Module flowcon.CNF.cnf

Classes

class CNF (odefunc, T=1.0, train_T=False, regularization_fns=None, solver='dopri5', atol=1e-05, rtol=1e-05)

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 CNF(nn.Module):
    def __init__(self, odefunc, T=1.0, train_T=False, regularization_fns=None, solver='dopri5', atol=1e-5, rtol=1e-5):
        super(CNF, self).__init__()
        if train_T:
            self.register_parameter("sqrt_end_time", nn.Parameter(torch.sqrt(torch.tensor(T))))
        else:
            self.register_buffer("sqrt_end_time", torch.sqrt(torch.tensor(T)))

        nreg = 0
        if regularization_fns is not None:
            odefunc = RegularizedODEfunc(odefunc, regularization_fns)
            nreg = len(regularization_fns)
        self.odefunc = odefunc
        self.nreg = nreg
        self.regularization_states = None
        self.solver = solver
        self.atol = atol
        self.rtol = rtol
        self.test_solver = solver
        self.test_atol = atol
        self.test_rtol = rtol
        self.solver_options = {}

    def forward(self, z, logpz=None, integration_times=None, reverse=False):

        if logpz is None:
            _logpz = torch.zeros(z.shape[0], 1).to(z)
        else:
            _logpz = logpz

        if integration_times is None:
            integration_times = torch.tensor([0.0, self.sqrt_end_time * self.sqrt_end_time]).to(z)
        if reverse:
            integration_times = _flip(integration_times, 0)

        # Refresh the odefunc statistics.
        self.odefunc.before_odeint()

        # Add regularization states.
        reg_states = tuple(torch.tensor(0).to(z) for _ in range(self.nreg))

        if self.training:
            state_t = odeint(
                self.odefunc,
                (z, _logpz) + reg_states,
                integration_times.to(z),
                atol=self.atol,
                rtol=self.rtol,
                method=self.solver,
                options=self.solver_options,
                adjoint_options={"norm": "seminorm"}
                # step_size = self.solver_options["step_size"]
            )
        else:
            state_t = odeint(
                self.odefunc,
                (z, _logpz),
                integration_times.to(z),
                atol=self.test_atol,
                rtol=self.test_rtol,
                method=self.test_solver,
                adjoint_options={"norm": "seminorm"}
                # step_size=self.solver_options["step_size"]
            )

        if len(integration_times) == 2:
            state_t = tuple(s[1] for s in state_t)

        z_t, logpz_t = state_t[:2]
        self.regularization_states = state_t[2:]

        if logpz is not None:
            return z_t, logpz_t
        else:
            return z_t

    def get_regularization_states(self):
        reg_states = self.regularization_states
        self.regularization_states = None
        return reg_states

    def num_evals(self):
        return self.odefunc._num_evals.item()

Ancestors

• torch.nn.modules.module.Module

Class variables

var call_super_init : bool

var dump_patches : bool

var training : bool

Methods

def forward(self, z, logpz=None, integration_times=None, reverse=False) ‑> 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 get_regularization_states(self)

def num_evals(self)