),(20,),(21,)
]
}
]
由于你曾经用三个神经元的隐藏层获得100％的准确度，你可以尝试优化学习率和动量等参数而不是隐藏层。
像这样使用Gridsearch：
clf = GridSearchCV(MLPClassifier(), param_grid, cv=3,
scoring='accuracy')
clf.fit(X,y)
print("Best parameters set found on development set:")
print(clf.best_params_)