[mlp]Keras Multilayer Perceptron for scikit-learn

是 否 +2 论坛币

k人 参与回答

经管之家送您一份

应届毕业生专属福利!

求职就业群

赵安豆老师微信：zhaoandou666

经管之家联合CDA

送您一个全额奖学金名额~ !

立即领取

感谢您参与论坛问题回答

经管之家送您两个论坛币！

+2 论坛币

Keras Multilayer Perceptron for scikit-learn

Keras makes it very easy to implement deep-learning models, however these are not compatible with scikit-learn out-of-the-box. This prohibits our models to be tested using sciki-learn's methods GridSearchCV and cross_val_score.

I created this MLP class to be compatible with scikit-learn that contains the fit, predict, and predict_proba methods.

Install

pip install mlp

Quick guide

Initialize your classifier

from mlp import MLPclf = MLP(n_hidden=10, n_deep=3, l1_norm=0, drop=0.1, verbose=0)

Now evaluate your classifier with scikit-learn's cross_val_score

from sklearn.cross_validation import cross_val_scorescores = cross_val_score(clf, data, label, cv=5, n_jobs=-1, scoring='f1_weighted')print(scores)

本帖隐藏的内容

https://github.com/alvarouc/mlp

扫码加我 拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

分享0 收藏0 回帖

关键词：scikit-learn multilayer Learn multi Earn

1. import warnings
   
2. warnings.filterwarnings('ignore')
   
3. from mlp import MLP
   
4. import matplotlib.pyplot as plt
   
5. from sklearn.datasets import make_moons
   
6. import numpy as np
   
7. import pandas as pd
   
8. import seaborn as sns
   
9. from sklearn.preprocessing import StandardScaler as SS
   
10. %matplotlib inline
    
11. data, label = make_moons(n_samples=500, noise=0.4)
    
12. 
    
13. data_s = SS().fit_transform(data)
    
14. 
    
15. plt.plot(data_s[np.where(label==0)[0],0],data[np.where(label==0)[0],1],'r.', label='Label = 0')
    
16. plt.plot(data_s[np.where(label==1)[0],0],data[np.where(label==1)[0],1],'b.', label='Label = 1')
    
17. 
    
18. plt.title('Moons')
    
19. plt.xlabel('Feature 1')
    
20. plt.ylabel('Feature 2')
    
21. plt.legend()

复制代码

1. from sklearn.cross_validation import cross_val_score
   
2. 
   
3. clf = MLP(n_hidden=10, n_deep=3, l1_norm=0, drop=0.1, verbose=1)
   
4. 
   
5. scores = cross_val_score(clf, data, label, cv=5, n_jobs=1, scoring='roc_auc')
   
6. print(scores)

复制代码

1. Using MLP with sklearn's GridSearchCV
   
2. In [3]:
   
3. from sklearn.grid_search import GridSearchCV
   
4. 
   
5. params = {'n_hidden': [10, 15],
   
6.          'l1_norm': [0.0],
   
7.          'n_deep': [2, 3],
   
8.          'drop': [0.1]}
   
9. 
   
10. clf_grid = GridSearchCV(clf, param_grid=params, scoring='roc_auc', cv=3, n_jobs=3)
    
11. scores = cross_val_score(clf_grid, data, label, cv=5, n_jobs=1, scoring='roc_auc')
    
12. print(scores)

复制代码

1. Extracting layer representation
   
2. In [4]:
   
3. clf = MLP(n_hidden=8, n_deep=3, l2_norm=0.0, l1_norm=0.0, drop=0.0, verbose=0, early_stop=True,
   
4.           max_epoch=10000, patience=1000)
   
5. clf.fit(data, label)
   
6. layers = clf.feed_forward(data)
   
7. plt.plot(layers[2][label==0,0], layers[2][label==0,1],'.')
   
8. plt.plot(layers[2][label==1,0], layers[2][label==1,1],'r.')
   
9. plt.title('Trainning AUC {:.2f}'.format(clf.auc(data,label)))

复制代码

1. Plotting decision boundary
   
2. In [2]:
   
3. def plot_decision(clf, data):
   
4.     xmin, ymin = data.min(axis=0)
   
5.     xmax, ymax = data.max(axis=0)
   
6. 
   
7.     x,y = np.meshgrid(np.linspace(xmin,xmax,100),np.linspace(ymin,ymax,100))
   
8.     mesh = np.vstack((x.ravel(), y.ravel())).T
   
9.     ypred = clf.predict(mesh)
   
10. 
    
11.     plt.contourf(x, y, ypred.reshape(x.shape), alpha=0.5)
    
12.     plt.scatter(data[label==0,0], data[label==0,1], marker='+')
    
13.     plt.scatter(data[label==1,0], data[label==1,1], c= 'red', marker='x')
    
14.     plt.xlim(xmin, xmax)
    
15.     plt.ylim(ymin, ymax)

复制代码

1. L2 norm effects
   
2. In [5]:
   
3. plt.figure(figsize=(25,5))
   
4. for n, l2 in enumerate([0, 0.0001, 0.001, 0.01]):
   
5.     clf = MLP(n_hidden=30, n_deep=3, l2_norm=l2, l1_norm=0, early_stop=False, max_epoch=10000)
   
6.     clf.fit(data, label)
   
7.     plt.subplot(1,5, n+1)
   
8.     plot_decision(clf, data)
   
9.     plt.title('L2 norm: {}'.format(l2))

复制代码

谢谢楼主分享！

谢谢楼主分享！