forceschedule

Force schedule functions for driven DDPM diffusion models.

This module defines the base abstract class for force schedules and two concrete implementations that map a normalized time variable \(t \in [0,1]\) to a force scaling coefficient \(\alpha_{\mathrm{force}}(t)\).

Classes:

• ForceSchedule –

  Abstract base class for force schedules.

• LinearForceSchedule –

  Fades in force linearly: \(\alpha_{\mathrm{force}}(t) = 1 - t\).

• ConstantForceSchedule –

  Constant force: \(\alpha_{\mathrm{force}}(t) = 1\).

ForceSchedule

Bases: ABC

Abstract base class for force schedules.

Defines the interface for the time-dependent force scaling coefficient \(\alpha_{\mathrm{force}}(t)\) in driven DDPM models.

Methods:

• alpha_force –

  Compute force scaling at time \(t\).

alpha_force(t) abstractmethod

Compute the force scaling coefficient at time \(t\).

Parameters:

• t ((array_like, shape(dim1))) –

  Normalized time steps, where \(t \in [0, 1]\).

Returns:

• alpha_t ( (array_like, shape(dim1)) ) –

  Force scaling coefficients.

Source code in src/fpsl/ddm/forceschedule.py

@abstractmethod
def alpha_force(self, t: Float[ArrayLike, ' dim1']) -> Float[ArrayLike, ' dim1']:
    r"""Compute the force scaling coefficient at time $t$.

    Parameters
    ----------
    t : array_like, shape (dim1,)
        Normalized time steps, where $t \in [0, 1]$.

    Returns
    -------
    alpha_t : array_like, shape (dim1,)
        Force scaling coefficients.
    """
    raise NotImplementedError

LinearForceSchedule

Bases: ForceSchedule

Linearly fading force schedule.

Implements:

\[ \alpha_{\mathrm{force}}(t) = 1 - t \]

Methods:

• alpha_force –

  Returns \(1 - t\) for each time \(t\).

alpha_force(t)

Compute linearly fading force scaling.

Parameters:

• t ((array_like, shape(dim1))) –

  Normalized time steps.

Returns:

• alpha_t ( (array_like, shape(dim1)) ) –

  Force scaling coefficients equal to \(1 - t\).

Source code in src/fpsl/ddm/forceschedule.py

def alpha_force(self, t: Float[ArrayLike, ' dim1']) -> Float[ArrayLike, ' dim1']:
    r"""Compute linearly fading force scaling.

    Parameters
    ----------
    t : array_like, shape (dim1,)
        Normalized time steps.

    Returns
    -------
    alpha_t : array_like, shape (dim1,)
        Force scaling coefficients equal to $1 - t$.
    """
    return 1 - t

ConstantForceSchedule

Bases: ForceSchedule

Constant force schedule.

Implements:

\[ \alpha_{\mathrm{force}}(t) = 1 \]

Methods:

• alpha_force –

  Returns 1 for each time \(t\).

alpha_force(t)

Compute constant force scaling.

Parameters:

• t ((array_like, shape(dim1))) –

  Normalized time steps.

Returns:

• alpha_t ( (array_like, shape(dim1)) ) –

  Force scaling coefficients all equal to 1.

Source code in src/fpsl/ddm/forceschedule.py

def alpha_force(self, t: Float[ArrayLike, ' dim1']) -> Float[ArrayLike, ' dim1']:
    r"""Compute constant force scaling.

    Parameters
    ----------
    t : array_like, shape (dim1,)
        Normalized time steps.

    Returns
    -------
    alpha_t : array_like, shape (dim1,)
        Force scaling coefficients all equal to 1.
    """
    return jnp.ones_like(t)