Module flowcon.transforms.UMNN.MonotonicNormalizer

Classes

class ELUPlus

Base class for all neural network modules.

Your models should also subclass this class.

Modules can also contain other Modules, allowing to nest them in a tree structure. You can assign the submodules as regular attributes::

import torch.nn as nn
import torch.nn.functional as F

class Model(nn.Module):
    def __init__(self):
        super().__init__()
        self.conv1 = nn.Conv2d(1, 20, 5)
        self.conv2 = nn.Conv2d(20, 20, 5)

    def forward(self, x):
        x = F.relu(self.conv1(x))
        return F.relu(self.conv2(x))

Submodules assigned in this way will be registered, and will have their parameters converted too when you call :meth:to, etc.

Note

As per the example above, an __init__() call to the parent class must be made before assignment on the child.

:ivar training: Boolean represents whether this module is in training or evaluation mode. :vartype training: bool

Initialize internal Module state, shared by both nn.Module and ScriptModule.

Expand source code

class ELUPlus(nn.Module):
    def __init__(self):
        super().__init__()
        self.elu = nn.ELU()

    def forward(self, x):
        return self.elu(x) + 1.

Ancestors

• torch.nn.modules.module.Module

Class variables

var call_super_init : bool

var dump_patches : bool

var training : bool

Methods

def forward(self, x) ‑> Callable[..., Any]

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the :class:Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class IntegrandNet (hidden, cond_in)

Base class for all neural network modules.

Your models should also subclass this class.

Modules can also contain other Modules, allowing to nest them in a tree structure. You can assign the submodules as regular attributes::

import torch.nn as nn
import torch.nn.functional as F

class Model(nn.Module):
    def __init__(self):
        super().__init__()
        self.conv1 = nn.Conv2d(1, 20, 5)
        self.conv2 = nn.Conv2d(20, 20, 5)

    def forward(self, x):
        x = F.relu(self.conv1(x))
        return F.relu(self.conv2(x))

Submodules assigned in this way will be registered, and will have their parameters converted too when you call :meth:to, etc.

Note

As per the example above, an __init__() call to the parent class must be made before assignment on the child.

:ivar training: Boolean represents whether this module is in training or evaluation mode. :vartype training: bool

Initialize internal Module state, shared by both nn.Module and ScriptModule.

Expand source code

class IntegrandNet(nn.Module):
    def __init__(self, hidden, cond_in):
        super(IntegrandNet, self).__init__()
        l1 = [1 + cond_in] + hidden
        l2 = hidden + [1]
        layers = []
        for h1, h2 in zip(l1, l2):
            layers += [nn.Linear(h1, h2), nn.ReLU()]
        layers.pop()
        layers.append(ELUPlus())
        self.net = nn.Sequential(*layers)

    def forward(self, x, h):
        nb_batch, in_d = x.shape
        x = torch.cat((x, h), 1)
        x_he = x.view(nb_batch, -1, in_d).transpose(1, 2).contiguous().view(nb_batch * in_d, -1)
        y = self.net(x_he).view(nb_batch, -1)
        return y

Ancestors

• torch.nn.modules.module.Module

Class variables

var call_super_init : bool

var dump_patches : bool

var training : bool

Methods

def forward(self, x, h) ‑> Callable[..., Any]

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the :class:Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class MonotonicNormalizer (integrand_net, cond_size, nb_steps=20, solver='CC')

Base class for all neural network modules.

Your models should also subclass this class.

Modules can also contain other Modules, allowing to nest them in a tree structure. You can assign the submodules as regular attributes::

import torch.nn as nn
import torch.nn.functional as F

class Model(nn.Module):
    def __init__(self):
        super().__init__()
        self.conv1 = nn.Conv2d(1, 20, 5)
        self.conv2 = nn.Conv2d(20, 20, 5)

    def forward(self, x):
        x = F.relu(self.conv1(x))
        return F.relu(self.conv2(x))

Submodules assigned in this way will be registered, and will have their parameters converted too when you call :meth:to, etc.

Note

As per the example above, an __init__() call to the parent class must be made before assignment on the child.

:ivar training: Boolean represents whether this module is in training or evaluation mode. :vartype training: bool

Initialize internal Module state, shared by both nn.Module and ScriptModule.

Expand source code

class MonotonicNormalizer(nn.Module):
    def __init__(self, integrand_net, cond_size, nb_steps=20, solver="CC"):
        super(MonotonicNormalizer, self).__init__()
        if type(integrand_net) is list:
            self.integrand_net = IntegrandNet(integrand_net, cond_size)
        else:
            self.integrand_net = integrand_net
        self.solver = solver
        self.nb_steps = nb_steps

    def forward(self, x, h, context=None):
        x0 = torch.zeros(x.shape).to(x.device)
        xT = x
        z0 = h[:, :, 0]
        h = h.permute(0, 2, 1).contiguous().view(x.shape[0], -1)
        if self.solver == "CC":
            z = NeuralIntegral.apply(x0, xT, self.integrand_net, _flatten(self.integrand_net.parameters()),
                                     h, self.nb_steps) + z0
        elif self.solver == "CCParallel":
            z = ParallelNeuralIntegral.apply(x0, xT, self.integrand_net,
                                             _flatten(self.integrand_net.parameters()),
                                             h, self.nb_steps) + z0
        else:
            return None
        return z, self.integrand_net(x, h)

    def inverse_transform(self, z, h, context=None):
        # Old inversion by binary search
        x_max = torch.ones_like(z) * 20
        x_min = -torch.ones_like(z) * 20
        z_max, _ = self.forward(x_max, h, context)
        z_min, _ = self.forward(x_min, h, context)
        for i in range(25):
            x_middle = (x_max + x_min) / 2
            z_middle, _ = self.forward(x_middle, h, context)
            left = (z_middle > z).float()
            right = 1 - left
            x_max = left * x_middle + right * x_max
            x_min = right * x_middle + left * x_min
            z_max = left * z_middle + right * z_max
            z_min = right * z_middle + left * z_min
        return (x_max + x_min) / 2

Ancestors

• torch.nn.modules.module.Module

Class variables

var call_super_init : bool

var dump_patches : bool

var training : bool

Methods

def forward(self, x, h, context=None) ‑> Callable[..., Any]

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the :class:Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

def inverse_transform(self, z, h, context=None)