# Bruce A. Maxwell
# Spring 2015
# CS 251 Project 7 Clustering
#
# Executes PCA on the UCI activity recognition data set
# Identifies the # of eigenvectors required to represent at least 90% of the variation
# Clusters the projected data into 6 categories using the specified number of eigenvectors
# Classifies the data set
# Compares the classifications to actual classes using a confusion matrix
# 
#
import sys
import data
import analysis as an
import numpy as np

def main(argv):
    
    if len(argv) < 3:
        print 'usage: python %s <Train CSV file> <Train categories CSV file>' % (argv[0])
        exit(-1)

    # read the features and categories data sets
    print 'Reading %s and %s' % (argv[1], argv[2])
    try:
        d = data.Data(argv[1])
    except:
        print 'Unable to open %s' % (argv[1])
        exit(-1)

        
    try:
        catdata = data.Data(argv[2])
    except:
        print 'Unable to open %s' % (argv[2])
        exit(-1)

    
    # execute PCA analysis
    print 'Executing PCA'
    pcadata = an.pca( d, d.get_headers() )

    print 'Evaluating eigenvalues'
    # identify how many dimensions it takes to represent 90% of the variation
    evals = pcadata.get_eigenvalues()
    esum = np.sum(evals)
    cum = evals[0,0]
    cumper = cum / esum
    i = 1
    while cumper < 0.9:
        cum += evals[0,i]
        cumper = cum/esum
        i += 1

    print 'Dimensions to reach 90% of variation:', i

    cheaders = pcadata.get_headers()[:i]

    # cluster the data
    K = 6

    # Use the average of each category as the initial means
    truecats = catdata.get_data(catdata.get_headers()[0:1])
    #tmpcats = truecats - 1 
    tmpcats = truecats # Don't adjust if we're using corrected labels
    
    print 'Clustering to %d clusters' % (K)
    codebook, codes, errors = an.kmeans(pcadata, cheaders, K, categories = tmpcats)
        
    # build a confusion matrix
    confmtx = [[0,0,0,0,0,0],[0,0,0,0,0,0],[0,0,0,0,0,0],[0,0,0,0,0,0],[0,0,0,0,0,0],[0,0,0,0,0,0]]
    for i in range(codes.shape[0]):
        #confmtx[codes[i,0]][int(truecats[i,0])-1] += 1
        confmtx[codes[i,0]][int(truecats[i,0])] += 1 # don't adjust

    print "\nConfusion Matrix:\n"
    print 'Actual->     Walking   Walk-up   Walk-dwn  Sitting   Standing   Laying'
    for i in range(len(confmtx)):
        s = 'Cluster %d' % (i)
        for val in confmtx[i]:
            s += "%10d" % (val)
        print s
    print

if __name__ == "__main__":
    main(sys.argv)