Skip to content

Latest commit

 

History

History
121 lines (69 loc) · 4.06 KB

README.md

File metadata and controls

121 lines (69 loc) · 4.06 KB
Raw

Perceptron Implementation

							      Author: Jasper Fang

							      Python version: 3.0
							      Matplotlib version: 3.0.3
							      Numpy version: 1.16.2
							      IDE: Sublime Text

1. PLA and visualisation

The first and second .py files are named PLA_2D_visualisation and

PLA_3D_visualisation respectively. Which contain a PLA model for two different

datasets for implement and validation (one is a 2-d dataset, one is high dimensional dataset).

1.1. 2D visualisation

In the first one, it contains a function genLinearSeparableData(w,b, numlines)

for creating random linear separable data for PLA model, and a PLA_train() and PLA_test()

for training model, visualisation and model evaluation. for the PLA_train(trainSet,plot=True),

It can return the final weight and bias for the Hyperplane, and decided to visualise the 2D dataset

and the decision region or not. PLA_test(w,b,testSet,DataVolume) receives the previous trained

'weight' and 'bias', and use a LinearSeparableDataSet to test the model, compute TP,TN,FP,FN

accuracy, FPR, TPR, precision, recall, and F-Score, then return the final evaluation results separately.

and the way to use this algorithm has been shown below:

dataScale = 20
trainSet = genLinearSeparableData([1,-2],7,int(dataScale*0.8)) # get a 2-dim vector data set
testSet  = genLinearSeparableData([1,-2],7,int(dataScale*0.2)) # use 20 percent for testing
w,b = PLA_train(trainSet,plot=True) 
accuracy,FPR,TPR,F1Score = PLA_test(w,b,testSet,int(dataScale*0.2))

1.2. 3D visualisation

In the second file, it's called PLA_3D_visualisation, which specifically designed for

3D visualisation of a high-dimensional Dataset and Hyperplane for validating the PLA model in a more

intuition way. The overall Algorithm is pretty same as the 1.1., But we use the training dataset from the

Assignment released, so that we build a function named data_load(dataPath), which receive the path of

data file to process and return an N-D_Array for our PLA model. Visulation3D() function is designed for

visualise the data and Hyperplane in a 3-Dimensional way, which receives the trainSet, trained weight, bias,

and the 3 different dimensions you want to choose. Then to plot a 3-D space for visualising the overall model.

The basic of the hyperplane construction is based on the plane equation w1x2+w2x2+w3x3-w0x0=0 in which z-axis

equals (bias - w1x1 - w2x2)/w3), and the way to use it has been shown as follow:

trainSet = data_load("train.data") # load 4-dim dataSet from Iris' data
testSet  = data_load("test.data")
w,b = PLA_train(trainSet)
Visualisation3D(trainSet,w,b,dim1=1,dim2=2,dim3=3) # or use dim1=2,dim2=3,dim3=4
accuracy,FPR,TPR,F1Score = PLA_test(w,b,testSet)

2. Binary Perceptron Class

The .py file of this part is named myPerceptron.py

This part is designed for classify the Assignment dataset 'class-1','class-2', and 'class-3'.

It aims to indentify 'class-1' and 'class-2', 'class1' and 'class-3', 'class-2' and 'class-3' respectivily.

The overall workflow like the previous two PLA models, but we built a PLA() class be used to create

each classifier. And PLA.train() function will return a status to identify the final status of current PLA

after 20 iterations (or more, you can change the itr times). Because 'maybe' one of these classifiers cannot

go converge forever (data may not belongs to linear separable). and the way to use as the following example:

trainSet_a,testSet_a = batchLoad('a') # only for classifier(a), but samiliar to (b) and (c)

maxIter = 20

Classifier_a = PLA(trainSet_a,testSet_a)
# train
Classifier_a.train(maxIter) 
# test
Classifier_a.test()
# output result
output(Classifier_a,'a')

If the code cannot be run successfully

  1. BrokenPipeError delete print() function and replaced by an output Stream or run in Anaconda or Colab

  2. ERROR at import matplotlib.pyplot as plt add a code before import plt, mpl.use('TkAgg') for changing the backend of plt 3.0.3 in macOS

  3. Unknown Error contact the author [email protected]