Python numpy.hstack() Examples

def add_intercept(self, X):
        """Add 1's to data as last features."""
        # Data shape
        N, D = X.shape

        # Check if there's not already an intercept column
        if np.any(np.sum(X, axis=0) == N):

            # Report
            print('Intercept is not the last feature. Swapping..')

            # Find which column contains the intercept
            intercept_index = np.argwhere(np.sum(X, axis=0) == N)

            # Swap intercept to last
            X = X[:, np.setdiff1d(np.arange(D), intercept_index)]

        # Add intercept as last column
        X = np.hstack((X, np.ones((N, 1))))

        # Append column of 1's to data, and increment dimensionality
        return X, D+1 

def regression_data():
    f = open(path + "regression_data1.txt")
    data = np.loadtxt(f, delimiter=",")
    x1 = np.insert(data[:, 0].reshape(len(data), 1), 0, np.ones(len(data)), axis=1)
    y1 = data[:, 1]
    f = open(path + "regression_data2.txt")
    data = np.loadtxt(f, delimiter=",")
    x2 = np.insert(data[:, 0].reshape(len(data), 1), 0, np.ones(len(data)), axis=1)
    y2 = data[:, 1]
    x1 = np.vstack((x1, x2))
    y1 = np.hstack((y1, y2))

    f = open(path + "regression_data_test1.txt")
    data = np.loadtxt(f, delimiter=",")
    x1_test = np.insert(data[:, 0].reshape(len(data), 1), 0, np.ones(len(data)), axis=1)
    y1_test = data[:, 1]
    f = open(path + "regression_data_test2.txt")
    data = np.loadtxt(f, delimiter=",")
    x2_test = np.insert(data[:, 0].reshape(len(data), 1), 0, np.ones(len(data)), axis=1)
    y2_test = data[:, 1]
    x1_test = np.vstack((x1_test, x2_test))
    y1_test = np.hstack((y1_test, y2_test))
    return x1, y1, x1_test, y1_test 

def test_all_columns_accounted_for(self):
    # Note: deliberately setting to small number so not always
    # all possibilities appear (matched, unmatched, ignored)
    num_matches = 10
    match_results = tf.random_uniform(
        [num_matches], minval=-2, maxval=5, dtype=tf.int32)
    match = matcher.Match(match_results)
    matched_column_indices = match.matched_column_indices()
    unmatched_column_indices = match.unmatched_column_indices()
    ignored_column_indices = match.ignored_column_indices()
    with self.test_session() as sess:
      matched, unmatched, ignored = sess.run([
          matched_column_indices, unmatched_column_indices,
          ignored_column_indices
      ])
      all_indices = np.hstack((matched, unmatched, ignored))
      all_indices_sorted = np.sort(all_indices)
      self.assertAllEqual(all_indices_sorted,
                          np.arange(num_matches, dtype=np.int32)) 

def scale(boxlist, y_scale, x_scale):
  """Scale box coordinates in x and y dimensions.

  Args:
    boxlist: BoxList holding N boxes
    y_scale: float
    x_scale: float

  Returns:
    boxlist: BoxList holding N boxes
  """
  y_min, x_min, y_max, x_max = np.array_split(boxlist.get(), 4, axis=1)
  y_min = y_scale * y_min
  y_max = y_scale * y_max
  x_min = x_scale * x_min
  x_max = x_scale * x_max
  scaled_boxlist = np_box_list.BoxList(np.hstack([y_min, x_min, y_max, x_max]))

  fields = boxlist.get_extra_fields()
  for field in fields:
    extra_field_data = boxlist.get_field(field)
    scaled_boxlist.add_field(field, extra_field_data)

  return scaled_boxlist 

def ex3(replication=2):
    f = open(path + "ex3.txt")
    train_data = np.loadtxt(f, delimiter=",")
    f = open(path + "ex3_test.txt")
    test_data = np.loadtxt(f, delimiter=",")

    x_train = np.insert(train_data[:, (0, 1)], 0, np.ones(len(train_data)), axis=1)
    y_train = train_data[:, 2]
    x_test = np.insert(test_data[:, (0, 1)], 0, np.ones(len(test_data)), axis=1)
    y_test = test_data[:, 2]

    for i in range(replication - 1):
        x_train = np.vstack((x_train, np.insert(train_data[:, (0, 1)], 0, np.ones(len(train_data)), axis=1)))
        y_train = np.hstack((y_train, train_data[:, 2]))

        x_test = np.vstack((x_test, np.insert(test_data[:, (0, 1)], 0, np.ones(len(test_data)), axis=1)))
        y_test = np.hstack((y_test, test_data[:, 2]))

    return x_train, y_train, x_test, y_test 

def iris(replication=2):
    f = open(path + "iris.txt")
    data = np.loadtxt(f, delimiter=",", dtype=np.string0)
    x_train = np.array(data[:, range(0, 4)], dtype=np.float)
    y_train = data[:, 4]

    for j in range(replication - 1):
        x_train = np.vstack([x_train, data[:, range(0, 4)]])
        y_train = np.hstack([y_train, data[:, 4]])
    x_train = np.array(x_train, dtype=np.float)

    f = open(path + "iris_test.txt")
    data = np.loadtxt(f, delimiter=",", dtype=np.string0)
    x_test = np.array(data[:, range(0, 4)], dtype=np.float)
    y_test = data[:, 4]

    for j in range(replication - 1):
        x_test = np.vstack([x_test, data[:, range(0, 4)]])
        y_test = np.hstack([y_test, data[:, 4]])
    x_test = np.array(x_test, dtype=np.float)

    return x_train, y_train, x_test, y_test 

def test_one_hot():
    """Check if one_hot returns correct label matrices."""
    # Generate label vector
    y = np.hstack((np.ones((10,))*0,
                   np.ones((10,))*1,
                   np.ones((10,))*2))

    # Map to matrix
    Y, labels = one_hot(y)

    # Check for only 0's and 1's
    assert len(np.setdiff1d(np.unique(Y), [0, 1])) == 0

    # Check for correct labels
    assert np.all(labels == np.unique(y))

    # Check correct shape of matrix
    assert Y.shape[0] == y.shape[0]
    assert Y.shape[1] == len(labels) 

def breastcancer_disc(replication=2):
    f = open(path + "breast_cancer_wisconsin_disc.txt")
    data = np.loadtxt(f, delimiter=",")
    x_train = data[:, range(1, 10)]
    y_train = data[:, 10]
    for j in range(replication - 1):
        x_train = np.vstack([x_train, data[:, range(1, 10)]])
        y_train = np.hstack([y_train, data[:, 10]])

    f = open(path + "breast_cancer_wisconsin_disc_test.txt")
    data = np.loadtxt(f, delimiter=",")
    x_test = data[:, range(1, 10)]
    y_test = data[:, 10]
    for j in range(replication - 1):
        x_test = np.vstack([x_test, data[:, range(1, 10)]])
        y_test = np.hstack([y_test, data[:, 10]])

    return x_train, y_train, x_test, y_test 

def breastcancer_cont(replication=2):
    f = open(path + "breast_cancer_wisconsin_cont.txt", "r")
    data = np.loadtxt(f, delimiter=",", dtype=np.string0)
    x_train = np.array(data[:, range(0, 9)])
    y_train = np.array(data[:, 9])
    for j in range(replication - 1):
        x_train = np.vstack([x_train, data[:, range(0, 9)]])
        y_train = np.hstack([y_train, data[:, 9]])
    x_train = np.array(x_train, dtype=np.float)

    f = open(path + "breast_cancer_wisconsin_cont_test.txt")
    data = np.loadtxt(f, delimiter=",", dtype=np.string0)
    x_test = np.array(data[:, range(0, 9)])
    y_test = np.array(data[:, 9])
    for j in range(replication - 1):
        x_test = np.vstack([x_test, data[:, range(0, 9)]])
        y_test = np.hstack([y_test, data[:, 9]])
    x_test = np.array(x_test, dtype=np.float)

    return x_train, y_train, x_test, y_test 

def test_lstm_forget_bias():
    forget_bias = 2.0
    stack = mx.rnn.SequentialRNNCell()
    stack.add(mx.rnn.LSTMCell(100, forget_bias=forget_bias, prefix='l0_'))
    stack.add(mx.rnn.LSTMCell(100, forget_bias=forget_bias, prefix='l1_'))

    dshape = (32, 1, 200)
    data = mx.sym.Variable('data')

    sym, _ = stack.unroll(1, data, merge_outputs=True)
    mod = mx.mod.Module(sym, label_names=None, context=mx.cpu(0))
    mod.bind(data_shapes=[('data', dshape)], label_shapes=None)

    mod.init_params()

    bias_argument = next(x for x in sym.list_arguments() if x.endswith('i2h_bias'))
    expected_bias = np.hstack([np.zeros((100,)),
                               forget_bias * np.ones(100, ), np.zeros((2 * 100,))])
    assert_allclose(mod.get_params()[0][bias_argument].asnumpy(), expected_bias) 

def build(self):
#{{{
        import numpy as np;
        self.W = shared((self.input_dim, 4 * self.output_dim),
                               name='{}_W'.format(self.name))
        self.U = shared((self.output_dim, 4 * self.output_dim),
                                     name='{}_U'.format(self.name))

        self.b = K.variable(np.hstack((np.zeros(self.output_dim),
                                        K.get_value(self.forget_bias_init(
                                                (self.output_dim,))),
                                        np.zeros(self.output_dim),
                                        np.zeros(self.output_dim))),
                                name='{}_b'.format(self.name))
        #self.c_0 = shared((self.output_dim,), name='{}_c_0'.format(self.name)  )
        #self.h_0 = shared((self.output_dim,), name='{}_h_0'.format(self.name)  )
        self.c_0=np.zeros(self.output_dim).astype(theano.config.floatX);
        self.h_0=np.zeros(self.output_dim).astype(theano.config.floatX);
        self.params=[self.W,self.U,
                        self.b,
                    # self.c_0,self.h_0
                    ];
        #}}} 

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 test_lstm_forget_bias():
    forget_bias = 2.0
    stack = gluon.rnn.SequentialRNNCell()
    stack.add(gluon.rnn.LSTMCell(100, i2h_bias_initializer=mx.init.LSTMBias(forget_bias), prefix='l0_'))
    stack.add(gluon.rnn.LSTMCell(100, i2h_bias_initializer=mx.init.LSTMBias(forget_bias), prefix='l1_'))

    dshape = (32, 1, 200)
    data = mx.sym.Variable('data')

    sym, _ = stack.unroll(1, data, merge_outputs=True)
    mod = mx.mod.Module(sym, label_names=None, context=mx.cpu(0))
    mod.bind(data_shapes=[('data', dshape)], label_shapes=None)

    mod.init_params()

    bias_argument = next(x for x in sym.list_arguments() if x.endswith('i2h_bias'))
    expected_bias = np.hstack([np.zeros((100,)),
                               forget_bias * np.ones(100, ), np.zeros((2 * 100,))])
    assert_allclose(mod.get_params()[0][bias_argument].asnumpy(), expected_bias) 

def make_data_iter_plan(self):
        "make a random data iteration plan"
        # truncate each bucket into multiple of batch-size
        bucket_n_batches = []
        for i in range(len(self.data)):
            bucket_n_batches.append(len(self.data[i]) / self.batch_size)
            self.data[i] = self.data[i][:int(bucket_n_batches[i]*self.batch_size)]

        bucket_plan = np.hstack([np.zeros(n, int)+i for i, n in enumerate(bucket_n_batches)])
        np.random.shuffle(bucket_plan)

        bucket_idx_all = [np.random.permutation(len(x)) for x in self.data]

        self.bucket_plan = bucket_plan
        self.bucket_idx_all = bucket_idx_all
        self.bucket_curr_idx = [0 for x in self.data]

        self.data_buffer = []
        self.label_buffer = []
        for i_bucket in range(len(self.data)):
            data = np.zeros((self.batch_size, self.buckets[i_bucket]))
            label = np.zeros((self.batch_size, self.buckets[i_bucket]))
            self.data_buffer.append(data)
            self.label_buffer.append(label) 

def output_shrink(K, L):
    """
    shrink the convolution output to half the size.
    used when both the annihilating filter and the uniform samples of sinusoids satisfy
    Hermitian symmetric.
    :param K: the annihilating filter size: K + 1
    :param L: length of the (complex-valued) b vector
    :return:
    """
    out_len = L - K
    if out_len % 2 == 0:
        half_out_len = np.int(out_len / 2.)
        mtx_r = np.hstack((np.eye(half_out_len),
                           np.zeros((half_out_len, half_out_len))))
        mtx_i = mtx_r
    else:
        half_out_len = np.int((out_len + 1) / 2.)
        mtx_r = np.hstack((np.eye(half_out_len),
                           np.zeros((half_out_len, half_out_len - 1))))
        mtx_i = np.hstack((np.eye(half_out_len - 1),
                           np.zeros((half_out_len - 1, half_out_len))))
    return linalg.block_diag(mtx_r, mtx_i) 

def backward_process(self, x, y, W, neuron_output):
        backward_output = []
        layer_num = len(neuron_output)
        score = np.dot(np.hstack((1, neuron_output[layer_num - 2])), W[layer_num - 1])
        error_gradient = np.array([-2 * (y - neuron_output[layer_num - 1][0]) * self.tanh_prime(score)])
        # error_gradient = np.array([np.sum(-2 * (y - score) * np.hstack((1, neuron_output[layer_num-2])))])
        backward_output.insert(0, error_gradient)
        # Hidden layer
        for i in range(layer_num - 2, -1, -1):
            if i == 0:
                score = np.dot(x, W[i])
            else:
                score = np.dot(np.hstack((1, neuron_output[i - 1])), W[i])
            error_gradient = np.dot(error_gradient, W[i + 1][1:].transpose()) * self.tanh_prime(score)
            backward_output.insert(0, error_gradient)
        return backward_output 

def _N(self,s,r):
        """
        Lagrange's interpolate function
        params:
            s,r:natural position of evalue point.2-array.
        returns:
            2x(2x4) shape function matrix.
        """
        la1=(1-s)/2
        la2=(1+s)/2
        lb1=(1-r)/2
        lb2=(1+r)/2
        N1=la1*lb1
        N2=la1*lb2
        N3=la2*lb1
        N4=la2*lb2

        N=np.hstack(N1*np.eye(2),N2*np.eye(2),N3*np.eye(2),N4*np.eye(2))
        return N 

def wave2input_image(wave, window, pos=0, pad=0):
    wave_image = np.hstack([wave[pos+i*sride:pos+(i+pad*2)*sride+dif].reshape(height+pad*2, sride) for i in range(256//sride)])[:,:254]
    wave_image *= window
    spectrum_image = np.fft.fft(wave_image, axis=1)
    input_image = np.abs(spectrum_image[:,:128].reshape(1, height+pad*2, 128), dtype=np.float32)

    np.clip(input_image, 1000, None, out=input_image)
    np.log(input_image, out=input_image)
    input_image += bias
    input_image /= scale

    if np.max(input_image) > 0.95:
        print('input image max bigger than 0.95', np.max(input_image))
    if np.min(input_image) < 0.05:
        print('input image min smaller than 0.05', np.min(input_image))

    return input_image 

def repmat(x, r, c):
    '''
    repmat(x, r, c) is equivalent to numpy.matlib.repmat(x, r, c) except that it works correctly for
      sparse matrices.
    '''
    if sps.issparse(x):
        row = sps.hstack([x for _ in range(c)])
        return sps.vstack([row for _ in range(r)], format=x.format)
    else: return np.matlib.repmat(x, r, c) 

def close_curves(*crvs, **kw):
    '''
    close_curves(crv1, crv2...) yields a single curve that merges all of the given list of curves
      together. The curves must be given in order, such that the i'th curve should be connected to
      to the (i+1)'th curve circularly to form a perimeter.

    The following optional parameters may be given:
      * grid may specify the number of grid-points to use in the initial search for a start-point
        (default: 16).
      * order may specify the order of the resulting curve; by default (None) uses the lowest order
        of all curves.
      * smoothing (None) the amount to smooth the points.
      * even_out (True) whether to even out the distances along the curve.
      * meta_data (None) an optional map of meta-data to give the spline representation.
    '''
    for k in six.iterkeys(kw):
        if k not in close_curves.default_options: raise ValueError('Unrecognized option: %s' % k)
    kw = {k:(kw[k] if k in kw else v) for (k,v) in six.iteritems(close_curves.default_options)}
    (grid, order) = (kw['grid'], kw['order'])
    crvs = [(crv if is_curve_spline(crv) else to_curve_spline(crv)).even_out() for crv in crvs]
    # find all intersections:
    isects = [curve_intersection(u,v, grid=grid)
              for (u,v) in zip(crvs, np.roll(crvs,-1))]
    # subsample curves
    crds = np.hstack([crv.subcurve(s1[1], s0[0]).coordinates[:,:-1]
                      for (crv,s0,s1) in zip(crvs, isects, np.roll(isects,1,0))])
    kw['order'] = np.min([crv.order for crv in crvs]) if order is None else order
    kw = {k:v for (k,v) in six.iteritems(kw)
          if v is not None and k in ('order','smoothing','even_out','meta_data')}
    return curve_spline(crds, periodic=True, **kw) 

def colors_to_cmap(colors):
    colors = np.asarray(colors)
    if len(colors.shape) == 1: return colors_to_cmap([colors])[0]
    if colors.shape[1] == 3:
        colors = np.hstack((colors, np.ones((len(colors),1))))
    steps = (0.5 + np.asarray(range(len(colors)-1), dtype=np.float))/(len(colors) - 1)
    return matplotlib.colors.LinearSegmentedColormap(
        'auto_cmap',
        {clrname: ([(0, col[0], col[0])] +
                   [(step, c0, c1) for (step,c0,c1) in zip(steps, col[:-1], col[1:])] +
                   [(1, col[-1], col[-1])])
         for (clridx,clrname) in enumerate(['red', 'green', 'blue', 'alpha'])
         for col in [colors[:,clridx]]},
        N=(len(colors))) 

def plotOutput(layer,feed_dict,fieldShape=None,channel=None,figOffset=1,cmap=None):
	# Output summary
	W = layer.output
	wp = W.eval(feed_dict=feed_dict);
	if len(np.shape(wp)) < 4:		# Fully connected layer, has no shape
		temp = np.zeros(np.product(fieldShape)); temp[0:np.shape(wp.ravel())[0]] = wp.ravel()
		fields = np.reshape(temp,[1]+fieldShape)
	else:			# Convolutional layer already has shape
		wp = np.rollaxis(wp,3,0)
		features, channels, iy,ix = np.shape(wp)
		if channel is not None:
			fields = wp[:,channel,:,:]
		else:
			fields = np.reshape(wp,[features*channels,iy,ix])

	perRow = int(math.floor(math.sqrt(fields.shape[0])))
	perColumn = int(math.ceil(fields.shape[0]/float(perRow)))
	fields2 = np.vstack([fields,np.zeros([perRow*perColumn-fields.shape[0]] + list(fields.shape[1:]))])
	tiled = []
	for i in range(0,perColumn*perRow,perColumn):
		tiled.append(np.hstack(fields2[i:i+perColumn]))

	tiled = np.vstack(tiled)
	if figOffset is not None:
		mpl.figure(figOffset); mpl.clf();

	mpl.imshow(tiled,cmap=cmap); mpl.title('%s Output' % layer.name); mpl.colorbar(); 

def hstack(tup):
    '''
    hstack(x) is equivalent to numpy.hstack(x) or scipy.sparse.hstack(x) except that it works
      correctly with both sparse and dense arrays (if any inputs are dense, it converts all inputs
      to dense arrays).
    '''
    if all([sps.issparse(u) for u in tup]): return sps.hstack(tup, format=tup[0].format)
    else: return np.hstack([u.toarray() if sps.issparse(u) else u for u in tup]) 

def plotFields(layer,fieldShape=None,channel=None,maxFields=25,figName='ReceptiveFields',cmap=None,padding=0.01):
	# Receptive Fields Summary
	W = layer.W
	wp = W.eval().transpose();
	if len(np.shape(wp)) < 4:		# Fully connected layer, has no shape
		fields = np.reshape(wp,list(wp.shape[0:-1])+fieldShape)
	else:			# Convolutional layer already has shape
		features, channels, iy, ix = np.shape(wp)
		if channel is not None:
			fields = wp[:,channel,:,:]
		else:
			fields = np.reshape(wp,[features*channels,iy,ix])

	fieldsN = min(fields.shape[0],maxFields)
	perRow = int(math.floor(math.sqrt(fieldsN)))
	perColumn = int(math.ceil(fieldsN/float(perRow)))

	fig = mpl.figure(figName); mpl.clf()

	# Using image grid
	from mpl_toolkits.axes_grid1 import ImageGrid
	grid = ImageGrid(fig,111,nrows_ncols=(perRow,perColumn),axes_pad=padding,cbar_mode='single')
	for i in range(0,fieldsN):
		im = grid[i].imshow(fields[i],cmap=cmap);

	grid.cbar_axes[0].colorbar(im)
	mpl.title('%s Receptive Fields' % layer.name)

	# old way
	# fields2 = np.vstack([fields,np.zeros([perRow*perColumn-fields.shape[0]] + list(fields.shape[1:]))])
	# tiled = []
	# for i in range(0,perColumn*perRow,perColumn):
	# 	tiled.append(np.hstack(fields2[i:i+perColumn]))
	#
	# tiled = np.vstack(tiled)
	# mpl.figure(figOffset); mpl.clf(); mpl.imshow(tiled,cmap=cmap); mpl.title('%s Receptive Fields' % layer.name); mpl.colorbar();
	mpl.figure(figName+' Total'); mpl.clf(); mpl.imshow(np.sum(np.abs(fields),0),cmap=cmap); mpl.title('%s Total Absolute Input Dependency' % layer.name); mpl.colorbar() 

def _siblings_to_pairs(rs):
        subject_list = [u for v in six.itervalues(rs)
                        for uuu in [[six.iterkeys(v)], six.itervalues(v)]
                        for uu in uuu for u in ([uu] if pimms.is_int(uu) else uu)]
        subject_list = np.unique(subject_list)
        # setup twin numbers so that we can export anonymized twin data (i.e.,
        # files containing twin data but not the subject IDs)
        twin_pairs = {tw: pimms.imm_array(list(sorted(dat)))
                      for tw  in ['MZ','DZ']
                      for dat in [set([tuple(sorted([k,v])) for (k,v) in six.iteritems(rs[tw])])]}
        # also get a list of all siblings so we can track who is/isn't related
        siblings = {}
        for s1 in subject_list:
            q = []
            for sibs in six.itervalues(rs):
                if s1 not in sibs: continue
                ss = sibs[s1]
                if pimms.is_int(ss): ss = [ss]
                for s2 in ss: q.append(s2)
            if len(q) > 0: siblings[s1] = q
        # Make up a list of all possible unrelated pairs
        unrelated_pairs = []
        for sid in subject_list:
            # find a random subject to pair them with
            urs = np.setdiff1d(subject_list, [sid] + siblings.get(sid,[]))
            unrelated_pairs.append([urs, np.full(len(urs), sid)])
        unrelated_pairs = np.unique(np.sort(np.hstack(unrelated_pairs), axis=0), axis=1).T
        unrelated_pairs.setflags(write=False)
        # Having made those unrelated pairs, we can add them to the twin pairs
        twin_pairs['UR'] = unrelated_pairs
        # finally, let's figure out the non-twin siblings:
        sibs = [(k,v) for (k,vv) in six.iteritems(rs['']) for v in vv]
        twin_pairs['SB'] = np.unique(np.sort(sibs, axis=1), axis=0)
        twin_pairs['SB'].setflags(write=False)
        return pyr.pmap({'monozygotic_twins': twin_pairs['MZ'],
                         'dizygotic_twins':   twin_pairs['DZ'],
                         'nontwin_siblings':  twin_pairs['SB'],
                         'unrelated_pairs':   twin_pairs['UR']}) 

def score_function(self, x, W):

        y_predict = x[1:]
        for i in range(0, len(W), 1):
            y_predict = np.tanh(np.dot(np.hstack((1, y_predict)), W[i]))

        score = y_predict[0]

        return score 

def forward_process(self, x, y, W):
        forward_output = []
        pre_x = x
        for i in range(len(W)):
            pre_x = np.tanh(np.dot(pre_x, W[i]))
            forward_output.append(pre_x)
            pre_x = np.hstack((1, pre_x))
        return forward_output 

def plot_saliency(raw_img, image_var, img_embedding_var, caption_var):
    dis = (caption_var.squeeze() * img_embedding_var.squeeze()).sum()
    dis.backward(retain_graph=True)

    grad = image_var.grad.data.cpu().squeeze().numpy().transpose((1, 2, 0))
    grad = normalize_grad(grad, stat=True)
    grad = imresize((grad * 255).astype('uint8'), (raw_img.height, raw_img.width)) / 255
    grad = normalize_grad(grad.mean(axis=-1, keepdims=True).repeat(3, axis=-1))
    grad = np.float_power(grad, args.grad_power)

    np_img = np.array(raw_img)
    masked_img = np_img * grad
    final = np.hstack([np_img, masked_img.astype('uint8'), (grad * 255).astype('uint8')])
    return Image.fromarray(final.astype('uint8')) 

def plotOutput(layer,feed_dict,fieldShape=None,channel=None,figOffset=1,cmap=None):
	# Output summary
	try:
		W = layer.output
	except:
		W = layer
	wp = W.eval(feed_dict=feed_dict);
	if len(np.shape(wp)) < 4:		# Fully connected layer, has no shape
		temp = np.zeros(np.product(fieldShape)); temp[0:np.shape(wp.ravel())[0]] = wp.ravel()
		fields = np.reshape(temp,[1]+fieldShape)
	else:			# Convolutional layer already has shape
		wp = np.rollaxis(wp,3,0)
		features, channels, iy,ix = np.shape(wp)
		if channel is not None:
			fields = wp[:,channel,:,:]
		else:
			fields = np.reshape(wp,[features*channels,iy,ix])

	perRow = int(math.floor(math.sqrt(fields.shape[0])))
	perColumn = int(math.ceil(fields.shape[0]/float(perRow)))
	fields2 = np.vstack([fields,np.zeros([perRow*perColumn-fields.shape[0]] + list(fields.shape[1:]))])
	tiled = []
	for i in range(0,perColumn*perRow,perColumn):
		tiled.append(np.hstack(fields2[i:i+perColumn]))

	tiled = np.vstack(tiled)
	if figOffset is not None:
		mpl.figure(figOffset); mpl.clf(); 

	mpl.imshow(tiled,cmap=cmap); mpl.title('%s Output' % layer.name); mpl.colorbar();