Jx-MLT : A Machine Learning Toolbox for Classification

---

"Toward Talent Scientist: Sharing and Learning Together" --- Jingwei Too

---

Introduction

• This toolbox contains 6 widely used machine learning algorithms
• The Demo_KNN and Demo_LDA provide the examples of how to use these methods on benchmark dataset

Usage

You may switch the algorithm by changing the knn in from ML.knn import jkfold to other abbreviations

• If you wish to use linear discriminate analysis ( LDA ) classifier then you may write

from ML.lda import jkfold 

• If you want to use naive bayes ( NB ) classifier then you may write

from ML.nb import jkfold  

Input

• feat : feature vector matrix ( Instance x Features )
• label : label matrix ( Instance x 1 )
• opts : parameter settings
  • ho : ratio of testing data in hold-out validation
  • kfold : number of folds in k-fold cross-validation

Output

• mdl : Machine learning model ( It contains several results )
  • acc : classification accuracy
  • con : confusion matrix
  • r : precision and recall

How to choose the validation scheme?

There are three types of performance validations. These validation strategies are listed as following ( KNN is adopted as an example ).

• Hold-out cross-validation

from ML.knn import jho

• K-fold cross-validation

from ML.knn import jkfold

• Leave-one-out cross-validation

from ML.knn import jloo

Example 1 : K-nearest neighbor ( KNN ) with k-fold cross-validation

import numpy as np
import pandas as pd
# change this to switch algorithm & types of validation (jho, jkfold, jloo)
from ML.knn import jkfold 
import matplotlib.pyplot as plt
import seaborn as sns


# load data
data  = pd.read_csv('ionosphere.csv')
data  = data.values
feat  = np.asarray(data[:, 0:-1])
label = np.asarray(data[:, -1])

# parameters
k     = 5
kfold = 10
opts  = {'k':k, 'kfold':kfold}
# KNN with k-fold
mdl   = jkfold(feat, label, opts) 

# overall accuracy
accuracy = mdl['acc']

# confusion matrix
confmat  = mdl['con']
print(confmat)

# precision & recall
result   = mdl['r']
print(result)


# plot confusion matrix
uni     = np.unique(label)
# Normalise
con     = confmat.astype('float') / confmat.sum(axis=1)[:, np.newaxis]
fig, ax = plt.subplots()
sns.heatmap(con, annot=True, fmt='.2f', xticklabels=uni, yticklabels=uni, cmap="YlGnBu")
plt.ylabel('Actual')
plt.xlabel('Predicted')
plt.title('KNN')
plt.show()

Example 2 : Support vector machine ( SVM ) with hold-out validation

import numpy as np
from sklearn import datasets
# change this to switch algorithm & types of validation (jho, jkfold, jloo)
from ML.svm import jho  
import matplotlib.pyplot as plt
import seaborn as sns


iris  = datasets.load_iris()
feat  = iris.data 
label = iris.target

# parameters
ho     = 0.3    # 30% testing set
kernel = 'rbf'
opts   = {'ho':ho, 'kernel':kernel}
# machine learning
mdl    = jho(feat, label, opts) 

# overall accuracy
accuracy = mdl['acc']

# confusion matrix
confmat  = mdl['con']
print(confmat)

# precision & recall
result   = mdl['r']
print(result)

# plot confusion matrix
uni     = np.unique(label)
# Normalise
con     = confmat.astype('float') / confmat.sum(axis=1)[:, np.newaxis]
fig, ax = plt.subplots()
sns.heatmap(con, annot=True, fmt='.2f', xticklabels=uni, yticklabels=uni, cmap="YlGnBu")
plt.ylabel('Actual')
plt.xlabel('Predicted')
plt.title('SVM')
plt.show()

Example 3 : Linear Discriminate Analysis ( LDA ) with leave-one-out validation

import numpy as np
from sklearn import datasets
# change this to switch algorithm & types of validation (jho, jkfold, jloo)
from ML.lda import jloo  
import matplotlib.pyplot as plt
import seaborn as sns


iris  = datasets.load_iris()
feat  = iris.data 
label = iris.target

# parameters
opts  = {}
# machine learning
mdl   = jloo(feat, label, opts) 

# overall accuracy
accuracy = mdl['acc']

# confusion matrix
confmat  = mdl['con']
print(confmat)

# precision & recall
result   = mdl['r']
print(result)

# plot confusion matrix
uni     = np.unique(label)
# Normalise
con     = confmat.astype('float') / confmat.sum(axis=1)[:, np.newaxis]
fig, ax = plt.subplots()
sns.heatmap(con, annot=True, fmt='.2f', xticklabels=uni, yticklabels=uni, cmap="YlGnBu")
plt.ylabel('Actual')
plt.xlabel('Predicted')
plt.title('LDA')
plt.show()

Requirement

• Python 3
• Numpy
• Pandas
• Scikit-learn
• Matplotlib
• Seaborn

List of available machine learning methods

• Click on the name of algorithm to check the parameters
• Use the opts to set the specific parameters
• If you do not set extra parameters then the algorithm will use default setting in here

No.	Abbreviation	Name	Support
06	knn	K-nearest Neighbor	Multi-class
05	svm	Support Vector Machine	Multi-class
04	dt	Decision Tree	Multi-class
03	lda	Linear Discriminate Analysis	Multi-class
02	nb	Naive Bayes	Multi-class
01	rf	Random Forest	Multi-class