Python numpy.float32() Examples

def sgd(self, cost, params,constraints={}, lr=0.01):
#{{{
        """
        Stochatic gradient descent.
        """
        updates = []
        
        lr = theano.shared(np.float32(lr).astype(floatX))
        gradients = self.get_gradients(cost, params)
        
        for p, g in zip(params, gradients):
            v=-lr*g;
            new_p=p+v;
            # apply constraints
            if p in constraints:
                c=constraints[p];
                new_p=c(new_p);
            updates.append((p, new_p))

        return updates
#}}} 

def convert_answer(answer, num_words, format_spec, maxlen):
    """
    Convert an answer into an appropriate answer matrix given
    a ModelOutputFormat.

    num_words should be after processing with get_effective_answer_words,
    so that the last word is the "stop" word
    """
    assert format_spec in model.ModelOutputFormat
    if format_spec == model.ModelOutputFormat.subset:
        ans_mat = np.zeros((1,num_words), np.float32)
        for word in answer:
            ans_mat[0, word] = 1.0
    elif format_spec == model.ModelOutputFormat.category:
        ans_mat = np.zeros((1,num_words), np.float32)
        ans_mat[0,answer[0]] = 1.0
    elif format_spec == model.ModelOutputFormat.sequence:
        ans_mat = np.zeros((maxlen+1,num_words), np.float32)
        for i,word in enumerate(answer+[num_words-1]*(maxlen+1-len(answer))):
            ans_mat[i, word] = 1.0
    return ans_mat 

def _maybe_cast_to_float64(da):
    """Cast DataArrays to np.float64 if they are of type np.float32.

    Parameters
    ----------
    da : xr.DataArray
        Input DataArray

    Returns
    -------
    DataArray

    """
    if da.dtype == np.float32:
        logging.warning('Datapoints were stored using the np.float32 datatype.'
                        'For accurate reduction operations using bottleneck, '
                        'datapoints are being cast to the np.float64 datatype.'
                        ' For more information see: https://github.com/pydata/'
                        'xarray/issues/1346')
        return da.astype(np.float64)
    else:
        return da 

def add_image(self, img):
        if self.print_progress and self.cur_images % self.progress_interval == 0:
            print('%d / %d\r' % (self.cur_images, self.expected_images), end='', flush=True)
            sys.stdout.flush()
        if self.shape is None:
            self.shape = img.shape
            self.resolution_log2 = int(np.log2(self.shape[1]))
            assert self.shape[0] in [1, 3]
            assert self.shape[1] == self.shape[2]
            assert self.shape[1] == 2**self.resolution_log2
            tfr_opt = tf.python_io.TFRecordOptions(tf.python_io.TFRecordCompressionType.NONE)
            for lod in range(self.resolution_log2 - 1):
                tfr_file = self.tfr_prefix + '-r%02d.tfrecords' % (self.resolution_log2 - lod)
                self.tfr_writers.append(tf.python_io.TFRecordWriter(tfr_file, tfr_opt))
        assert img.shape == self.shape
        for lod, tfr_writer in enumerate(self.tfr_writers):
            if lod:
                img = img.astype(np.float32)
                img = (img[:, 0::2, 0::2] + img[:, 0::2, 1::2] + img[:, 1::2, 0::2] + img[:, 1::2, 1::2]) * 0.25
            quant = np.rint(img).clip(0, 255).astype(np.uint8)
            ex = tf.train.Example(features=tf.train.Features(feature={
                'shape': tf.train.Feature(int64_list=tf.train.Int64List(value=quant.shape)),
                'data': tf.train.Feature(bytes_list=tf.train.BytesList(value=[quant.tostring()]))}))
            tfr_writer.write(ex.SerializeToString())
        self.cur_images += 1 

def create_mnist(tfrecord_dir, mnist_dir):
    print('Loading MNIST from "%s"' % mnist_dir)
    import gzip
    with gzip.open(os.path.join(mnist_dir, 'train-images-idx3-ubyte.gz'), 'rb') as file:
        images = np.frombuffer(file.read(), np.uint8, offset=16)
    with gzip.open(os.path.join(mnist_dir, 'train-labels-idx1-ubyte.gz'), 'rb') as file:
        labels = np.frombuffer(file.read(), np.uint8, offset=8)
    images = images.reshape(-1, 1, 28, 28)
    images = np.pad(images, [(0,0), (0,0), (2,2), (2,2)], 'constant', constant_values=0)
    assert images.shape == (60000, 1, 32, 32) and images.dtype == np.uint8
    assert labels.shape == (60000,) and labels.dtype == np.uint8
    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 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 draw_on_image(self, img, color=[0, 255, 0], alpha=1.0, copy=True):
        if copy:
            img = np.copy(img)

        orig_dtype = img.dtype
        if alpha != 1.0 and img.dtype != np.float32:
            img = img.astype(np.float32, copy=False)

        for rectimg in self:
            if rectimg is not None:
                rectimg.draw_on_image(img, color=color, alpha=alpha, copy=False)

        if orig_dtype != img.dtype:
            img = img.astype(orig_dtype, copy=False)

        return img 

def draw_on_image(self, img, color=[0, 255, 0], alpha=1.0, copy=True, from_img=None):
        if copy:
            img = np.copy(img)

        orig_dtype = img.dtype
        if alpha != 1.0 and img.dtype != np.float32:
            img = img.astype(np.float32, copy=False)

        for rect in self:
            if from_img is not None:
                rect.resize(from_img, img).draw_on_image(img, color=color, alpha=alpha, copy=False)
            else:
                rect.draw_on_image(img, color=color, alpha=alpha, copy=False)

        if orig_dtype != img.dtype:
            img = img.astype(orig_dtype, copy=False)

        return img 

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 generate_fake_images(run_id, snapshot=None, grid_size=[1,1], num_pngs=1, image_shrink=1, png_prefix=None, random_seed=1000, minibatch_size=8):
    network_pkl = misc.locate_network_pkl(run_id, snapshot)
    if png_prefix is None:
        png_prefix = misc.get_id_string_for_network_pkl(network_pkl) + '-'
    random_state = np.random.RandomState(random_seed)

    print('Loading network from "%s"...' % network_pkl)
    G, D, Gs = misc.load_network_pkl(run_id, snapshot)

    result_subdir = misc.create_result_subdir(config.result_dir, config.desc)
    for png_idx in range(num_pngs):
        print('Generating png %d / %d...' % (png_idx, num_pngs))
        latents = misc.random_latents(np.prod(grid_size), Gs, random_state=random_state)
        labels = np.zeros([latents.shape[0], 0], np.float32)
        images = Gs.run(latents, labels, minibatch_size=minibatch_size, num_gpus=config.num_gpus, out_mul=127.5, out_add=127.5, out_shrink=image_shrink, out_dtype=np.uint8)
        misc.save_image_grid(images, os.path.join(result_subdir, '%s%06d.png' % (png_prefix, png_idx)), [0,255], grid_size)
    open(os.path.join(result_subdir, '_done.txt'), 'wt').close()

#----------------------------------------------------------------------------
# Generate MP4 video of random interpolations using a previously trained network.
# To run, uncomment the appropriate line in config.py and launch train.py. 

def minibatch_stddev_layer(x, group_size=4):
    with tf.variable_scope('MinibatchStddev'):
        group_size = tf.minimum(group_size, tf.shape(x)[0])     # Minibatch must be divisible by (or smaller than) group_size.
        s = x.shape                                             # [NCHW]  Input shape.
        y = tf.reshape(x, [group_size, -1, s[1], s[2], s[3]])   # [GMCHW] Split minibatch into M groups of size G.
        y = tf.cast(y, tf.float32)                              # [GMCHW] Cast to FP32.
        y -= tf.reduce_mean(y, axis=0, keep_dims=True)           # [GMCHW] Subtract mean over group.
        y = tf.reduce_mean(tf.square(y), axis=0)                # [MCHW]  Calc variance over group.
        y = tf.sqrt(y + 1e-8)                                   # [MCHW]  Calc stddev over group.
        y = tf.reduce_mean(y, axis=[1,2,3], keep_dims=True)      # [M111]  Take average over fmaps and pixels.
        y = tf.cast(y, x.dtype)                                 # [M111]  Cast back to original data type.
        y = tf.tile(y, [group_size, 1, s[2], s[3]])             # [N1HW]  Replicate over group and pixels.
        return tf.concat([x, y], axis=1)                        # [NCHW]  Append as new fmap.

#----------------------------------------------------------------------------
# Generator network used in the paper. 

def set_values(name, param, pretrained):
#{{{
    """
    Initialize a network parameter with pretrained values.
    We check that sizes are compatible.
    """
    param_value = param.get_value()
    if pretrained.size != param_value.size:
        raise Exception(
            "Size mismatch for parameter %s. Expected %i, found %i."
            % (name, param_value.size, pretrained.size)
        )
    param.set_value(np.reshape(
        pretrained, param_value.shape
    ).astype(np.float32))
#}}} 

def sgdmomentum(self, cost, params,constraints={}, lr=0.01,consider_constant=None, momentum=0.):
        """
        Stochatic gradient descent with momentum. Momentum has to be in [0, 1)
        """
        # Check that the momentum is a correct value
        assert 0 <= momentum < 1

        lr = theano.shared(np.float32(lr).astype(floatX))
        momentum = theano.shared(np.float32(momentum).astype(floatX))

        gradients = self.get_gradients(cost, params)
        velocities = [theano.shared(np.zeros_like(param.get_value(borrow=True)).astype(floatX)) for param in params]

        updates = []
        for param, gradient, velocity in zip(params, gradients, velocities):
            new_velocity = momentum * velocity - lr * gradient
            updates.append((velocity, new_velocity))
            new_p=param+new_velocity;
            # apply constraints
            if param in constraints:
                c=constraints[param];
                new_p=c(new_p);
            updates.append((param, new_p))
        return updates 

def adagrad(self, cost, params, lr=1.0, epsilon=1e-6,consider_constant=None):
        """
        Adagrad. Based on http://www.ark.cs.cmu.edu/cdyer/adagrad.pdf
        """
        lr = theano.shared(np.float32(lr).astype(floatX))
        epsilon = theano.shared(np.float32(epsilon).astype(floatX))

        gradients = self.get_gradients(cost, params,consider_constant)
        gsums = [theano.shared(np.zeros_like(param.get_value(borrow=True)).astype(floatX)) for param in params]

        updates = []
        for param, gradient, gsum in zip(params, gradients, gsums):
            new_gsum = gsum + gradient ** 2.
            updates.append((gsum, new_gsum))
            updates.append((param, param - lr * gradient / (T.sqrt(gsum + epsilon))))
        return updates 

def adadelta(self, cost, params, rho=0.95, epsilon=1e-6,consider_constant=None):
        """
        Adadelta. Based on:
        http://www.matthewzeiler.com/pubs/googleTR2012/googleTR2012.pdf
        """
        rho = theano.shared(np.float32(rho).astype(floatX))
        epsilon = theano.shared(np.float32(epsilon).astype(floatX))

        gradients = self.get_gradients(cost, params,consider_constant)
        accu_gradients = [theano.shared(np.zeros_like(param.get_value(borrow=True)).astype(floatX)) for param in params]
        accu_deltas = [theano.shared(np.zeros_like(param.get_value(borrow=True)).astype(floatX)) for param in params]

        updates = []
        for param, gradient, accu_gradient, accu_delta in zip(params, gradients, accu_gradients, accu_deltas):
            new_accu_gradient = rho * accu_gradient + (1. - rho) * gradient ** 2.
            delta_x = - T.sqrt((accu_delta + epsilon) / (new_accu_gradient + epsilon)) * gradient
            new_accu_delta = rho * accu_delta + (1. - rho) * delta_x ** 2.
            updates.append((accu_gradient, new_accu_gradient))
            updates.append((accu_delta, new_accu_delta))
            updates.append((param, param + delta_x))
        return updates 

def rmsprop(self, cost, params, lr=0.001, rho=0.9, eps=1e-6,consider_constant=None):
        """
        RMSProp.
        """
        lr = theano.shared(np.float32(lr).astype(floatX))

        gradients = self.get_gradients(cost, params,consider_constant)
        accumulators = [theano.shared(np.zeros_like(p.get_value()).astype(np.float32)) for p in params]

        updates = []

        for param, gradient, accumulator in zip(params, gradients, accumulators):
            new_accumulator = rho * accumulator + (1 - rho) * gradient ** 2
            updates.append((accumulator, new_accumulator))

            new_param = param - lr * gradient / T.sqrt(new_accumulator + eps)
            updates.append((param, new_param))

        return updates 

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_autosummary_var(name, value_expr):
    assert not _autosummary_finalized
    v = tf.cast(value_expr, tf.float32)
    if v.shape.ndims is 0:
        v = [v, np.float32(1.0)]
    elif v.shape.ndims is 1:
        v = [tf.reduce_sum(v), tf.cast(tf.shape(v)[0], tf.float32)]
    else:
        v = [tf.reduce_sum(v), tf.reduce_prod(tf.cast(tf.shape(v), tf.float32))]
    v = tf.cond(tf.is_finite(v[0]), lambda: tf.stack(v), lambda: tf.zeros(2))
    with tf.control_dependencies(None):
        var = tf.Variable(tf.zeros(2)) # [numerator, denominator]
    update_op = tf.cond(tf.is_variable_initialized(var), lambda: tf.assign_add(var, v), lambda: tf.assign(var, v))
    if name in _autosummary_vars:
        _autosummary_vars[name].append(var)
    else:
        _autosummary_vars[name] = [var]
    return update_op

#----------------------------------------------------------------------------
# Call filewriter.add_summary() with all summaries in the default graph,
# automatically finalizing and merging them on the first call. 

def autosummary(name, value):
    id = name.replace('/', '_')
    if is_tf_expression(value):
        with tf.name_scope('summary_' + id), tf.device(value.device):
            update_op = _create_autosummary_var(name, value)
            with tf.control_dependencies([update_op]):
                return tf.identity(value)
    else: # python scalar or numpy array
        if name not in _autosummary_immediate:
            with absolute_name_scope('Autosummary/' + id), tf.device(None), tf.control_dependencies(None):
                update_value = tf.placeholder(tf.float32)
                update_op = _create_autosummary_var(name, update_value)
                _autosummary_immediate[name] = update_op, update_value
        update_op, update_value = _autosummary_immediate[name]
        run(update_op, {update_value: np.float32(value)})
        return value

# Create the necessary ops to include autosummaries in TensorBoard report.
# Note: This should be done only once per graph. 

def _get_rois_blob(im_rois, im_scale_factors):
    """Converts RoIs into network inputs.
    Arguments:
        im_rois (ndarray): R x 4 matrix of RoIs in original image coordinates
        im_scale_factors (list): scale factors as returned by _get_image_blob
    Returns:
        blob (ndarray): R x 5 matrix of RoIs in the image pyramid
    """
    rois_blob_real = []

    for i in range(len(im_scale_factors)):
        rois, levels = _project_im_rois(im_rois, np.array([im_scale_factors[i]]))
        rois_blob = np.hstack((levels, rois))
        rois_blob_real.append(rois_blob.astype(np.float32, copy=False))

    return rois_blob_real 

def _get_rois_blob(im_rois, im_scale_factors):
    """Converts RoIs into network inputs.
    Arguments:
        im_rois (ndarray): R x 4 matrix of RoIs in original image coordinates
        im_scale_factors (list): scale factors as returned by _get_image_blob
    Returns:
        blob (ndarray): R x 5 matrix of RoIs in the image pyramid
    """
    rois_blob_real = []

    for i in range(len(im_scale_factors)):
        rois, levels = _project_im_rois(im_rois, np.array([im_scale_factors[i]]))
        rois_blob = np.hstack((levels, rois))
        rois_blob_real.append(rois_blob.astype(np.float32, copy=False))

    return rois_blob_real 

def read_float_vec(fd, direct_access=False):
    """
        Read float vector(for class Vector in kaldi setup)
        see matrix/kaldi-vector.cc
    """
    if direct_access:
        expect_binary(fd)
    vec_type = read_token(fd)
    print_info(f'\tType of the common vector: {vec_type}')
    if vec_type not in ["FV", "DV"]:
        raise RuntimeError(f"Unknown matrix type in kaldi: {vec_type}")
    float_size = 4 if vec_type == 'FV' else 8
    float_type = np.float32 if vec_type == 'FV' else np.float64
    dim = read_int32(fd)
    print_info(f'\tDim of the common vector: {dim}')
    vec_data = fd.read(float_size * dim)
    return np.fromstring(vec_data, dtype=float_type) 

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 read_common_mat(fd):
    """ 
        Read common matrix(for class Matrix in kaldi setup)
        see matrix/kaldi-matrix.cc::
            void Matrix<Real>::Read(std::istream & is, bool binary, bool add)
        Return a numpy ndarray object
    """
    mat_type = read_token(fd)
    print_info(f'\tType of the common matrix: {mat_type}')
    if mat_type not in ["FM", "DM"]:
        raise RuntimeError(f"Unknown matrix type in kaldi: {mat_type}")
    float_size = 4 if mat_type == 'FM' else 8
    float_type = np.float32 if mat_type == 'FM' else np.float64
    num_rows = read_int32(fd)
    num_cols = read_int32(fd)
    print_info(f'\tSize of the common matrix: {num_rows} x {num_cols}')
    mat_data = fd.read(float_size * num_cols * num_rows)
    mat = np.fromstring(mat_data, dtype=float_type)
    return mat.reshape(num_rows, num_cols) 

def wrap_variable(self, var):
        """wrap layer.w into variables"""
        val = self.lay.w.get(var, None)
        if val is None:
            shape = self.lay.wshape[var]
            args = [0., 1e-2, shape]
            if 'moving_mean' in var:
                val = np.zeros(shape)
            elif 'moving_variance' in var:
                val = np.ones(shape)
            else:
                val = np.random.normal(*args)
            self.lay.w[var] = val.astype(np.float32)
            self.act = 'Init '
        if not self.var: return

        val = self.lay.w[var]
        self.lay.w[var] = tf.constant_initializer(val)
        if var in self._SLIM: return
        with tf.variable_scope(self.scope):
            self.lay.w[var] = tf.get_variable(var,
                shape = self.lay.wshape[var],
                dtype = tf.float32,
                initializer = self.lay.w[var]) 

def generate_moving_mnist(self, num_digits=2):
    '''
    Get random trajectories for the digits and generate a video.
    '''
    data = np.zeros((self.n_frames_total, self.image_size_, self.image_size_), dtype=np.float32)
    for n in range(num_digits):
      # Trajectory
      start_y, start_x = self.get_random_trajectory(self.n_frames_total)
      ind = random.randint(0, self.mnist.shape[0] - 1)
      digit_image = self.mnist[ind]
      for i in range(self.n_frames_total):
        top    = start_y[i]
        left   = start_x[i]
        bottom = top + self.digit_size_
        right  = left + self.digit_size_
        # Draw digit
        data[i, top:bottom, left:right] = np.maximum(data[i, top:bottom, left:right], digit_image)

    data = data[..., np.newaxis]
    return data 

def __init__(self, input_wave_file, output_wave_file, target_phrase):
        self.pop_size = 100
        self.elite_size = 10
        self.mutation_p = 0.005
        self.noise_stdev = 40
        self.noise_threshold = 1
        self.mu = 0.9
        self.alpha = 0.001
        self.max_iters = 3000
        self.num_points_estimate = 100
        self.delta_for_gradient = 100
        self.delta_for_perturbation = 1e3
        self.input_audio = load_wav(input_wave_file).astype(np.float32)
        self.pop = np.expand_dims(self.input_audio, axis=0)
        self.pop = np.tile(self.pop, (self.pop_size, 1))
        self.output_wave_file = output_wave_file
        self.target_phrase = target_phrase
        self.funcs = self.setup_graph(self.pop, np.array([toks.index(x) for x in target_phrase])) 

def write_common_mat(fd, mat):
    """
        Write a common matrix
    """
    if mat.dtype not in [np.float32, np.float64]:
        raise RuntimeError(f"Unsupported numpy dtype: {mat.dtype}")
    mat_type = 'FM' if mat.dtype == np.float32 else 'DM'
    write_token(fd, mat_type)
    num_rows, num_cols = mat.shape
    write_int32(fd, num_rows)
    write_int32(fd, num_cols)
    fd.write(mat.tobytes()) 

def process_reals(x, lod, mirror_augment, drange_data, drange_net):
    with tf.name_scope('ProcessReals'):
        with tf.name_scope('DynamicRange'):
            x = tf.cast(x, tf.float32)
            x = misc.adjust_dynamic_range(x, drange_data, drange_net)
        if mirror_augment:
            with tf.name_scope('MirrorAugment'):
                s = tf.shape(x)
                mask = tf.random_uniform([s[0], 1, 1, 1], 0.0, 1.0)
                mask = tf.tile(mask, [1, s[1], s[2], s[3]])
                x = tf.where(mask < 0.5, x, tf.reverse(x, axis=[3]))
        with tf.name_scope('FadeLOD'): # Smooth crossfade between consecutive levels-of-detail.
            s = tf.shape(x)
            y = tf.reshape(x, [-1, s[1], s[2]//2, 2, s[3]//2, 2])
            y = tf.reduce_mean(y, axis=[3, 5], keep_dims=True)
            y = tf.tile(y, [1, 1, 1, 2, 1, 2])
            y = tf.reshape(y, [-1, s[1], s[2], s[3]])
            x = tfutil.lerp(x, y, lod - tf.floor(lod))
        with tf.name_scope('UpscaleLOD'): # Upscale to match the expected input/output size of the networks.
            s = tf.shape(x)
            factor = tf.cast(2 ** tf.floor(lod), tf.int32)
            x = tf.reshape(x, [-1, s[1], s[2], 1, s[3], 1])
            x = tf.tile(x, [1, 1, 1, factor, 1, factor])
            x = tf.reshape(x, [-1, s[1], s[2] * factor, s[3] * factor])
        return x

#----------------------------------------------------------------------------
# Just-in-time processing of masks before feeding them to the networks.