Python numpy.int32() Examples

def rocstories(data_dir, n_train=1497, n_valid=374):
    storys, comps1, comps2, ys = _rocstories(os.path.join(
        data_dir, 'cloze_test_val__spring2016 - cloze_test_ALL_val.csv'))
    teX1, teX2, teX3, _ = _rocstories(os.path.join(
        data_dir, 'cloze_test_test__spring2016 - cloze_test_ALL_test.csv'))
    tr_storys, va_storys, tr_comps1, va_comps1, tr_comps2, va_comps2, tr_ys, va_ys = train_test_split(
        storys, comps1, comps2, ys, test_size=n_valid, random_state=seed)
    trX1, trX2, trX3 = [], [], []
    trY = []
    for s, c1, c2, y in zip(tr_storys, tr_comps1, tr_comps2, tr_ys):
        trX1.append(s)
        trX2.append(c1)
        trX3.append(c2)
        trY.append(y)

    vaX1, vaX2, vaX3 = [], [], []
    vaY = []
    for s, c1, c2, y in zip(va_storys, va_comps1, va_comps2, va_ys):
        vaX1.append(s)
        vaX2.append(c1)
        vaX3.append(c2)
        vaY.append(y)
    trY = np.asarray(trY, dtype=np.int32)
    vaY = np.asarray(vaY, dtype=np.int32)
    return (trX1, trX2, trX3, trY), (vaX1, vaX2, vaX3, vaY), (teX1, teX2, teX3) 

def sst():
    train_url = 'https://raw.githubusercontent.com/harvardnlp/sent-conv-torch/master/data/stsa.binary.train'
    valid_url = 'https://raw.githubusercontent.com/harvardnlp/sent-conv-torch/master/data/stsa.binary.dev'
    test_url = 'https://raw.githubusercontent.com/harvardnlp/sent-conv-torch/master/data/stsa.binary.test'

    path = chainer.dataset.cached_download(train_url)
    trX, trY = _sst(path)
    sys.stderr.write('train data is {}\n'.format(path))
    path = chainer.dataset.cached_download(valid_url)
    vaX, vaY = _sst(path)
    sys.stderr.write('valid data is {}\n'.format(path))
    path = chainer.dataset.cached_download(test_url)
    teX, teY = _sst(path)
    sys.stderr.write('test data is {}\n'.format(path))

    trY = np.asarray(trY, dtype=np.int32)
    vaY = np.asarray(vaY, dtype=np.int32)
    teY = np.asarray(teY, dtype=np.int32)
    return (trX, trY), (vaX, vaY), (teX, teY) 

def load_keypoints(image_filepath, image_height, image_width):
    """Load facial keypoints of one image."""
    fp_keypoints = "%s.cat" % (image_filepath,)
    if not os.path.isfile(fp_keypoints):
        raise Exception("Could not find keypoint coordinates for image '%s'." \
                        % (image_filepath,))
    else:
        coords_raw = open(fp_keypoints, "r").readlines()[0].strip().split(" ")
        coords_raw = [abs(int(coord)) for coord in coords_raw]
        keypoints = []
        #keypoints_arr = np.zeros((9*2,), dtype=np.int32)
        for i in range(1, len(coords_raw), 2): # first element is the number of coords
            x = np.clip(coords_raw[i], 0, image_width-1)
            y = np.clip(coords_raw[i+1], 0, image_height-1)
            keypoints.append((x, y))

        return keypoints 

def batch_iter(data, batch_size, num_epochs, shuffle = True):
		"""
		Generates a batch iterator for a dataset.
		"""
		#data = np.array(data, dtype = np.int32)
		data_size = len(data)

		num_batches_per_epoch = int(data_size/batch_size) + 1
		for epoch in range(num_epochs):
			# Shuffle the data at each epoch
			if shuffle:
				#shuffle_indices = np.random.permutation(np.arange(data_size))
				#shuffled_data = data[shuffle_indices]
				random.shuffle(data)
			#else:
			#	shuffled_data = data

			for batch_num in range(num_batches_per_epoch):
				start_index = batch_num * batch_size
				end_index = min((batch_num + 1) * batch_size, data_size)
				yield data[start_index:end_index] 

def in_top_k(predictions, targets, k):
    '''Returns whether the `targets` are in the top `k` `predictions`

    # Arguments
        predictions: A tensor of shape batch_size x classess and type float32.
        targets: A tensor of shape batch_size and type int32 or int64.
        k: An int, number of top elements to consider.

    # Returns
        A tensor of shape batch_size and type int. output_i is 1 if
        targets_i is within top-k values of predictions_i
    '''
    predictions_top_k = T.argsort(predictions)[:, -k:]
    result, _ = theano.map(lambda prediction, target: any(equal(prediction, target)), sequences=[predictions_top_k, targets])
    return result


# CONVOLUTIONS 

def ctc_path_probs(predict, Y, alpha=1e-4):
    smoothed_predict = (1 - alpha) * predict[:, Y] + alpha * np.float32(1.) / Y.shape[0]
    L = T.log(smoothed_predict)
    zeros = T.zeros_like(L[0])
    log_first = zeros

    f_skip_idxs = ctc_create_skip_idxs(Y)
    b_skip_idxs = ctc_create_skip_idxs(Y[::-1])  # there should be a shortcut to calculating this

    def step(log_f_curr, log_b_curr, f_active, log_f_prev, b_active, log_b_prev):
        f_active_next, log_f_next = ctc_update_log_p(f_skip_idxs, zeros, f_active, log_f_curr, log_f_prev)
        b_active_next, log_b_next = ctc_update_log_p(b_skip_idxs, zeros, b_active, log_b_curr, log_b_prev)
        return f_active_next, log_f_next, b_active_next, log_b_next

    [f_active, log_f_probs, b_active, log_b_probs], _ = theano.scan(
        step, sequences=[L, L[::-1, ::-1]], outputs_info=[np.int32(1), log_first, np.int32(1), log_first])

    idxs = T.arange(L.shape[1]).dimshuffle('x', 0)
    mask = (idxs < f_active.dimshuffle(0, 'x')) & (idxs < b_active.dimshuffle(0, 'x'))[::-1, ::-1]
    log_probs = log_f_probs + log_b_probs[::-1, ::-1] - L
    return log_probs, mask 

def create_roidb_from_box_list(self, box_list, gt_roidb):
    assert len(box_list) == self.num_images, \
            'Number of boxes must match number of ground-truth images'
    roidb = []

    if gt_roidb is not None:
        for i in range(self.num_images):
            boxes = box_list[i]

            real_label = gt_roidb[i]['labels']

            roidb.append({'boxes' : boxes,
                          'labels' : np.array([real_label], dtype=np.int32),
                          'flipped' : False})
    else:
        for i in range(self.num_images):
            boxes = box_list[i]

            roidb.append({'boxes' : boxes,
                          'labels' : np.zeros((1, 0), dtype=np.int32),
                          'flipped' : False})

    return roidb 

def generate_anchors_pre(height, width, feat_stride, anchor_scales=(8,16,32), anchor_ratios=(0.5,1,2)):
  """ A wrapper function to generate anchors given different scales
    Also return the number of anchors in variable 'length'
  """
  anchors = generate_anchors(ratios=np.array(anchor_ratios), scales=np.array(anchor_scales))
  A = anchors.shape[0]
  shift_x = np.arange(0, width) * feat_stride
  shift_y = np.arange(0, height) * feat_stride
  shift_x, shift_y = np.meshgrid(shift_x, shift_y)
  shifts = np.vstack((shift_x.ravel(), shift_y.ravel(), shift_x.ravel(), shift_y.ravel())).transpose()
  K = shifts.shape[0]
  # width changes faster, so here it is H, W, C
  anchors = anchors.reshape((1, A, 4)) + shifts.reshape((1, K, 4)).transpose((1, 0, 2))
  anchors = anchors.reshape((K * A, 4)).astype(np.float32, copy=False)
  length = np.int32(anchors.shape[0])

  return anchors, length 

def init_topics_KV__rand_active_words(
        n_states=10,
        frac_words_active=0.5,
        blend_frac_active=0.5,
        n_vocabs=144,
        seed=0):
    prng = np.random.RandomState(int(seed))
    unif_topics_KV = np.ones((n_states, n_vocabs)) / float(n_vocabs)
    active_topics_KV = np.zeros((n_states, n_vocabs))
    for k in xrange(n_states):
        active_words_U = prng.choice(
            np.arange(n_vocabs, dtype=np.int32),
            int(frac_words_active * n_vocabs),
            replace=False)
        active_topics_KV[k, active_words_U] = 1.0 / active_words_U.size
    topics_KV = (1 - blend_frac_active) * unif_topics_KV \
        + blend_frac_active * active_topics_KV
    topics_KV /= topics_KV.sum(axis=1)[:,np.newaxis]
    return topics_KV 

def init_topics_KV__rand_docs(
        dataset=None,
        n_states=10,
        n_vocabs=144,
        blend_frac_doc=0.5,
        seed=0):
    prng = np.random.RandomState(int(seed))
    unif_topics_KV = np.ones((n_states, n_vocabs)) / float(n_vocabs)
    doc_KV = np.zeros((n_states, n_vocabs))
    chosen_doc_ids = prng.choice(
        np.arange(dataset['n_docs'], dtype=np.int32),
        n_states,
        replace=False)
    for k in xrange(n_states):
        start_d = dataset['doc_indptr_Dp1'][chosen_doc_ids[k]]
        stop_d = dataset['doc_indptr_Dp1'][chosen_doc_ids[k] + 1]
        active_words_U = dataset['word_id_U'][start_d:stop_d]
        doc_KV[k, active_words_U] = dataset['word_ct_U'][start_d:stop_d]
    doc_KV /= doc_KV.sum(axis=1)[:,np.newaxis]
    topics_KV = (1 - blend_frac_doc) * unif_topics_KV \
        + blend_frac_doc * doc_KV
    topics_KV /= topics_KV.sum(axis=1)[:,np.newaxis]
    return topics_KV 

def generate_anchors_pre(height, width, feat_stride, anchor_scales=(8, 16, 32), anchor_ratios=(0.5, 1, 2)):
    """ A wrapper function to generate anchors given different scales
      Also return the number of anchors in variable 'length'
    """
    anchors = generate_anchors(ratios=np.array(anchor_ratios), scales=np.array(anchor_scales))
    A = anchors.shape[0]
    shift_x = np.arange(0, width) * feat_stride
    shift_y = np.arange(0, height) * feat_stride
    shift_x, shift_y = np.meshgrid(shift_x, shift_y)
    shifts = np.vstack((shift_x.ravel(), shift_y.ravel(), shift_x.ravel(), shift_y.ravel())).transpose()
    K = shifts.shape[0]
    # width changes faster, so here it is H, W, C
    anchors = anchors.reshape((1, A, 4)) + shifts.reshape((1, K, 4)).transpose((1, 0, 2))
    anchors = anchors.reshape((K * A, 4)).astype(np.float32, copy=False)
    length = np.int32(anchors.shape[0])

    return anchors, length 

def create_cifar100(tfrecord_dir, cifar100_dir):
    print('Loading CIFAR-100 from "%s"' % cifar100_dir)
    import pickle
    with open(os.path.join(cifar100_dir, 'train'), 'rb') as file:
        data = pickle.load(file, encoding='latin1')
    images = data['data'].reshape(-1, 3, 32, 32)
    labels = np.array(data['fine_labels'])
    assert images.shape == (50000, 3, 32, 32) and images.dtype == np.uint8
    assert labels.shape == (50000,) and labels.dtype == np.int32
    assert np.min(images) == 0 and np.max(images) == 255
    assert np.min(labels) == 0 and np.max(labels) == 99
    onehot = np.zeros((labels.size, np.max(labels) + 1), dtype=np.float32)
    onehot[np.arange(labels.size), labels] = 1.0

    with TFRecordExporter(tfrecord_dir, images.shape[0]) as tfr:
        order = tfr.choose_shuffled_order()
        for idx in range(order.size):
            tfr.add_image(images[order[idx]])
        tfr.add_labels(onehot[order])

#---------------------------------------------------------------------------- 

def __init__(self, resolution=1024, num_channels=3, dtype='uint8', dynamic_range=[0,255], label_size=0, label_dtype='float32'):
        self.resolution         = resolution
        self.resolution_log2    = int(np.log2(resolution))
        self.shape              = [num_channels, resolution, resolution]
        self.dtype              = dtype
        self.dynamic_range      = dynamic_range
        self.label_size         = label_size
        self.label_dtype        = label_dtype
        self._tf_minibatch_var  = None
        self._tf_lod_var        = None
        self._tf_minibatch_np   = None
        self._tf_labels_np      = None

        assert self.resolution == 2 ** self.resolution_log2
        with tf.name_scope('Dataset'):
            self._tf_minibatch_var = tf.Variable(np.int32(0), name='minibatch_var')
            self._tf_lod_var = tf.Variable(np.int32(0), name='lod_var') 

def _get_area_ratio(self, pos_proposals, pos_assigned_gt_inds, gt_masks):
        """Compute area ratio of the gt mask inside the proposal and the gt
        mask of the corresponding instance"""
        num_pos = pos_proposals.size(0)
        if num_pos > 0:
            area_ratios = []
            proposals_np = pos_proposals.cpu().numpy()
            pos_assigned_gt_inds = pos_assigned_gt_inds.cpu().numpy()
            # compute mask areas of gt instances (batch processing for speedup)
            gt_instance_mask_area = gt_masks.sum((-1, -2))
            for i in range(num_pos):
                gt_mask = gt_masks[pos_assigned_gt_inds[i]]

                # crop the gt mask inside the proposal
                x1, y1, x2, y2 = proposals_np[i, :].astype(np.int32)
                gt_mask_in_proposal = gt_mask[y1:y2 + 1, x1:x2 + 1]

                ratio = gt_mask_in_proposal.sum() / (
                    gt_instance_mask_area[pos_assigned_gt_inds[i]] + 1e-7)
                area_ratios.append(ratio)
            area_ratios = torch.from_numpy(np.stack(area_ratios)).float().to(
                pos_proposals.device)
        else:
            area_ratios = pos_proposals.new_zeros((0, ))
        return area_ratios 

def get_cls_results(det_results, gt_bboxes, gt_labels, gt_ignore, class_id):
    """Get det results and gt information of a certain class."""
    cls_dets = [det[class_id]
                for det in det_results]  # det bboxes of this class
    cls_gts = []  # gt bboxes of this class
    cls_gt_ignore = []
    for j in range(len(gt_bboxes)):
        gt_bbox = gt_bboxes[j]
        cls_inds = (gt_labels[j] == class_id + 1)
        cls_gt = gt_bbox[cls_inds, :] if gt_bbox.shape[0] > 0 else gt_bbox
        cls_gts.append(cls_gt)
        if gt_ignore is None:
            cls_gt_ignore.append(np.zeros(cls_gt.shape[0], dtype=np.int32))
        else:
            cls_gt_ignore.append(gt_ignore[j][cls_inds])
    return cls_dets, cls_gts, cls_gt_ignore 

def _load_dataset_clipping(self, dataset_dir, epsilon):
    """Helper method which loads dataset and determines clipping range.

    Args:
      dataset_dir: location of the dataset.
      epsilon: maximum allowed size of adversarial perturbation.
    """
    self.dataset_max_clip = {}
    self.dataset_min_clip = {}
    self._dataset_image_count = 0
    for fname in os.listdir(dataset_dir):
      if not fname.endswith('.png'):
        continue
      image_id = fname[:-4]
      image = np.array(
          Image.open(os.path.join(dataset_dir, fname)).convert('RGB'))
      image = image.astype('int32')
      self._dataset_image_count += 1
      self.dataset_max_clip[image_id] = np.clip(image + epsilon,
                                                0,
                                                255).astype('uint8')
      self.dataset_min_clip[image_id] = np.clip(image - epsilon,
                                                0,
                                                255).astype('uint8') 

def load_images(input_dir, metadata_file_path, batch_shape):
    """Retrieve numpy arrays of images and labels, read from a directory."""
    num_images = batch_shape[0]
    with open(metadata_file_path) as input_file:
        reader = csv.reader(input_file)
        header_row = next(reader)
        rows = list(reader)

    row_idx_image_id = header_row.index('ImageId')
    row_idx_true_label = header_row.index('TrueLabel')
    images = np.zeros(batch_shape)
    labels = np.zeros(num_images, dtype=np.int32)
    for idx in xrange(num_images):
        row = rows[idx]
        filepath = os.path.join(input_dir, row[row_idx_image_id] + '.png')

        with tf.gfile.Open(filepath, 'rb') as f:
            image = np.array(
                Image.open(f).convert('RGB')).astype(np.float) / 255.0
        images[idx, :, :, :] = image
        labels[idx] = int(row[row_idx_true_label])
    return images, labels 

def setup_graph(self, input_audio_batch, target_phrase): 
        batch_size = input_audio_batch.shape[0]
        weird = (input_audio_batch.shape[1] - 1) // 320 
        logits_arg2 = np.tile(weird, batch_size)
        dense_arg1 = np.array(np.tile(target_phrase, (batch_size, 1)), dtype=np.int32)
        dense_arg2 = np.array(np.tile(target_phrase.shape[0], batch_size), dtype=np.int32)
        
        pass_in = np.clip(input_audio_batch, -2**15, 2**15-1)
        seq_len = np.tile(weird, batch_size).astype(np.int32)
        
        with tf.variable_scope('', reuse=tf.AUTO_REUSE):
            
            inputs = tf.placeholder(tf.float32, shape=pass_in.shape, name='a')
            len_batch = tf.placeholder(tf.float32, name='b')
            arg2_logits = tf.placeholder(tf.int32, shape=logits_arg2.shape, name='c')
            arg1_dense = tf.placeholder(tf.float32, shape=dense_arg1.shape, name='d')
            arg2_dense = tf.placeholder(tf.int32, shape=dense_arg2.shape, name='e')
            len_seq = tf.placeholder(tf.int32, shape=seq_len.shape, name='f')
            
            logits = get_logits(inputs, arg2_logits)
            target = ctc_label_dense_to_sparse(arg1_dense, arg2_dense, len_batch)
            ctcloss = tf.nn.ctc_loss(labels=tf.cast(target, tf.int32), inputs=logits, sequence_length=len_seq)
            decoded, _ = tf.nn.ctc_greedy_decoder(logits, arg2_logits, merge_repeated=True)
            
            sess = tf.Session()
            saver = tf.train.Saver(tf.global_variables())
            saver.restore(sess, "models/session_dump")
            
        func1 = lambda a, b, c, d, e, f: sess.run(ctcloss, 
            feed_dict={inputs: a, len_batch: b, arg2_logits: c, arg1_dense: d, arg2_dense: e, len_seq: f})
        func2 = lambda a, b, c, d, e, f: sess.run([ctcloss, decoded], 
            feed_dict={inputs: a, len_batch: b, arg2_logits: c, arg1_dense: d, arg2_dense: e, len_seq: f})
        return (func1, func2) 

def getctcloss(self, input_audio_batch, target_phrase, decode=False):
        batch_size = input_audio_batch.shape[0]
        weird = (input_audio_batch.shape[1] - 1) // 320 
        logits_arg2 = np.tile(weird, batch_size)
        dense_arg1 = np.array(np.tile(target_phrase, (batch_size, 1)), dtype=np.int32)
        dense_arg2 = np.array(np.tile(target_phrase.shape[0], batch_size), dtype=np.int32)
        
        pass_in = np.clip(input_audio_batch, -2**15, 2**15-1)
        seq_len = np.tile(weird, batch_size).astype(np.int32)

        if decode:
            return self.funcs[1](pass_in, batch_size, logits_arg2, dense_arg1, dense_arg2, seq_len)
        else:
            return self.funcs[0](pass_in, batch_size, logits_arg2, dense_arg1, dense_arg2, seq_len) 

def _split_into_chunks(signal, chunks_per_second=24):
    # TODO currently broken
    raise NotImplemented("Splitting to chunks is currently broken.")
    window_length_ms = 1/chunks_per_second * 1000
    intervals = np.arange(window_length_ms, signal.shape[0], window_length_ms, dtype=np.int32)
    chunks = np.array_split(signal, intervals, axis=0)
    pad_to = _next_power_of_two(np.max([chunk.shape[0] for chunk in chunks]))
    padded_chunks = np.stack(np.concatenate([chunk, np.zeros((pad_to - chunk.shape[0],))]) for chunk in chunks)
    return padded_chunks 

def udacity_val_gen(data, batch_size):

    """
    Generate training image give image paths and associated steering angles
    """

    images = np.empty([batch_size, configs.LENGTH, configs.IMG_HEIGHT, configs.IMG_WIDTH, 3], dtype=np.int32)
    labels = np.empty([batch_size])

    while True:

        c = 0

        for index in np.random.permutation(data.shape[0]):

            imgs = np.empty([configs.LENGTH, configs.IMG_HEIGHT, configs.IMG_WIDTH, 3], dtype=np.int32)

            if index < configs.LENGTH:
                start = 0
                end = configs.LENGTH
            elif index + configs.LENGTH >= len(data):
                start = len(data) - configs.LENGTH - 1
                end = len(data) - 1
            else:
                start = index
                end = index + configs.LENGTH

            for i in range(start, end):
                center_path = '/home/neil/dataset/steering/test/center/' + str(data['frame_id'].loc[i]) + ".jpg"
                image = load_image(center_path)
                imgs[i - start] = image

            images[c] = imgs
            labels[c] = data['steering_angle'].loc[end]

            c += 1

            if c == batch_size:
                break

        yield images, labels 

def transform_roc(X1, X2, X3):
    n_batch = len(X1)
    xmb = np.zeros((n_batch, 2, n_ctx, 2), dtype=np.int32)
    mmb = np.zeros((n_batch, 2, n_ctx), dtype=np.float32)
    start = encoder['_start_']
    delimiter = encoder['_delimiter_']
    for i, (x1, x2, x3), in enumerate(zip(X1, X2, X3)):
        x12 = [start] + x1[:max_len] + [delimiter] + x2[:max_len] + [clf_token]
        x13 = [start] + x1[:max_len] + [delimiter] + x3[:max_len] + [clf_token]
        l12 = len(x12)
        l13 = len(x13)
        xmb[i, 0, :l12, 0] = x12
        xmb[i, 1, :l13, 0] = x13
        mmb[i, 0, :l12] = 1
        mmb[i, 1, :l13] = 1
    xmb[:, :, :, 1] = np.arange(
        n_vocab + n_special, n_vocab + n_special + n_ctx)
    return xmb, mmb 

def transform_sst(X1):
    n_batch = len(X1)
    xmb = np.zeros((n_batch, 1, n_ctx, 2), dtype=np.int32)
    mmb = np.zeros((n_batch, 1, n_ctx), dtype=np.float32)
    start = encoder['_start_']
    delimiter = encoder['_delimiter_']
    for i, x1, in enumerate(X1):
        x1 = [start] + x1[:max_len] + [clf_token]
        l1 = len(x1)
        xmb[i, 0, :l1, 0] = x1
        mmb[i, 0, :l1] = 1
    xmb[:, :, :, 1] = np.arange(
        n_vocab + n_special, n_vocab + n_special + n_ctx)
    return xmb, mmb 

def draw_on_image(self, img, color=[0, 255, 0], alpha=1.0, thickness=1, copy=copy):
        assert img.dtype in [np.uint8, np.float32, np.int32, np.int64]

        result = np.copy(img) if copy else img
        for i in range(thickness):
            y = [self.y1-i, self.y1-i, self.y2+i, self.y2+i]
            x = [self.x1-i, self.x2+i, self.x2+i, self.x1-i]
            rr, cc = draw.polygon_perimeter(y, x, shape=img.shape)
            if alpha >= 0.99:
                result[rr, cc, 0] = color[0]
                result[rr, cc, 1] = color[1]
                result[rr, cc, 2] = color[2]
            else:
                if result.dtype == np.float32:
                    result[rr, cc, 0] = (1 - alpha) * result[rr, cc, 0] + alpha * color[0]
                    result[rr, cc, 1] = (1 - alpha) * result[rr, cc, 1] + alpha * color[1]
                    result[rr, cc, 2] = (1 - alpha) * result[rr, cc, 2] + alpha * color[2]
                    result = np.clip(result, 0, 255)
                else:
                    result = result.astype(np.float32)
                    result[rr, cc, 0] = (1 - alpha) * result[rr, cc, 0] + alpha * color[0]
                    result[rr, cc, 1] = (1 - alpha) * result[rr, cc, 1] + alpha * color[1]
                    result[rr, cc, 2] = (1 - alpha) * result[rr, cc, 2] + alpha * color[2]
                    result = np.clip(result, 0, 255).astype(np.uint8)

        return result 

def draw_heatmap(img, heatmap, alpha=0.5):
    """Draw a heatmap overlay over an image."""
    assert len(heatmap.shape) == 2 or \
        (len(heatmap.shape) == 3 and heatmap.shape[2] == 1)
    assert img.dtype in [np.uint8, np.int32, np.int64]
    assert heatmap.dtype in [np.float32, np.float64]

    if img.shape[0:2] != heatmap.shape[0:2]:
        heatmap_rs = np.clip(heatmap * 255, 0, 255).astype(np.uint8)
        heatmap_rs = ia.imresize_single_image(
            heatmap_rs[..., np.newaxis],
            img.shape[0:2],
            interpolation="nearest"
        )
        heatmap = np.squeeze(heatmap_rs) / 255.0

    cmap = plt.get_cmap('jet')
    heatmap_cmapped = cmap(heatmap)
    heatmap_cmapped = np.delete(heatmap_cmapped, 3, 2)
    heatmap_cmapped = heatmap_cmapped * 255
    mix = (1-alpha) * img + alpha * heatmap_cmapped
    mix = np.clip(mix, 0, 255).astype(np.uint8)
    return mix 

def test_steps_cumulative_cooccurrence_graph(self, simple_cumco_graph):
        g, o, o_cumsum, co, co_cumsum = simple_cumco_graph
        steps_expected = np.array([0, 1, 2], dtype=np.int32)
        assert_array_equal(g.gp['steps'], steps_expected) 

def evaluate(parameters, f_eval, raw_sentences, parsed_sentences,
             id_to_tag, dictionary_tags,filename,
             useAttend=True):
#{{{
    """
    Evaluate current model using CoNLL script.
    """
    n_tags = len(id_to_tag)
    predictions = []
    count = np.zeros((n_tags, n_tags), dtype=np.int32)

    for raw_sentence, data in zip(raw_sentences, parsed_sentences):
        input = create_input(data, parameters, False,useAttend=useAttend)
        if parameters['crf']:
            y_preds = np.array(f_eval(*input))
        else:
            y_preds = f_eval(*input).argmax(axis=1)
        y_reals = np.array(data['tags']).astype(np.int32)
        assert len(y_preds) == len(y_reals)
        p_tags = [id_to_tag[y_pred] for y_pred in y_preds]
        r_tags = [id_to_tag[y_real] for y_real in y_reals]
        if parameters['tag_scheme'] == 'iobes':
            p_tags = iobes_iob(p_tags)
            r_tags = iobes_iob(r_tags)
        for i, (y_pred, y_real) in enumerate(zip(y_preds, y_reals)):
            new_line = " ".join(raw_sentence[i][:-1] + [r_tags[i], p_tags[i]])
            predictions.append(new_line)
            count[y_real, y_pred] += 1
        predictions.append("")
    #write to file 
    with codecs.open(filename, 'w', 'utf8') as f:
        f.write("\n".join(predictions))
    return get_perf(filename) 
#}}} 

def modelScore(self,tag_ids,scores,s_len):
    #{{{
        """
            ATTENTATION THIS FUNCTION IS SYMBOL PROGRAMMING
            this function is to return the score of our model at a fixed sentence label 
        @param:
            scores:        the scores matrix ,the output of our model
            tag:           a numpy array, which represent one sentence label 
            sent_lens:     a scalar number, the length of sentence.
                because our sentence label will be expand to max sentence length,
                so we will use this to get the original sentence label. 
        @return: 
            a scalar number ,the score;
        """
    #{{{
        n_tags=self.output_dim;
        transitions=self.transitions;
        #score from tags_scores
        real_path_score = scores[T.arange(s_len), tag_ids].sum()

        # Score from transitions
        b_id = theano.shared(value=np.array([n_tags], dtype=np.int32))
        e_id = theano.shared(value=np.array([n_tags + 1], dtype=np.int32))
        padded_tags_ids = T.concatenate([b_id, tag_ids, e_id], axis=0)
        real_path_score += transitions[
                padded_tags_ids[T.arange(s_len + 1)],
                padded_tags_ids[T.arange(s_len + 1) + 1]
            ].sum()
        #to prevent T.exp(real_path_score) to be inf 
        #return real_path_score;
        return real_path_score/s_len;
    #}}}
    #}}} 

def arange(start, stop=None, step=1, dtype='int32'):
    '''Creates a 1-D tensor containing a sequence of integers.

    The function arguments use the same convention as
    Theano's arange: if only one argument is provided,
    it is in fact the "stop" argument.

    The default type of the returned tensor is 'int32' to
    match TensorFlow's default.
    '''
    return T.arange(start, stop=stop, step=step, dtype=dtype) 

def sparse_categorical_crossentropy(output, target, from_logits=False):
    target = T.cast(T.flatten(target), 'int32')
    target = T.extra_ops.to_one_hot(target, nb_class=output.shape[-1])
    target = reshape(target, shape(output))
    return categorical_crossentropy(output, target, from_logits) 

def __init__(self, config):

		entity_total = config.entity
		relation_total = config.relation
		batch_size = config.batch_size
		size = config.hidden_size
		margin = config.margin

		self.pos_h = tf.placeholder(tf.int32, [None])
		self.pos_t = tf.placeholder(tf.int32, [None])
		self.pos_r = tf.placeholder(tf.int32, [None])

		self.neg_h = tf.placeholder(tf.int32, [None])
		self.neg_t = tf.placeholder(tf.int32, [None])
		self.neg_r = tf.placeholder(tf.int32, [None])

		with tf.name_scope("embedding"):
			self.ent_embeddings = tf.get_variable(name = "ent_embedding", shape = [entity_total, size], initializer = tf.contrib.layers.xavier_initializer(uniform = False))
			self.rel_embeddings = tf.get_variable(name = "rel_embedding", shape = [relation_total, size], initializer = tf.contrib.layers.xavier_initializer(uniform = False))
			pos_h_e = tf.nn.embedding_lookup(self.ent_embeddings, self.pos_h)
			pos_t_e = tf.nn.embedding_lookup(self.ent_embeddings, self.pos_t)
			pos_r_e = tf.nn.embedding_lookup(self.rel_embeddings, self.pos_r)
			neg_h_e = tf.nn.embedding_lookup(self.ent_embeddings, self.neg_h)
			neg_t_e = tf.nn.embedding_lookup(self.ent_embeddings, self.neg_t)
			neg_r_e = tf.nn.embedding_lookup(self.rel_embeddings, self.neg_r)

		if config.L1_flag:
			pos = tf.reduce_sum(abs(pos_h_e + pos_r_e - pos_t_e), 1, keep_dims = True)
			neg = tf.reduce_sum(abs(neg_h_e + neg_r_e - neg_t_e), 1, keep_dims = True)
			self.predict = pos
		else:
			pos = tf.reduce_sum((pos_h_e + pos_r_e - pos_t_e) ** 2, 1, keep_dims = True)
			neg = tf.reduce_sum((neg_h_e + neg_r_e - neg_t_e) ** 2, 1, keep_dims = True)
			self.predict = pos

		with tf.name_scope("output"):
			self.loss = tf.reduce_sum(tf.maximum(pos - neg + margin, 0)) 

def _load_pascal_labels(self, index):
    """
    Load image labels.
    """
    filename = os.path.join(self._data_path, 'Annotations', index + '.xml')
    tree = ET.parse(filename)
    objs = tree.findall('object')
    if not self.config['use_diff']:
        # Exclude the samples labeled as difficult
        non_diff_objs = [
            obj for obj in objs if int(obj.find('difficult').text) == 0]
        # if len(non_diff_objs) != len(objs):
        #     print 'Removed {} difficult objects'.format(
        #         len(objs) - len(non_diff_objs))
        objs = non_diff_objs
    num_objs = len(objs)

    gt_classes = np.zeros((num_objs), dtype=np.int32)

    # Load object bounding boxes into a data frame.
    for ix, obj in enumerate(objs):
        cls = self._class_to_ind[obj.find('name').text.lower().strip()]
        gt_classes[ix] = cls

    real_label = np.zeros(self.num_classes).astype(np.float32)
    for label in gt_classes:
        real_label[label] = 1

    return {'labels' : real_label} 

def make_dataset_subset(doc_ids=None, dataset=None, **kwargs):
    ''' Make dataset using subset of the examples in a larger dataset.

    Returns
    -------
    subset : dict
    '''
    doc_ids = np.asarray(doc_ids, dtype=np.int32)

    n_vocabs = dataset['n_vocabs']
    doc_indptr_Dp1 = dataset['doc_indptr_Dp1']
    word_id_U = dataset['word_id_U']
    word_ct_U = dataset['word_ct_U']

    subset_wid_list = list()
    subset_wct_list = list()
    doclen_list = list()
    for d in doc_ids:
        start_d = doc_indptr_Dp1[d]
        stop_d = doc_indptr_Dp1[d+1]
        subset_wid_list.append(word_id_U[start_d:stop_d])
        subset_wct_list.append(word_ct_U[start_d:stop_d])
        doclen_list.append(stop_d - start_d)
    subset = dict(
        word_id_U=np.hstack(subset_wid_list),
        word_ct_U=np.hstack(subset_wct_list),
        doc_indptr_Dp1=np.hstack([0,np.cumsum(doclen_list)]),
        n_docs=len(doc_ids),
        n_vocabs=n_vocabs)
    subset['x_csr_DV'] = scipy.sparse.csr_matrix(
        (subset['word_ct_U'],
         subset['word_id_U'],
         subset['doc_indptr_Dp1']),
        shape=(subset['n_docs'], subset['n_vocabs']))
    if 'y_DC' in dataset:
        subset['y_DC'] = dataset['y_DC'][doc_ids]
        subset['n_labels'] = dataset['n_labels']
    if 'y_rowmask' in dataset:
        subset['y_rowmask'] = dataset['y_rowmask'][doc_ids]
    return subset 

def process_ldac_line_into_preallocated_arrays(line, word_id, word_ct, start):
    """

    Returns
    -------

    Examples
    --------
    >>> word_id = np.zeros(5, dtype=np.int32)
    >>> word_ct = np.zeros(5, dtype=np.float64)
    >>> a = process_ldac_line_into_preallocated_arrays(
    ...    '5 66:6 77:7 88:8',
    ...    word_id, word_ct, 0)
    >>> a
    5
    >>> word_id.tolist()
    [66, 77, 88, 0, 0]
    >>> word_ct.tolist()
    [ 6.,7., 8., 0., 0.]
    """
    line = line.replace(':', ' ')
    data = np.fromstring(line, sep=' ', dtype=np.int32)
    stop = start + (len(data) - 1) // 2
    if stop >= word_id.size:
        raise IndexError("Provided array not large enough")    
    word_id[start:stop] = data[1::2]
    word_ct[start:stop] = data[2::2]
    return data[0] 

def toCArray(X, dtype=np.float64):
    """ Convert input into numpy array of C-contiguous order.

    Ensures returned array is aligned and owns its own data,
    not a view of another array.

    Returns
    -------
    X : ND array

    Examples
    -------
    >>> Q = np.zeros(10, dtype=np.int32, order='F')
    >>> toCArray(Q).flags.c_contiguous
    True
    >>> toCArray(Q).dtype.byteorder
    '='
    """
    X = np.asarray_chkfinite(X, dtype=dtype, order='C')
    if X.dtype.byteorder != '=':
        X = X.newbyteorder('=').copy()
    if not X.flags.owndata or X.flags.aligned:
        X = X.copy()
    assert X.flags.owndata
    assert X.flags.aligned
    return X 

def make_best_task_df(
        df,
        target_query="SPLIT_NAME == 'VALID' and LAP > 50",
        score_colname='Y_ERROR_RATE',
        score_ranking_func=np.argmin,
        default_score=None,
        verbose=False):
    ''' Find best task for each unique job in provided df.

    Returns
    -------
    best_df : dataframe of best tasks for each unique job
    '''
    if default_score is None:
        default_score = fetch_default_score(score_ranking_func.__name__)
    best_task_df_list = list()
    job_paths = np.unique(df['JOB_PATH'].values)
    for job_path in job_paths:
        if job_path is None:
            continue
        job_df = df.query("JOB_PATH == '%s'" % job_path)
        taskids = np.unique(job_df['TASKID'].values)
        best_score_idx = np.zeros_like(taskids, dtype=np.int32)
        best_score = default_score * np.ones_like(taskids, dtype=np.float64)
        for tt, taskidstr in enumerate(taskids):
            task_df = job_df.query(target_query + " and TASKID == '%s'" % taskidstr)
            if task_df.shape[0] < 1:
                continue
            if not np.all(np.isfinite(task_df[score_colname].values)):
                pprint(task_df[score_colname].values)
            best_score_idx[tt] = score_ranking_func(task_df[score_colname].values)
            best_score[tt] = task_df[score_colname].values[best_score_idx[tt]]
        best_task_idx = score_ranking_func(best_score)
        best_task_df = job_df.query("TASKID == '%s'" % taskids[best_task_idx])
        best_task_df_list.append(best_task_df)
        if verbose:
            pprint(job_path)
            pprint("best task: %s" % best_task_idx)
    return pd.concat(best_task_df_list) 

def struct_unpack(bytes):    
    lists = struct.unpack('IIII', bytes)
    shape = tf.TensorShape(lists)    
    return np.array(shape.as_list(), np.int32) 

def convert_shape_to_tensor(shape):
    return tf.convert_to_tensor(shape.as_list(), dtype=tf.int32) 

def test_pack():
    x = tf.placeholder(tf.int32, shape=(4,))
    shapepack = tf.py_func(struct_pack, [x],tf.string)
    with tf.Session() as sess:
       v = sess.run(shapepack, feed_dict={x:[1,2,3,4]})
       print((type(v),v)) 

def test_unpack():
    z = tf.placeholder(tf.string, shape=())
    shapeunpack = tf.py_func(struct_unpack, [z], tf.int32)
    
    v = b'\x01\x00\x00\x00\x02\x00\x00\x00\x03\x00\x00\x00\x04\x00\x00\x00'
    with tf.Session() as sess:
       v = sess.run([shapeunpack], feed_dict={z:v})
       print(v) 

def test_paradigm_citation_dist(self):

        paradigm = sc("E:O:O:.")
        cells = [
            Term("E:U:U:."),
            Term("E:U:A:."),
            Term("E:A:U:."),
            Term("E:A:A:.")
        ]
        headers = [Term("E:O:U:."), Term("E:O:A:."), Term("E:U:O:."), Term("E:A:O:.")]

        usl_collection = [
            usl(Word(Morpheme([cells[0], cells[1]]), Morpheme([cells[3], cells[1], cells[2]]))),
            usl(Word(Morpheme([cells[0], cells[2]]), Morpheme([cells[3]]))),
            usl(Word(Morpheme([headers[0], cells[3]]), Morpheme([headers[3], headers[2]]))),
            usl(Word(Morpheme([cells[1]]), Morpheme([cells[1]])))
        ]

        result = paradigm_usl_distribution(paradigm, usl_collection)
        correct_result = np.zeros((2, 2), dtype=np.int32)

        correct_result[0][0] = 4
        correct_result[0][1] = 5
        correct_result[1][0] = 4
        correct_result[1][1] = 4

        self.assertEqual(len(result), 1, "The paradigm has one table so we should have one distribution table")
        self.assertTrue(np.array_equal(result[0], correct_result)) 

def create_roidb_from_box_list(self, box_list, gt_roidb):
        assert len(box_list) == self.num_images, \
            'Number of boxes must match number of ground-truth images'
        roidb = []
        for i in range(self.num_images):
            boxes = box_list[i]
            num_boxes = boxes.shape[0]
            overlaps = np.zeros((num_boxes, self.num_classes), dtype=np.float32)

            if gt_roidb is not None and gt_roidb[i]['boxes'].size > 0:
                gt_boxes = gt_roidb[i]['boxes']
                gt_classes = gt_roidb[i]['gt_classes']
                gt_overlaps = bbox_overlaps(boxes.astype(np.float),
                                            gt_boxes.astype(np.float))
                argmaxes = gt_overlaps.argmax(axis=1)
                maxes = gt_overlaps.max(axis=1)
                I = np.where(maxes > 0)[0]
                overlaps[I, gt_classes[argmaxes[I]]] = maxes[I]

            overlaps = scipy.sparse.csr_matrix(overlaps)
            roidb.append({
                'boxes': boxes,
                'gt_classes': np.zeros((num_boxes,), dtype=np.int32),
                'gt_overlaps': overlaps,
                'flipped': False,
                'seg_areas': np.zeros((num_boxes,), dtype=np.float32),
            })
        return roidb 

def draw_outputs(img, outputs, class_names):
    boxes, objectness, classes, nums = outputs
    boxes, objectness, classes, nums = boxes[0], objectness[0], classes[0], nums[0]
    wh = np.flip(img.shape[0:2])
    for i in range(nums):
        x1y1 = tuple((np.array(boxes[i][0:2]) * wh).astype(np.int32))
        x2y2 = tuple((np.array(boxes[i][2:4]) * wh).astype(np.int32))
        img = cv2.rectangle(img, x1y1, x2y2, (255, 0, 0), 2)
        img = cv2.putText(img, '{} {:.4f}'.format(
            class_names[int(classes[i])], objectness[i]),
            x1y1, cv2.FONT_HERSHEY_COMPLEX_SMALL, 1, (0, 0, 255), 2)
    return img 

def draw_labels(x, y, class_names):
    img = x.numpy()
    boxes, classes = tf.split(y, (4, 1), axis=-1)
    classes = classes[..., 0]
    wh = np.flip(img.shape[0:2])
    for i in range(len(boxes)):
        x1y1 = tuple((np.array(boxes[i][0:2]) * wh).astype(np.int32))
        x2y2 = tuple((np.array(boxes[i][2:4]) * wh).astype(np.int32))
        img = cv2.rectangle(img, x1y1, x2y2, (255, 0, 0), 2)
        img = cv2.putText(img, class_names[classes[i]],
                          x1y1, cv2.FONT_HERSHEY_COMPLEX_SMALL,
                          1, (0, 0, 255), 2)
    return img 

def create_cifar10(tfrecord_dir, cifar10_dir):
    print('Loading CIFAR-10 from "%s"' % cifar10_dir)
    import pickle
    images = []
    labels = []
    for batch in range(1, 6):
        with open(os.path.join(cifar10_dir, 'data_batch_%d' % batch), 'rb') as file:
            data = pickle.load(file, encoding='latin1')
        images.append(data['data'].reshape(-1, 3, 32, 32))
        labels.append(data['labels'])
    images = np.concatenate(images)
    labels = np.concatenate(labels)
    assert images.shape == (50000, 3, 32, 32) and images.dtype == np.uint8
    assert labels.shape == (50000,) and labels.dtype == np.int32
    assert np.min(images) == 0 and np.max(images) == 255
    assert np.min(labels) == 0 and np.max(labels) == 9
    onehot = np.zeros((labels.size, np.max(labels) + 1), dtype=np.float32)
    onehot[np.arange(labels.size), labels] = 1.0

    with TFRecordExporter(tfrecord_dir, images.shape[0]) as tfr:
        order = tfr.choose_shuffled_order()
        for idx in range(order.size):
            tfr.add_image(images[order[idx]])
        tfr.add_labels(onehot[order])

#---------------------------------------------------------------------------- 

def get_random_labels_tf(self, minibatch_size): # => labels
        if self.label_size > 0:
            return tf.gather(self._tf_labels_var, tf.random_uniform([minibatch_size], 0, self._np_labels.shape[0], dtype=tf.int32))
        else:
            return tf.zeros([minibatch_size, 0], self.label_dtype)

    # Get random labels as NumPy array. 

def get_minibatch_tf(self): # => images, labels
        with tf.name_scope('SyntheticDataset'):
            shrink = tf.cast(2.0 ** tf.cast(self._tf_lod_var, tf.float32), tf.int32)
            shape = [self.shape[0], self.shape[1] // shrink, self.shape[2] // shrink]
            images = self._generate_images(self._tf_minibatch_var, self._tf_lod_var, shape)
            labels = self._generate_labels(self._tf_minibatch_var)
            return images, labels 

def __init__(self, n_samples=None, list_of_indecies=None):
        if not n_samples is None:
            self.GetFresh_list = numpy.arange(n_samples, dtype=numpy.int32)
            self.get_fresh_count = n_samples
        elif not list_of_indecies is None:
            self.GetFresh_list = numpy.array(copy.deepcopy(list_of_indecies))
            self.get_fresh_count = len(self.GetFresh_list)
        else:
            raise Exception('Invalid Input')

        numpy.random.shuffle(self.GetFresh_list)
        self.get_fresh_pos = 0 

def get(self, n):
        """
        will return a list of n random numbers in self.GetFresh_list
        - Samuel Buteau, October 2018
        """
        if n >= self.get_fresh_count:
            return numpy.concatenate((self.get(int(n/2)),self.get(n- int(n/2))))


        reshuffle_flag = False

        n_immediate_fulfill = min(n, self.get_fresh_count - self.get_fresh_pos)
        batch_of_indecies = numpy.empty([n], dtype=numpy.int32)
        for i in range(0, n_immediate_fulfill):
            batch_of_indecies[i] = self.GetFresh_list[i + self.get_fresh_pos]

        self.get_fresh_pos += n_immediate_fulfill
        if self.get_fresh_pos >= self.get_fresh_count:
            self.get_fresh_pos -= self.get_fresh_count
            reshuffle_flag = True

            # Now, the orders that needed to be satisfied are satisfied.
        n_delayed_fulfill = max(0, n - n_immediate_fulfill)
        if reshuffle_flag:
            numpy.random.shuffle(self.GetFresh_list)

        if n_delayed_fulfill > 0:
            for i in range(0, n_delayed_fulfill):
                batch_of_indecies[i + n_immediate_fulfill] = self.GetFresh_list[i]
            self.get_fresh_pos = n_delayed_fulfill

        return batch_of_indecies 

def parser(transcript, ref_path):
    transcript = tf.decode_raw(transcript, tf.int32) # decode string to int
    transcript = tf.pad(transcript, ([0, Hp.num_charac], ))[:Hp.num_charac] # pad charac[0] to fill in the whole max length Tx

    def load_audio(audio_path):
        sub_paths = audio_path.strip().split('/')
        main_path = '/'.join(sub_paths[:-2])
        fname = os.path.basename(audio_path)
        
        mel_path = main_path + "/mels/{}".format(sub_paths[-1].replace('wav', 'npy'))  # mel spectrogrom
        mag_path = main_path + "/mags/{}".format(sub_paths[-1].replace('wav', 'npy'))  # wavform

        mel = np.load(mel_path)
        mag = np.load(mag_path)
        ref_len = mel.shape[0] if mel.shape[1]==Hp.num_mels*Hp.reduction_factor \
                                else mel.shape[0]*mel.shape[1]//Hp.num_mels*Hp.reduction_factor
        ref_len = np.array(ref_len, dtype = np.int32)
        
        return mel, mag, ref_len 

    spectrogrom, wavform, ref_len = tf.py_func(load_audio, [ref_path], [tf.float32, tf.float32, tf.int32])
    
    transcript = tf.reshape(transcript, [Hp.num_charac,])
    
    spectrogrom = tf.reshape(spectrogrom, [-1, Hp.num_mels*Hp.reduction_factor])
    
    #ref_len = tf.cast(spectrogrom.get_shape()[0], tf.int32)
    wavform = tf.reshape(wavform, [-1, Hp.num_fft//2+1])

    inputs = list((transcript, spectrogrom, tf.reshape(ref_len, [1]), tf.reshape(tf.constant(0), [1]), spectrogrom, wavform))
    #inputs = {'transcript':transcript, 'reference':spectrogrom, 'ref_len':tf.reshape(ref_len,[1]),
    #'speaker':tf.reshape(tf.constant(0), [1]), 'decoder':spectrogrom, 'labels':wavform} 
    
    return inputs