def plot():
C = [1, 10, 100, 1000]
test_accuracy = []
word_acr = []
X_train, y_train = read_train_struct()
X_test, y_test = read_test_struct()
for i in C:
y_train = y_train.ravel()
clf = train(X_train, y_train, i / len(y_train))
y_pred, score = test(clf, X_test, y_test)
test_accuracy.append(score * 100)
y_train = y_train.reshape(len(y_train, ))
given_words, pred_words = form_words(y_test, y_pred)
w_acc = word_accuracy(given_words, pred_words)
word_acr.append(w_acc * 100)
mp.figure(1)
mp.title('Letter wise Accuracy vs C - SVM-MC ')
mp.plot(C, test_accuracy)
mp.ylabel('Accuracy')
mp.xlabel('C')
mp.figure(2)
mp.plot(C, word_acr)
mp.title('Word wise Accuracy vs C - SVM-MC ')
mp.ylabel('Accuracy')
mp.xlabel('C')
def plot():
C = [1, 10]
test_accuracy = []
word_acr = []
X_train, y_train = read_train_struct()
X_test, y_test = read_test_struct()
for i in C:
train(i)
test()
my_path = os.path.abspath(os.path.dirname(__file__))
path = os.path.join(my_path, "test.outtags")
y_pred = np.loadtxt(path, usecols=(0, ))
y_test = y_test.reshape(len(y_test), )
test_acc = get_test_accuracy(y_test, y_pred)
test_accuracy.append(test_acc * 100)
given_words, pred_words = form_words(y_test, y_pred)
w_acc = word_accuracy(given_words, pred_words)
word_acr.append(w_acc * 100)
mp.figure(1)
mp.plot(C, test_accuracy)
mp.ylabel('Accuracy')
mp.xlabel('C')
mp.figure(2)
mp.plot(C, word_acr)
mp.ylabel('Accuracy')
mp.xlabel('C')
import translation as t
import rotation as r
import readInput as ri
import SVM_MC as svm
import os.path
import numpy as np
import matplotlib.pyplot as mp
my_path = os.path.abspath(os.path.dirname(__file__))
path = os.path.join(my_path, "../../data/transform.txt")
f = open(path, 'r')
#raw_data = file.read()
x=[0,50]
X_train ,y_train=ri.read_train_struct();
X_test,y_test=ri.read_test_struct()
test_accuracy =[]
word_acr =[]
train_data={}
for i in range(len(X_train)):
train_data[i+1]=y_train[i]
for num in x:
x_trans=[]
y_trans=[]
for i in range(num):
line=next(f)
line=line[0:len(line)-1]
def tamper(model):
test_acc = []
wrd_acr = []
y_pred = []
X_train, y_train = ri.read_train_struct()
X_test, y_test = ri.read_test_struct()
for num in x:
print(num)
X_trans = copy.deepcopy(X_train)
for i in range(num):
line = next(f)
line = line.split(' ')
offset = []
c = X_trans[int(line[1]), :]
example = np.array(c).reshape(16, 8)
if (line[0] == 'r'):
alpha = float(line[2])
x_result = r.rotate(example, alpha)
x_result = x_result.reshape(128, )
elif (line[0] == 't'):
offset.append(int(line[2]))
offset.append(int(line[3]))
x_result = t.translate(example, offset)
x_result = x_result.reshape(128, )
x_max_old = 0
x_min_old = 255
x_min_new = 0
x_max_new = 1
new_value = restore_range(x_result, x_max_old, x_min_old,
x_min_new, x_max_new)
X_trans[int(line[1])] = new_value
#training
if (num == 0):
if (model == 'svm'):
clf = svm.train(X_train, y_train, 1000 / len(y_train))
else:
x_y = X_train, y_train
optimize.get_params(x_y)
a = np.loadtxt("best_Weights_tampered", usecols=(0, ))
W = np.array(a[:26 * 128].reshape(26, 128))
T = np.array(a[26 * 128:26 * 128 + 26 * 26].reshape(26, 26))
#training with more than 1 transformation
else:
if (model == 'svm'):
clf = svm.train(X_trans, y_train, 1000 / len(y_train))
else:
x_y = X_trans, y_train
print(type(x_y))
optimize.get_params(x_y)
a = np.loadtxt("best_Weights_tampered", usecols=(0, ))
W = np.array(a[:26 * 128].reshape(26, 128))
T = np.array(a[26 * 128:26 * 128 + 26 * 26].reshape(26, 26))
#testing
if (model == 'svm'):
y_pred, score = svm.test(clf, X_test, y_test)
else:
y_pred = decode.max_sum(X_test, W, T)
y_pred = [y + 1 for y in y_pred]
y_test = y_test.reshape(26198, )
y_pred = np.array(y_pred).reshape(len(y_pred, ))
print((y_test))
print((y_pred))
score = max_sum_decode.get_test_accuracy(y_test, y_pred)
test_acc.append(score * 100)
y_test = y_test.reshape(len(y_test, ))
given_words, pred_words = svm.form_words(y_test, y_pred)
w_acc = svm.word_accuracy(given_words, pred_words)
wrd_acr.append(w_acc * 100)
return test_acc, wrd_acr