pca-kmeans.py

#!/usr/bin/python
# encoding: utf-8
# code: pca
# purpose: run pca on validators to start grouping validators

# thanks Sebastial for how to conduct PCA with sklearn bit.ly/2npnTT8
# thanks Yangki for guiding how to take a sklearn PCA and make a df of principal components bit.ly/2oNPtfj
# thanks sentdex for showing how to conduct kmeans classifcation bit.ly/2oBvLS1

import pandas as pd
import numpy as np
from sklearn.preprocessing import StandardScaler
from sklearn.decomposition import PCA as sklearnPCA
from sklearn.cluster import KMeans
import matplotlib.pyplot as plt
import itertools
import os
%matplotlib inline


# if not in the correct directory, this function helps the user get there
def getRightDirectory():
    if('whovalidates' in os.getcwd()):
        pass
    else:
        neededDirectory = input("Provide path to 'whovalidates' folder: ")
        try:
            os.chdir(neededDirectory)
        except:
            print('\nThat directory does not exist! Try again')
            getRightDirectory()
        else:
            if('/output' in [x[0] for x in os.walk(os.getcwd())][-1]):
                pass
            else:
                print("\nThat folder is not correct, please provide the whovalidates folder!")
                getRightDirectory()
getRightDirectory()
    

#%% CLEAN DATAFRAME, SCALE FOR PCA


# read in validators, set index to user name, rid the uid
validators = pd.read_csv("output/validators_analysis.csv").drop(['validations','changesets'], axis=1)
# csv holding # validated sqaures by 1400 known validators
trueValidation = pd.read_csv ("output/trueValidations.csv")
validators = pd.merge(validators,trueValidation,how='inner',on=['user_name'])
validators['validations_age'] = validators['validations']/validators['a']
validatorsTMP = validators.drop(['user_id','user_name','mapping_freq','validations'],axis=1).dropna()
validatorsFIN = validators.drop(['user_id'],axis=1)

# 0 = mean, +-1 = +-standard devation
vStd = StandardScaler().fit_transform(validatorsTMP)

#%% EIGENDECOMPOSITION

vMeanVec = np.mean(vStd, axis=0)
covMat = (vStd - vMeanVec).T.dot((vStd - vMeanVec)) / (vStd.shape[0]-1)


#%% CORRELATION MATRIX
covMat = np.cov(vStd.T)
eigVals, eigVecs = np.linalg.eig(covMat)

#%% SORT EIGENPAIRS

for ev in eigVecs:
    np.testing.assert_array_almost_equal(1.0, np.linalg.norm(ev))

# Make a list of (eigenvalue, eigenvector) tuples
eigPairs = [(np.abs(eigVals[i]), eigVecs[:,i]) for i in range(len(eigVals))]
# Sort the (eigenvalue, eigenvector) tuples from high to low
eigPairs.sort(key=lambda x: x[0], reverse=True)



#%% EXPLAINED VARIANCE
tot = sum(eigVals)
varExp = [(i / tot)*100 for i in sorted(eigVals, reverse=True)]
cumVarExp = np.cumsum(varExp)

# create and write out dataframe with eigenVec/Explained variance info for PCs

# this dataframe's columns are principal components. the first row is
# the explained variance, then following rows are each variable's and its eigenvector

EigVar =  pd.concat([pd.DataFrame(np.array(varExp)).T,pd.DataFrame(eigVecs)])

# set index so rows are better explained
Index = ['explained_variance'] + validatorsTMP.columns.tolist()
EigVar['Index'] = Index 
EigVar = EigVar.set_index('Index')

ComponentNames = ["c" + str(x+1) for x in range(4)]
EigVar.columns = ComponentNames

# write it to a csv
EigVar.to_csv('output/EigVar.csv')

#%% PLOT EXPLAINED VARIANCE FOR EIGEN PAIRS

plt.plot(varExp)
plt.plot(cumVarExp)

#%% PUSH OUT A DATAFRAME

vPCA = sklearnPCA()
yPCA = vPCA.fit_transform(vStd)
validatorsPCA = pd.DataFrame(yPCA, columns = ComponentNames)
validatorsPCA=validatorsPCA.merge(
    validatorsTMP,
    left_index=True,
    right_index=True).merge(
    validatorsFIN,
    left_index=True,
    right_index=True)
validatorsPCA.to_csv('output/validatorsPCA-validatorsOnly.csv')

#%% CLASSIFY IT WITH KMEANS
# create nd.array for each possible combination of the first 4 principal
# components
componentCombos = list(itertools.combinations(range(4),2))
kMeans = KMeans(n_clusters=3,init='k-means++',n_init=300, verbose=1)
for i in range(len(componentCombos)):
    cA = validatorsPCA['c' + str(componentCombos[i][0]+1)].tolist()
    cB = validatorsPCA['c' + str(componentCombos[i][1]+1)].tolist()
    cAcBList = [[cA[i],cB[i]] for i in range(0,len(cA))]
    cAcBArray = np.array(cAcBList)
    kMeans.fit(cAcBArray)
    centroids = kMeans.cluster_centers_
    labels = kMeans.labels_
    labels
    validatorsLST = []
    for j in range(len(cAcBArray)):
        validatorsLST.append(labels[j])
    validatorsTMP = pd.DataFrame(validatorsLST)
    validatorsTMP.columns = [
            'kmeansClass'+'c'+ str(componentCombos[i][0]+1) +
                          'c'+ str(componentCombos[i][1]+1) 
    ]
    validatorsPCA=validatorsPCA.merge(
        validatorsTMP,
        left_index=True,
        right_index=True)
validatorsPCA.to_csv('output/validatorsPCAclassified-validatorsOnly.csv')
# next http://bit.ly/2nqWArk