Python numpy.argmax() Examples

def test(model, data):
    x_test, y_test = data
    y_pred, x_recon = model.predict(x_test, batch_size=100)
    print('-'*50)
    print('Test acc:', np.sum(np.argmax(y_pred, 1) == np.argmax(y_test, 1))/y_test.shape[0])

    import matplotlib.pyplot as plt
    from utils import combine_images
    from PIL import Image

    img = combine_images(np.concatenate([x_test[:50],x_recon[:50]]))
    image = img * 255
    Image.fromarray(image.astype(np.uint8)).save("real_and_recon.png")
    print()
    print('Reconstructed images are saved to ./real_and_recon.png')
    print('-'*50)
    plt.imshow(plt.imread("real_and_recon.png", ))
    plt.show() 

def log():
    global best_score
    print("Logging")
    tr_logits, tr_cost = iter_apply(
        trX[:n_valid], trM[:n_valid], trY[:n_valid])
    va_logits, va_cost = iter_apply(vaX, vaM, vaY)
    tr_cost = tr_cost / len(trY[:n_valid])
    va_cost = va_cost / n_valid
    tr_acc = accuracy_score(trY[:n_valid], np.argmax(tr_logits, 1)) * 100.
    va_acc = accuracy_score(vaY, np.argmax(va_logits, 1)) * 100.
    logger.log(n_epochs=n_epochs, n_updates=n_updates, tr_cost=tr_cost,
               va_cost=va_cost, tr_acc=tr_acc, va_acc=va_acc)
    print('%d %d %.3f %.3f %.2f %.2f' %
          (n_epochs, n_updates, tr_cost, va_cost, tr_acc, va_acc))
    if submit:
        score = va_acc
        if score > best_score:
            best_score = score
            path = os.path.join(save_dir, desc, 'best_params')
            chainer.serializers.save_npz(make_path(path), model) 

def find_main_orientation(segments):
    """
    Checks which segment is supported by the most points and returns
    the orientation of this segment.

    Parameters
    ----------
    segments : list of BoundarySegment
        The boundary (wall) segments of the building (part).

    Returns
    -------
    main_orientation : float
        The orientation of the segment supported by the most points.
        In radians.
    """
    longest_segment = np.argmax([len(s.points) for s in segments])
    main_orientation = segments[longest_segment].orientation
    return main_orientation 

def _peaks1D(self):
        if self.num_src == 1:
            self.src_idx[0] = np.argmax(self.P)
            self.sources[:, 0] = self.loc[:, self.src_idx[0]]
            self.phi_recon = self.theta[self.src_idx[0]]
        else:
            peak_idx = []
            n = self.P.shape[0]
            for i in range(self.num_loc):
                # straightforward peak finding
                if self.P[i] >= self.P[(i-1)%n] and self.P[i] > self.P[(i+1)%n]:
                    if len(peak_idx) == 0 or peak_idx[-1] != i-1:
                        if not (i == self.num_loc and self.P[i] == self.P[0]):
                            peak_idx.append(i)

            peaks = self.P[peak_idx]
            max_idx = np.argsort(peaks)[-self.num_src:]
            self.src_idx = [peak_idx[k] for k in max_idx]
            self.sources = self.loc[:, self.src_idx]
            self.phi_recon = self.theta[self.src_idx]
            self.num_src = len(self.src_idx)


# ------------------Miscellaneous Functions---------------------# 

def process_box(self, b, h, w, threshold):
	max_indx = np.argmax(b.probs)
	max_prob = b.probs[max_indx]
	label = self.meta['labels'][max_indx]
	if max_prob > threshold:
		left  = int ((b.x - b.w/2.) * w)
		right = int ((b.x + b.w/2.) * w)
		top   = int ((b.y - b.h/2.) * h)
		bot   = int ((b.y + b.h/2.) * h)
		if left  < 0    :  left = 0
		if right > w - 1: right = w - 1
		if top   < 0    :   top = 0
		if bot   > h - 1:   bot = h - 1
		mess = '{}'.format(label)
		return (left, right, top, bot, mess, max_indx, max_prob)
	return None 

def train(self):
        while (self.epoch < self.option.max_epoch and not self.early_stopped):
            self.one_epoch_train()
            self.one_epoch_valid()
            self.one_epoch_test()
            self.epoch += 1
            model_path = self.saver.save(self.sess, 
                                         self.option.model_path,
                                         global_step=self.epoch)
            print("Model saved at %s" % model_path)
            
            if self.early_stop():
                self.early_stopped = True
                print("Early stopped at epoch %d" % (self.epoch))
        
        all_test_in_top = [np.mean(x[1]) for x in self.test_stats]
        best_test_epoch = np.argmax(all_test_in_top)
        best_test = all_test_in_top[best_test_epoch]
        
        msg = "Best test in top: %0.4f at epoch %d." % (best_test, best_test_epoch + 1)       
        print(msg)
        self.log_file.write(msg + "\n")
        pickle.dump([self.train_stats, self.valid_stats, self.test_stats],
                    open(os.path.join(self.option.this_expsdir, "results.pckl"), "w")) 

def linear_subpixel_detection(image,thresh):
    framesize=image.shape
    edgeleft=np.zeros(framesize[0])
    edgeright=np.zeros(framesize[0])
    for y in range(framesize[0]-1): #edge detection, go line by line on horizontal
        edgeleft[y]=np.argmax(image[y,0:framesize[1]]<thresh) #edge detection on pixel scale
        if edgeleft[y]!=0:
            subpxcorr=(thresh-np.float_(image[y,np.int(edgeleft[y]-1)]))/(np.float_(image[y,np.int(edgeleft[y])])-np.float_(image[y,np.int(edgeleft[y]-1)])) #subpixel correction with using corr=(threshold-intensity(edge-1))/(intensity(edge)-intensity(edge-1))
            edgeleft[y]=edgeleft[y]+subpxcorr-1 #add the correction and shift 1 pixel left, to plot on the edge properly
        edgeright[y]=np.int(framesize[1]-np.argmax(image[y,range(framesize[1]-1,0,-1)]<thresh))
        #same scheme for right edge, except the edge detection is done flipped, since np.argmax gives the first instance of the maximum value
        if edgeright[y]!=framesize[1]:
            subpxcorr=(thresh-np.float_(image[y,np.int(edgeright[y]-1)]))/(np.float_(image[y,np.int(edgeright[y])])-np.float_(image[y,np.int(edgeright[y]-1)]))
            edgeright[y]=edgeright[y]+subpxcorr-1
        else:
            edgeright[y]=0
    return edgeleft, edgeright; 

def errorfunction_subpixel_detection(image,thresh):
    erffitsize=np.int(40)
    def errorfunction(x,xdata,y): #define errorfunction to fit with a least squares fit.
        return x[0]*(1+sp.special.erf(xdata*x[1]+x[2]))+x[3] - y;
    framesize=image.shape
    edgeleft=np.zeros(framesize[0])
    edgeright=np.zeros(framesize[0])
    for y in range(framesize[0]-1): #edge detection, go line by line on horizontal
        edgeleft[y]=np.argmax(image[y,0:framesize[1]]<thresh) #edge detection on pixel scale
        if (edgeleft[y]-erffitsize)>=0  and (edgeleft[y]-erffitsize)<=framesize[0]:
            fitparts=np.array(image[y,range(np.int(edgeleft[y])-erffitsize,np.int(edgeleft[y])+erffitsize)]) #take out part of the image around the edge to fit the error function
            guess=(max(fitparts)-min(fitparts))/2,-.22,0,min(fitparts) #initial guess for error function
            lstsqrsol=sp.optimize.least_squares(errorfunction,guess,args=(np.array(range(-erffitsize,erffitsize)),fitparts)) #least sqaures fit
            edgeleft[y]=edgeleft[y]-lstsqrsol.x[2]/lstsqrsol.x[1] #add the subpixel correction
        edgeright[y]=np.int(framesize[1]-np.argmax(image[y,range(framesize[1]-1,0,-1)]<thresh))#same scheme for right edge, except the edge detection is done flipped, since np.argmax gives the first instance of the maximum value
        if (edgeright[y]-erffitsize)>=0  and (edgeright[y]-erffitsize)<=framesize[0]:
            fitparts=np.array(image[y,range(np.int(edgeright[y])-erffitsize,np.int(edgeright[y])+erffitsize)])
            guess=(max(fitparts)-min(fitparts))/2,.22,0,min(fitparts)
            lstsqrsol=sp.optimize.least_squares(errorfunction,guess,args=(np.array(range(-erffitsize,erffitsize)),fitparts))
            edgeright[y]=edgeright[y]-lstsqrsol.x[2]/lstsqrsol.x[1]
        elif edgeright[y]==framesize[1]:
            edgeright[y]=0

    return edgeleft, edgeright; 

def __getitem__(self, index):

        img=self.adv_flat[self.sample_num,:]

        if(self.shuff == False):
            # shuff is true for non-pgd attacks
            img = torch.from_numpy(np.reshape(img,(3,32,32)))
        else:
            img = torch.from_numpy(img).type(torch.FloatTensor)
        target = np.argmax(self.adv_dict["adv_labels"],axis=1)[self.sample_num]
        # doing this so that it is consistent with all other datasets
        # to return a PIL Image
        if self.transform is not None:
            img = self.transform(img)

        if self.target_transform is not None:
            target = self.target_transform(target)

        self.sample_num = self.sample_num + 1
        return img, target 

def __getitem__(self, index):

        img=self.adv_flat[self.sample_num,:]

        if(self.shuff == False):
            # shuff is true for non-pgd attacks
            img = torch.from_numpy(np.reshape(img,(3,32,32)))
        else:
            img = torch.from_numpy(img).type(torch.FloatTensor)
        target = np.argmax(self.adv_dict["adv_labels"],axis=1)[self.sample_num]
        # doing this so that it is consistent with all other datasets
        # to return a PIL Image
        if self.transform is not None:
            img = self.transform(img)

        if self.target_transform is not None:
            target = self.target_transform(target)

        self.sample_num = self.sample_num + 1
        return img, target 

def __getitem__(self, index):
        img=self.adv_flat[self.sample_num,:]
        if(self.transp == False):
            # shuff is true for non-pgd attacks
            img = torch.from_numpy(np.reshape(img,(28,28)))
        else:
            img = torch.from_numpy(img).type(torch.FloatTensor)
        target = np.argmax(self.adv_dict["adv_labels"],axis=1)[self.sample_num]
        # doing this so that it is consistent with all other datasets
        # to return a PIL Image

        if self.transform is not None:
            img = self.transform(img)
        if self.target_transform is not None:
            target = self.target_transform(target)
        self.sample_num = self.sample_num + 1
        return img, target 

def test_attack_strength(self):
        """
        If clipping is not done at each iteration (not passing clip_min and
        clip_max to fgm), this attack fails by
        np.mean(orig_labels == new_labels) == .39.
        """
        x_val = np.random.rand(100, 2)
        x_val = np.array(x_val, dtype=np.float32)

        x_adv = self.attack.generate_np(x_val, eps=1.0, ord=np.inf,
                                        clip_min=0.5, clip_max=0.7,
                                        nb_iter=5)

        orig_labs = np.argmax(self.sess.run(self.model(x_val)), axis=1)
        new_labs = np.argmax(self.sess.run(self.model(x_adv)), axis=1)
        self.assertTrue(np.mean(orig_labs == new_labs) < 0.1) 

def test_generate_np_targeted_gives_adversarial_example(self):
        x_val = np.random.rand(100, 2)
        x_val = np.array(x_val, dtype=np.float32)

        feed_labs = np.zeros((100, 2))
        feed_labs[np.arange(100), np.random.randint(0, 1, 100)] = 1
        x_adv = self.attack.generate_np(x_val, max_iterations=100,
                                        binary_search_steps=3,
                                        initial_const=1,
                                        clip_min=-5, clip_max=5,
                                        batch_size=100, y_target=feed_labs)

        new_labs = np.argmax(self.sess.run(self.model(x_adv)), axis=1)

        self.assertTrue(np.mean(np.argmax(feed_labs, axis=1) == new_labs)
                        > 0.9) 

def test_generate_gives_adversarial_example(self):

        x_val = np.random.rand(100, 2)
        x_val = np.array(x_val, dtype=np.float32)

        orig_labs = np.argmax(self.sess.run(self.model(x_val)), axis=1)
        feed_labs = np.zeros((100, 2))
        feed_labs[np.arange(100), orig_labs] = 1
        x = tf.placeholder(tf.float32, x_val.shape)
        y = tf.placeholder(tf.float32, feed_labs.shape)

        x_adv_p = self.attack.generate(x, max_iterations=100,
                                       binary_search_steps=3,
                                       initial_const=1,
                                       clip_min=-5, clip_max=5,
                                       batch_size=100, y=y)
        self.assertEqual(x_val.shape, x_adv_p.shape)
        x_adv = self.sess.run(x_adv_p, {x: x_val, y: feed_labs})

        new_labs = np.argmax(self.sess.run(self.model(x_adv)), axis=1)

        self.assertTrue(np.mean(orig_labs == new_labs) < 0.1) 

def test_generate_np_targeted_gives_adversarial_example(self):
        x_val = np.random.rand(100, 2)
        x_val = np.array(x_val, dtype=np.float32)

        feed_labs = np.zeros((100, 2))
        feed_labs[np.arange(100), np.random.randint(0, 1, 100)] = 1
        x_adv = self.attack.generate_np(x_val, max_iterations=100,
                                        binary_search_steps=3,
                                        initial_const=1,
                                        clip_min=-5, clip_max=5,
                                        batch_size=100, y_target=feed_labs)

        new_labs = np.argmax(self.sess.run(self.model(x_adv)), axis=1)

        self.assertTrue(np.mean(np.argmax(feed_labs, axis=1) == new_labs) >
                        0.9) 

def test_generate_gives_adversarial_example(self):

        x_val = np.random.rand(100, 2)
        x_val = np.array(x_val, dtype=np.float32)

        orig_labs = np.argmax(self.sess.run(self.model(x_val)), axis=1)
        feed_labs = np.zeros((100, 2))
        feed_labs[np.arange(100), orig_labs] = 1
        x = tf.placeholder(tf.float32, x_val.shape)
        y = tf.placeholder(tf.float32, feed_labs.shape)

        x_adv_p = self.attack.generate(x, max_iterations=100,
                                       binary_search_steps=3,
                                       initial_const=1,
                                       clip_min=-5, clip_max=5,
                                       batch_size=100, y=y)
        self.assertEqual(x_val.shape, x_adv_p.shape)
        x_adv = self.sess.run(x_adv_p, {x: x_val, y: feed_labs})

        new_labs = np.argmax(self.sess.run(self.model(x_adv)), axis=1)

        self.assertTrue(np.mean(orig_labs == new_labs) < 0.1) 

def test_generate_gives_adversarial_example(self):

        x_val = np.random.rand(100, 2)
        x_val = np.array(x_val, dtype=np.float32)

        orig_labs = np.argmax(self.sess.run(self.model(x_val)), axis=1)
        x = tf.placeholder(tf.float32, x_val.shape)

        x_adv_p = self.attack.generate(x, over_shoot=0.02, max_iter=50,
                                       nb_candidate=2, clip_min=-5, clip_max=5)
        self.assertEqual(x_val.shape, x_adv_p.shape)
        x_adv = self.sess.run(x_adv_p, {x: x_val})

        new_labs = np.argmax(self.sess.run(self.model(x_adv)), axis=1)

        self.assertTrue(np.mean(orig_labs == new_labs) < 0.1) 

def test_attack_strength(self):
        """
        If clipping is not done at each iteration (not using clip_min and
        clip_max), this attack fails by
        np.mean(orig_labels == new_labels) == .5
        """
        x_val = np.random.rand(100, 2)
        x_val = np.array(x_val, dtype=np.float32)

        x_adv = self.attack.generate_np(x_val, eps=1.0, eps_iter=0.05,
                                        clip_min=0.5, clip_max=0.7,
                                        nb_iter=5)

        orig_labs = np.argmax(self.sess.run(self.model(x_val)), axis=1)
        new_labs = np.argmax(self.sess.run(self.model(x_adv)), axis=1)
        self.assertTrue(np.mean(orig_labs == new_labs) < 0.1) 

def test_generate_targeted_gives_adversarial_example(self):
        x_val = np.random.rand(100, 2)
        x_val = np.array(x_val, dtype=np.float32)

        feed_labs = np.zeros((100, 2))
        feed_labs[np.arange(100), np.random.randint(0, 1, 100)] = 1
        x = tf.placeholder(tf.float32, x_val.shape)
        y = tf.placeholder(tf.float32, feed_labs.shape)

        x_adv_p = self.attack.generate(x, max_iterations=100,
                                       binary_search_steps=3,
                                       initial_const=1,
                                       clip_min=-5, clip_max=5,
                                       batch_size=100, y_target=y)
        self.assertEqual(x_val.shape, x_adv_p.shape)
        x_adv = self.sess.run(x_adv_p, {x: x_val, y: feed_labs})

        new_labs = np.argmax(self.sess.run(self.model(x_adv)), axis=1)

        self.assertTrue(np.mean(np.argmax(feed_labs, axis=1) == new_labs)
                        > 0.9) 

def model_argmax(sess, x, predictions, samples, feed=None):
    """
    Helper function that computes the current class prediction
    :param sess: TF session
    :param x: the input placeholder
    :param predictions: the model's symbolic output
    :param samples: numpy array with input samples (dims must match x)
    :param feed: An optional dictionary that is appended to the feeding
             dictionary before the session runs. Can be used to feed
             the learning phase of a Keras model for instance.
    :return: the argmax output of predictions, i.e. the current predicted class
    """
    feed_dict = {x: samples}
    if feed is not None:
        feed_dict.update(feed)
    probabilities = sess.run(predictions, feed_dict)

    if samples.shape[0] == 1:
        return np.argmax(probabilities)
    else:
        return np.argmax(probabilities, axis=1) 

def predict(limit):
    _limit = limit if limit > 0 else 5

    td = TrainingData(LABEL_FILE, img_root=IMAGES_ROOT, mean_image_file=MEAN_IMAGE_FILE, image_property=IMAGE_PROP)
    label_def = LabelingMachine.read_label_def(LABEL_DEF_FILE)
    model = alex.Alex(len(label_def))
    serializers.load_npz(MODEL_FILE, model)

    i = 0
    for arr, im in td.generate():
        x = np.ndarray((1,) + arr.shape, arr.dtype)
        x[0] = arr
        x = chainer.Variable(np.asarray(x), volatile="on")
        y = model.predict(x)
        p = np.argmax(y.data)
        print("predict {0}, actual {1}".format(label_def[p], label_def[im.label]))
        im.image.show()
        i += 1
        if i >= _limit:
            break 

def _get_bp_indexes_labranchor(self, soi):
        """
        Get indexes of branch point regions in given sequences.

        :param soi: batch of sequences of interest for introns (intron-3..intron+6)
        :return: array of predicted bp indexes
        """
        encoded = [onehot(str(seq)[self.acc_i - 70:self.acc_i]) for seq in np.nditer(soi)]
        labr_in = np.stack(encoded, axis=0)
        out = self.labranchor.predict_on_batch(labr_in)
        # for each row, pick the base with max branchpoint probability, and get its index
        max_indexes = np.apply_along_axis(lambda x: self.acc_i - 70 + np.argmax(x), axis=1, arr=out)
        # self.write_bp(max_indexes)
        return max_indexes

# TODO boilerplate
#    def write_bp(self, max_indexes):
#        max_indexes = [str(seq) for seq in np.nditer(max_indexes)]
#        with open(''.join([this_dir, "/../customBP/example_files/bp_idx_chr21_labr.txt"]), "a") as bp_idx_file:
#            bp_idx_file.write('\n'.join(max_indexes))
#            bp_idx_file.write('\n')
#            bp_idx_file.close() 

def act(self, s):
        eps = self.getEpsilon()
        global frames
        frames = frames + 1

        if random.random() < eps:
            return random.randint(0, NUM_ACTIONS - 1)

        else:
            s = np.array([s])
            p = brain.predict_p(s)[0]

            # a = np.argmax(p)
            a = np.random.choice(NUM_ACTIONS, p=p)

            return a 

def choose_action(random_action, probability):
    random_value = randint(0, 5) #np.random.uniform()

    if (random_action):
        return random_value
    else:
        return np.argmax(probability)
        # if np.argmax(probability) == 1:
        #     return 4
        # else:
        #     return 5


    # if random_value < probability:
    #     # signifies up in openai gym
    #     return 2
    # else:
    #     # signifies down in openai gym
    #     return 3 

def choose_action(random_action, probability):
    random_value = randint(0, 5) #np.random.uniform()

    if (random_action):
        return random_value
    else:
        return np.argmax(probability)
        # if np.argmax(probability) == 1:
        #     return 4
        # else:
        #     return 5


    # if random_value < probability:
    #     # signifies up in openai gym
    #     return 2
    # else:
    #     # signifies down in openai gym
    #     return 3 

def choose_action(random_action, probability):
    random_value = randint(4, 5) #np.random.uniform()

    if (random_action):
        return random_value
    else:
        if np.argmax(probability) == 1:
            return 4
        else:
            return 5


    # if random_value < probability:
    #     # signifies up in openai gym
    #     return 2
    # else:
    #     # signifies down in openai gym
    #     return 3 

def forward(self, x):
        N, C, H, W = x.shape
        out_h = int(1 + (H - self.pool_h) / self.stride)
        out_w = int(1 + (W - self.pool_w) / self.stride)

        col = im2col(x, self.pool_h, self.pool_w, self.stride, self.pad)
        col = col.reshape(-1, self.pool_h * self.pool_w)

        arg_max = np.argmax(col, axis=1)
        out = np.max(col, axis=1)
        out = out.reshape(N, out_h, out_w, C).transpose(0, 3, 1, 2)

        self.x = x
        self.arg_max = arg_max

        return out 

def predict_all(X, all_theta):
    rows = X.shape[0]
    params = X.shape[1]
    num_labels = all_theta.shape[0]
    
    # same as before, insert ones to match the shape
    X = np.insert(X, 0, values=np.ones(rows), axis=1)
    
    # convert to matrices
    X = np.matrix(X)
    all_theta = np.matrix(all_theta)
    
    # compute the class probability for each class on each training instance
    h = sigmoid(X * all_theta.T)
    
    # create array of the index with the maximum probability
    h_argmax = np.argmax(h, axis=1)
    
    # because our array was zero-indexed we need to add one for the true label prediction
    h_argmax = h_argmax + 1
    
    return h_argmax 

def rocstories(data_dir, pred_path, log_path):
    preds = pd.read_csv(pred_path, delimiter='\t')[
        'prediction'].values.tolist()
    _, _, _, labels = _rocstories(os.path.join(
        data_dir, 'cloze_test_test__spring2016 - cloze_test_ALL_test.csv'))
    test_accuracy = accuracy_score(labels, preds) * 100.
    logs = [json.loads(line) for line in open(log_path)][1:]
    best_validation_index = np.argmax([log['va_acc'] for log in logs])
    valid_accuracy = logs[best_validation_index]['va_acc']
    print('ROCStories Valid Accuracy: %.2f' % (valid_accuracy))
    print('ROCStories Test Accuracy:  %.2f' % (test_accuracy)) 

def sst(data_dir, pred_path, log_path):
    preds = pd.read_csv(pred_path, delimiter='\t')[
        'prediction'].values.tolist()
    test_url = 'https://raw.githubusercontent.com/harvardnlp/sent-conv-torch/master/data/stsa.binary.test'
    path = chainer.dataset.cached_download(test_url)
    teX, teY = _sst(path)
    labels = teY
    test_accuracy = accuracy_score(labels, preds) * 100.
    logs = [json.loads(line) for line in open(log_path)][1:]
    best_validation_index = np.argmax([log['va_acc'] for log in logs])
    valid_accuracy = logs[best_validation_index]['va_acc']
    print('SST Valid Accuracy: %.2f' % (valid_accuracy))
    print('SST Test Accuracy:  %.2f' % (test_accuracy)) 

def argmax(x): return np.argmax(x, 1) 

def __init__(self, labels = [], predictions = [], gold = [], one_hot_encoding = False, class_indices = False):
		# rows are true labels, columns predictions
		self.matrix = np.zeros(shape = (len(labels), len(labels)))
		self.labels = labels

		if len(predictions) != len(gold):
			raise ValueError("Predictions and gold labels do not have the same count.")
		for i in range(len(predictions)):
			index_pred = np.argmax(predictions[i]) if one_hot_encoding else (predictions[i] if class_indices else labels.index(predictions[i]))
			index_gold = np.argmax(gold[i]) if one_hot_encoding else (gold[i] if class_indices else labels.index(gold[i]))
			self.matrix[index_gold][index_pred] += 1

		if len(predictions) > 0: 
			self.compute_all_scores() 

def get_accuracy_seqlab(golds, preds):
	correct = 0.0
	total = 0.0
	for i in range(len(golds)):
		for j in range(len(golds[i])):
			if np.count_nonzero(golds[i][j]) > 0:
				total += 1.0
				lgold = np.argmax(golds[i][j])
				lpred = np.argmax(preds[i][j])
				if lgold == lpred:
					correct += 1.0
	acc = correct / total;
	return acc 

def inference(self, scores, sequence_lengths=None):
        """
        Inference label sequence given scores.
        If transitions is given, then perform veterbi search, else perform greedy search.

        Args:
            scores: A numpy array with shape (batch, max_length, num_tags).
            sequence_lengths: A numpy array with shape (batch,).

        Returns:
            A numpy array with shape (batch, max_length).
        """

        if not self.parameters['use_crf']:
            return np.argmax(scores, 2)
        else:
            with tf.variable_scope(self.scope, reuse=True):
                transitions = tf.get_variable('transitions').eval(session=self.sess)
            paths = np.zeros(scores.shape[:2], dtype=INT_TYPE)
            for i in xrange(scores.shape[0]):
                tag_score, length = scores[i], sequence_lengths[i]
                if length == 0:
                    continue
                path, _ = crf.viterbi_decode(tag_score[:length], transitions)
                paths[i, :length] = path
            return paths 

def fetch_default_score(ranking_func_name):
    if ranking_func_name.count('argmax'):
        return -1.0 * np.inf
    elif ranking_func_name.count('argmin'):
        return +1.0 * np.inf
    else:
        raise ValueError("Unrecognized ranking_func_name %s" % ranking_func_name) 

def get_score_ranking_function_for_colname(score_name):
    if score_name.count("AUC"):
        return np.argmax
    elif score_name.count("LOSS"):
        return np.argmin
    else:
        return np.argmin 

def extract_quaternion(R):
    d = np.diagonal(R)
    t = np.sum(d)
    if t + 1 < 0.25:
        symmetric_mat = R + R.T
        asymmetric_mat = R - R.T
        symmetric_diag = np.diagonal(symmetric_mat)
        i_max = np.argmax(symmetric_diag)
        q = np.empty(4)
        if i_max == 0:
            q[1] = np.sqrt(symmetric_diag[0] - t + 1) / 2
            normalizer = 1 / q[1]
            q[2] = symmetric_mat[1, 0] / 4 * normalizer
            q[3] = symmetric_mat[2, 0] / 4 * normalizer
            q[0] = asymmetric_mat[2, 1] / 4 * normalizer
        elif i_max == 1:
            q[2] = np.sqrt(symmetric_diag[1] - t + 1) / 2
            normalizer = 1 / q[2]
            q[1] = symmetric_mat[1, 0] / 4 * normalizer
            q[3] = symmetric_mat[2, 1] / 4 * normalizer
            q[0] = asymmetric_mat[0, 2] / 4 * normalizer
        elif i_max == 2:
            q[3] = np.sqrt(symmetric_diag[2] - t + 1) / 2
            normalizer = 1 / q[3]
            q[1] = symmetric_mat[2, 0] / 4 * normalizer
            q[2] = symmetric_mat[1, 2] / 4 * normalizer
            q[0] = asymmetric_mat[1, 0] / 4 * normalizer
    else:
        r = np.sqrt(1+t)
        s = 0.5 / r
        q = np.array([0.5*r, (R[2, 1] - R[1, 2])*s, (R[0, 2] - R[2, 0])*s, (R[1, 0] - R[0, 1])*s])

    return q 

def error_rate(predictions, labels):
    """
    Return the error rate based on dense predictions and 1-hot labels.
    """

    return 100.0 - ( 100.0 *
        np.sum(np.argmax(predictions, 3) == np.argmax(labels, 3)) /
        (predictions.shape[0]*predictions.shape[1]*predictions.shape[2])) 

def predict(self, X, **args):
        svc = RFFSVC(self.dim, self.std, self.W, **args)
        return np.argmax(np.dot(svc.conv(X), self.coef.T) + self.icpt.flatten(), axis = 1) 

def blind_search(self, x, y, image):
        '''
        This function is applied in the first few frames and/or if the lane was not successfully detected
        in the previous frame. It uses a slinding window approach to detect peaks in a histogram of the
        binary thresholded image. Pixels in close proimity to the detected peaks are considered to belong
        to the lane lines.
        '''
        xvals = []
        yvals = []
        if self.found == False:
            i = 720
            j = 630
            histogram = np.sum(image[image.shape[0]//2:], axis=0)
            if self == Right:
                peak = np.argmax(histogram[image.shape[1]//2:]) + image.shape[1]//2
            else:
                peak = np.argmax(histogram[:image.shape[1]//2])
            while j >= 0:
                x_idx = np.where((((peak - 100) < x)&(x < (peak + 100))&((y > j) & (y < i))))
                x_window, y_window = x[x_idx], y[x_idx]
                if np.sum(x_window) != 0:
                    xvals.extend(x_window)
                    yvals.extend(y_window)
                if np.sum(x_window) > 100:
                    peak = np.int(np.mean(x_window))
                i -= 90
                j -= 90
        if np.sum(xvals) > 0:
            self.found = True
        else:
            yvals = self.Y
            xvals = self.X
        return xvals, yvals, self.found 

def attack(self, X, Y):
        """
        Perform the L_2 attack on the given images for the given targets.

        :param X: samples to generate advs
        :param Y: the original class labels
        If self.targeted is true, then the targets represents the target labels.
        If self.targeted is false, then targets are the original class labels.
        """
        nb_classes = Y.shape[1]

        # random select target class for targeted attack
        y_target = np.copy(Y)
        if self.TARGETED:
            for i in range(Y.shape[0]):
                current = int(np.argmax(Y[i]))
                target = np.random.choice(other_classes(nb_classes, current))
                y_target[i] = np.eye(nb_classes)[target]

        X_adv = np.zeros_like(X)
        for i in tqdm(range(0, X.shape[0], self.batch_size)):
            start = i
            end = i + self.batch_size
            end = np.minimum(end, X.shape[0])
            X_adv[start:end] = self.attack_batch(X[start:end], y_target[start:end])

        return X_adv 

def attack(self, X, Y):
        """
        Perform the L_2 attack on the given images for the given targets.

        :param X: samples to generate advs
        :param Y: the original class labels
        If self.targeted is true, then the targets represents the target labels.
        If self.targeted is false, then targets are the original class labels.
        """
        nb_classes = Y.shape[1]

        # random select target class for targeted attack
        y_target = np.copy(Y)
        if self.TARGETED:
            for i in range(Y.shape[0]):
                current = int(np.argmax(Y[i]))
                target = np.random.choice(other_classes(nb_classes, current))
                y_target[i] = np.eye(nb_classes)[target]

        X_adv = np.zeros_like(X)
        for i in tqdm(range(0, X.shape[0], self.batch_size)):
            start = i
            end = i + self.batch_size
            end = np.minimum(end, X.shape[0])
            X_adv[start:end] = self.attack_batch(X[start:end], y_target[start:end])

        return X_adv 

def saliency_map_method(sess, model, X, Y, theta, gamma, clip_min=None,
                        clip_max=None):
    """
    TODO
    :param sess:
    :param model: predictions or after-softmax
    :param X:
    :param Y:
    :param theta:
    :param gamma:
    :param clip_min:
    :param clip_max:
    :return:
    """
    nb_classes = Y.shape[1]
    # Define TF placeholder for the input
    x = tf.placeholder(tf.float32, shape=(None,) + X.shape[1:])
    # Define model gradients
    grads = jacobian_graph(model(x), x, nb_classes)
    X_adv = np.zeros_like(X)
    for i in tqdm(range(len(X))):
        current_class = int(np.argmax(Y[i]))
        target_class = np.random.choice(other_classes(nb_classes, current_class))
        X_adv[i], _, _ = jsma(
            sess, x, model(x), grads, X[i:(i+1)], target_class, theta=theta,
            gamma=gamma, increase=True, nb_classes=nb_classes,
            clip_min=clip_min, clip_max=clip_max
        )

    return X_adv 

def binary_refinement(sess,Best_X_adv,
                      X_adv, Y, ALPHA, ub, lb, model, dataset='cifar'):
    num_samples = np.shape(X_adv)[0]
    print(dataset)
    if(dataset=="mnist"):
        X_place = tf.placeholder(tf.float32, shape=[1, 1, 28, 28])
    else:
        X_place = tf.placeholder(tf.float32, shape=[1, 3, 32, 32])

    pred = model(X_place)
    for i in range(num_samples):
        logits_op = sess.run(pred,feed_dict={X_place:X_adv[i:i+1,:,:,:]})
        if(not np.argmax(logits_op) == np.argmax(Y[i,:])):
            # Success, increase alpha
            Best_X_adv[i,:,:,:] = X_adv[i,:,:,]
            lb[i] = ALPHA[i,0]
        else:
            ub[i] = ALPHA[i,0]
        ALPHA[i] = 0.5*(lb[i] + ub[i])
    return ALPHA, Best_X_adv 

def help_generate_np_gives_adversarial_example(self, ord, eps=.5, **kwargs):
        x_val, x_adv, delta = self.generate_adversarial_examples_np(ord, eps,
                                                                    **kwargs)
        self.assertClose(delta, eps)
        orig_labs = np.argmax(self.sess.run(self.model(x_val)), axis=1)
        new_labs = np.argmax(self.sess.run(self.model(x_adv)), axis=1)
        self.assertTrue(np.mean(orig_labs == new_labs) < 0.5) 

def test_targeted_generate_np_gives_adversarial_example(self):
        random_labs = np.random.random_integers(0, 1, 100)
        random_labs_one_hot = np.zeros((100, 2))
        random_labs_one_hot[np.arange(100), random_labs] = 1

        _, x_adv, delta = self.generate_adversarial_examples_np(
            eps=.5, ord=np.inf, y_target=random_labs_one_hot)

        self.assertClose(delta, 0.5)

        new_labs = np.argmax(self.sess.run(self.model(x_adv)), axis=1)
        self.assertTrue(np.mean(random_labs == new_labs) > 0.7) 

def test_generate_np_does_not_cache_graph_computation_for_nb_iter(self):

        x_val = np.random.rand(100, 2)
        x_val = np.array(x_val, dtype=np.float32)

        x_adv = self.attack.generate_np(x_val, eps=1.0, ord=np.inf,
                                        clip_min=-5.0, clip_max=5.0,
                                        nb_iter=10)

        orig_labs = np.argmax(self.sess.run(self.model(x_val)), axis=1)
        new_labs = np.argmax(self.sess.run(self.model(x_adv)), axis=1)
        self.assertTrue(np.mean(orig_labs == new_labs) < 0.1)

        ok = [False]
        old_grads = tf.gradients

        def fn(*x, **y):
            ok[0] = True
            return old_grads(*x, **y)
        tf.gradients = fn

        x_adv = self.attack.generate_np(x_val, eps=1.0, ord=np.inf,
                                        clip_min=-5.0, clip_max=5.0,
                                        nb_iter=11)

        orig_labs = np.argmax(self.sess.run(self.model(x_val)), axis=1)
        new_labs = np.argmax(self.sess.run(self.model(x_adv)), axis=1)
        self.assertTrue(np.mean(orig_labs == new_labs) < 0.1)

        tf.gradients = old_grads

        self.assertTrue(ok[0]) 

def test_generate_np_untargeted_gives_adversarial_example(self):
        x_val = np.random.rand(100, 2)
        x_val = np.array(x_val, dtype=np.float32)

        x_adv = self.attack.generate_np(x_val, max_iterations=100,
                                        binary_search_steps=3,
                                        initial_const=1,
                                        clip_min=-5, clip_max=5,
                                        batch_size=10)

        orig_labs = np.argmax(self.sess.run(self.model(x_val)), axis=1)
        new_labs = np.argmax(self.sess.run(self.model(x_adv)), axis=1)

        self.assertTrue(np.mean(orig_labs == new_labs) < 0.1) 

def test_generate_np_high_confidence_untargeted_examples(self):

        trivial_model = TrivialModel()

        for CONFIDENCE in [0, 2.3]:
            x_val = np.random.rand(10, 1) - .5
            x_val = np.array(x_val, dtype=np.float32)

            orig_labs = np.argmax(self.sess.run(trivial_model.get_logits(x_val)), axis=1)
            attack = CarliniWagnerL2(trivial_model, sess=self.sess)
            x_adv = attack.generate_np(x_val,
                                       max_iterations=100,
                                       binary_search_steps=2,
                                       learning_rate=1e-2,
                                       initial_const=1,
                                       clip_min=-10, clip_max=10,
                                       confidence=CONFIDENCE,
                                       batch_size=10)

            new_labs = self.sess.run(trivial_model.get_logits(x_adv))

            good_labs = new_labs[np.arange(10), 1 - orig_labs]
            bad_labs = new_labs[np.arange(10), orig_labs]

            self.assertTrue(np.mean(np.argmax(new_labs, axis=1) == orig_labs)
                            == 0)
            self.assertTrue(np.isclose(
                0, np.min(good_labs - (bad_labs + CONFIDENCE)), atol=1e-1)) 

def test_generate_np_untargeted_gives_adversarial_example(self):
        x_val = np.random.rand(100, 2)
        x_val = np.array(x_val, dtype=np.float32)

        x_adv = self.attack.generate_np(x_val, max_iterations=100,
                                        binary_search_steps=3,
                                        initial_const=1,
                                        clip_min=-5, clip_max=5,
                                        batch_size=10)

        orig_labs = np.argmax(self.sess.run(self.model(x_val)), axis=1)
        new_labs = np.argmax(self.sess.run(self.model(x_adv)), axis=1)

        self.assertTrue(np.mean(orig_labs == new_labs) < 0.1)