svmExperiment.py

import timeit
start = timeit.default_timer()
import itertools
import numpy as np
from scipy import stats
from scipy import linalg
import pylab as pl
from sklearn import svm, linear_model
from sklearn.model_selection import train_test_split
from random import random
from dataread import *



#reading the data from the excel


n_samples = 14970 #this is the number of images in our dataset
n_features = 34 #num of features per feature vector
data = np.empty((n_samples, n_features), dtype=np.int) #creates a 2d array with the first dimension number of samples and second number of features thus the matrix is 502 x 10 
target = np.empty((n_samples), dtype=np.int) #creates a 1d array that is of length sample number ie 502
qid = np.empty((n_samples), dtype=np.int)

trainDat = trainData()
trainLabel = trainLabels()
# print(trainDat)
for y in range(n_samples-1):
    for x in range(n_features):
        data[y][x] = trainDat[x][y]# X axis is first index and Y is second!!!!! why numpy...

for i in range(n_samples-1):
    if trainLabel[0][i] == 'triangle':# X axis is first index and Y is second!!!!! why numpy...
        target[i] = 0  
    if trainLabel[0][i] == 'circle':# X axis is first index and Y is second!!!!! why numpy...
        target[i] = 1
    if trainLabel[0][i] == 'square':# X axis is first index and Y is second!!!!! why numpy...
        target[i] = 2  
    if trainLabel[0][i] == 'star':# X axis is first index and Y is second!!!!! why numpy...
        target[i] = 3       
     

X_train, X_test, y_train, y_test = train_test_split(data, target, train_size = 0.8)
# split into train and test set
# cv = cross_validate.StratifiedShuffleSplit(target, test_size=.8) #TODO
# train, test = iter(cv).next() #creates a test mask array and a train mask array. these arrays are the indexes of x and y that are assigned to each set

# X_train, y_train = data[train], target[train] #using the train index mask create an copy array of both x and y respectivly that is only filled with the training items
# X_test, y_test, qid_test = data[test], target[test], qid[test] #using the test index mask create an copy array of both x and y respectivly that is only filled with the test items


svms = []
avgscore = 0
avgscores = 0
print("Initializing SVMs")
for i in range(10000): #make that many svms and test them
    print("Initializing SVM: "+str(i))
    thiscval = (i/10000)+0.0001
    svms.append(svm.SVC(kernel='poly', C=thiscval, cache_size=5000)) # C is random float between 0.0 and 1.0
    #cache_size is faster if bigger
    print("C for SVM "+str(i)+" is "+str(thiscval))
    print("SVM "+str(i)+" Initialized training")
    svms[i].fit(X_train, y_train) # Train the selected svm

    # coef = svms[i].coef_.ravel() / linalg.norm(svms[i].coef_)
    clf_predict = svms[i].predict(X_test) #Predict values
    score = svms[i].score(X_test, y_test) #Evaluate score

    # print("SVM "+str(i)+" coef: "+sstr(coef))
    print("SVM "+str(i)+" X_test prediction: "+str(clf_predict))
    print("SVM "+str(i)+" score: "+str(score))
    avgscore+=score
    avgscores+=1
avgscore = avgscore/avgscores
print("The average score for the SVMs")
print(avgscore)

stop = timeit.default_timer()
print("Execution Time: ")
print(stop-start)
#for i in range(len(y_test)):
    #clf_predict = clf.predict(X_test)
    #print(str(clf_predict[i]) + ' ' + str(y_test[i]))