def check_accuracy(self):
'''computes thetas on training set, saves them, and checks
accuracy on evaluation set'''
tinit = 0.005* np.random.rand(self.LABS, self.N)
thetas = soft.optimizeThetas(tinit, self.xt, self.gt, self.LABS, self.L)
thetas = thetas.reshape(self.LABS, -1)
np.savetxt('./data/kaggle/optimized_thetas.csv', thetas, delimiter=',')
h = soft.h(thetas, self.xe)
predictions = h.argmax(axis=1)
zeros_are_right = np.subtract(self.ye.T, predictions)
misses = 1.0 * np.count_nonzero(zeros_are_right)
acc = 1 - misses/len(predictions)
print 'accuracy:', acc
pass
def check_accuracy(self):
logit_thetas = {}
soft_thetas = np.array(pd.read_csv('./data/kaggle/optimized_thetas.csv', header=None))
soft_thetas = soft_thetas.reshape(self.LABS, -1)
h = soft.h(soft_thetas, self.xe)
m = h.shape[0]
misses = 0.00
count = 0.0
for i in range(m):
true_label = self.ye[i,0]
[ml_1, ml_2] = h[i,:].argsort()[-2:][::-1] # 1st and 2nd model choices
p1,p2 = h[i,:][ml_1], h[i,:][ml_2]
right_order = True
if ml_1 > ml_2:
right_order = False
s = `ml_2`+`ml_1`
else:
s = `ml_1`+`ml_2`
if p1<0.99 and p2>0.01:
if s not in logit_thetas:
count +=1
logit_thetas[s] = self.optimize_logit_for(s)
l_t = logit_thetas[s]
logix = np.hstack([1, self.xe[i,:]])
p = logit.h(l_t, logix)
if (p>0.5):
prediction = (ml_1 if right_order else ml_2)
else:
prediction = (ml_2 if right_order else ml_1)
else:
prediction = ml_1
#print prediction, true_label
if prediction!=true_label:
misses +=1.0
print 'misses', misses
print 'logit thetas searched', count
acc = 1 - misses/m
print 'accuracy:', acc
pass
def test_model_submit(self):
# compute thetas on whole training set
tinit = 0.005* np.random.rand(self.LABS, self.N)
x = np.vstack([self.xt, self.xe])
y = np.vstack([self.yt, self.ye])
g = np.vstack([self.gt, self.ge])
# find thetas and save them
thetas = soft.optimizeThetas(tinit, x, g, self.LABS, self.L)
thetas = thetas.reshape(self.LABS, -1)
np.savetxt('./data/kaggle/submit_optimized_thetas.csv', thetas, delimiter=',')
# compute predictions
m, n = self.x_test.shape
h = soft.h(thetas, self.x_test)
predictions = np.zeros((m,2))
for i in range(m):
a = h[i,:].argmax()
predictions[i,:]=[i+1, a]
print 'To submitt add header: ImageId,Label'
print predictions[0:10,:]
np.savetxt('./data/kaggle/predictions.csv', predictions, fmt='%i,%i')
pass
def test_model_submit(self):
logit_thetas = {}
soft_thetas = np.array(pd.read_csv('./data/kaggle/submit_optimized_thetas.csv', header=None))
soft_thetas = soft_thetas.reshape(self.LABS, -1)
m, n = self.x_test.shape
h = soft.h(soft_thetas, self.x_test)
predictions = np.zeros((m,2))
for i in range(m):
[ml_1, ml_2] = h[i,:].argsort()[-2:][::-1] # 1st and 2nd model choices
p1,p2 = h[i,:][ml_1], h[i,:][ml_2]
right_order = True
if ml_1 > ml_2:
right_order = False
s = `ml_2`+`ml_1`
else:
s = `ml_1`+`ml_2`
if p1<0.99 and p2>0.01:
if s not in logit_thetas:
logit_thetas[s] = self.optimize_logit_for(s)
l_t = logit_thetas[s]
logix = np.hstack([1, self.x_test[i,:]])
p = logit.h(l_t, logix)
if (p>0.5):
predictions[i,:] = ([i+1, ml_1] if right_order else [i+1, ml_2])
else:
predictions[i,:] = ([i+1, ml_2] if right_order else [i+1, ml_1])
else:
predictions[i,:]=[i+1, ml_1]
print 'To submitt add header: ImageId,Label'
print predictions[0:10,:]
np.savetxt('./data/kaggle/predictions_2steps.csv', predictions, fmt='%i,%i')
pass
