Source code for apax.train.loss

import dataclasses
from typing import List

import einops
import jax
import jax.numpy as jnp
import jax.scipy as jsc
import numpy as np

from apax.utils.math import normed_dotp


[docs] def weighted_squared_error( label: jnp.array, prediction: jnp.array, name, divisor: float = 1.0, parameters: dict = {}, ) -> jnp.array: """ Squared error function that allows weighting of individual contributions by the number of atoms in the system. """ label, prediction = label[name], prediction[name] return (label - prediction) ** 2 / divisor
[docs] def weighted_huber_loss( label: jnp.array, prediction: jnp.array, name, divisor: float = 1.0, parameters: dict = {}, ) -> jnp.array: """ Huber loss function that allows weighting of individual contributions by the number of atoms in the system. """ label, prediction = label[name], prediction[name] if "delta" not in parameters.keys(): raise KeyError("Huber loss function requires 'delta' parameter") delta = parameters["delta"] diff = jnp.abs(label - prediction) loss = jnp.where(diff > delta, delta * (diff - 0.5 * delta), 0.5 * diff**2) return loss / divisor
[docs] def crps_loss( label: jax.Array, prediction: jax.Array, name, divisor: float = 1.0, parameters: dict = {}, ) -> jax.Array: """Computes the CRPS of a gaussian distribution given means, targets and standard deviations (uncertainty estimate) """ label = label[name] means = prediction[name] sigmas = prediction[name + "_uncertainty"] sigmas = jnp.clip(sigmas, a_min=1e-6) norm_x = (label - means) / sigmas cdf = 0.5 * (1 + jsc.special.erf(norm_x / jnp.sqrt(2))) normalization = 1 / (jnp.sqrt(2.0 * np.pi)) pdf = normalization * jnp.exp(-(norm_x**2) / 2.0) crps = sigmas * (norm_x * (2 * cdf - 1) + 2 * pdf - 1 / jnp.sqrt(np.pi)) return crps / divisor
[docs] def nll_loss( label: jax.Array, prediction: jax.Array, name, divisor: float = 1.0, parameters: dict = {}, ) -> jax.Array: """Computes the gaussian NLL loss given means, targets and standard deviations (uncertainty estimate) """ label = label[name] means = prediction[name] sigmas = prediction[name + "_uncertainty"] eps = 1e-6 sigmas = jnp.clip(sigmas, a_min=eps) variances = jnp.pow(sigmas, 2) x1 = jnp.log(variances) x2 = ((means - label) ** 2) / variances nll = 0.5 * (x1 + x2) return nll / divisor
[docs] def force_angle_loss( label: jnp.array, prediction: jnp.array, name, divisor: float = 1.0, parameters: dict = {}, ) -> jnp.array: """ Consine similarity loss function. Contributions are summed in `Loss`. """ label, prediction = label[name], prediction[name] dotp = normed_dotp(label, prediction) return (1.0 - dotp) / divisor
[docs] def force_angle_div_force_label( label: jnp.array, prediction: jnp.array, name, divisor: float = 1.0, parameters: dict = {}, ): """ Consine similarity loss function weighted by the norm of the force labels. Contributions are summed in `Loss`. """ label, prediction = label[name], prediction[name] dotp = normed_dotp(label, prediction) F_0_norm = jnp.linalg.norm(label, ord=2, axis=2, keepdims=False) loss = jnp.where(F_0_norm > 1e-6, (1.0 - dotp) / F_0_norm, jnp.zeros_like(dotp)) return loss
[docs] def force_angle_exponential_weight( label: jnp.array, prediction: jnp.array, name, divisor: float = 1.0, parameters: dict = {}, ) -> jnp.array: """ Consine similarity loss function exponentially scaled by the norm of the force labels. Contributions are summed in `Loss`. """ label, prediction = label[name], prediction[name] dotp = normed_dotp(label, prediction) F_0_norm = jnp.linalg.norm(label, ord=2, axis=2, keepdims=False) return (1.0 - dotp) * jnp.exp(-F_0_norm) / divisor
def stress_tril(label, prediction, name, divisor=1.0, parameters: dict = {}): label, prediction = label[name], prediction[name] idxs = jnp.tril_indices(3) label_tril = label[:, idxs[0], idxs[1]] prediction_tril = prediction[:, idxs[0], idxs[1]] return (label_tril - prediction_tril) ** 2 / divisor loss_functions = { "mse": weighted_squared_error, "huber": weighted_huber_loss, "cosine_sim": force_angle_loss, "cosine_sim_div_magnitude": force_angle_div_force_label, "cosine_sim_exp_magnitude": force_angle_exponential_weight, "tril": stress_tril, "crps": crps_loss, "nll": nll_loss, }
[docs] @dataclasses.dataclass class Loss: """ Represents a single weighted loss function that is constructed from a `name` and a type of comparison metric. """ name: str loss_type: str weight: float = 1.0 atoms_exponent: float = 1.0 parameters: dict = dataclasses.field(default_factory=lambda: {}) def __post_init__(self): if self.loss_type not in loss_functions.keys(): raise NotImplementedError( f"the loss function '{self.loss_type}' is not known." ) if self.name not in ["energy", "forces", "stress"]: raise NotImplementedError(f"the quantity '{self.name}' is not known.") self.loss_fn = loss_functions[self.loss_type] def __call__(self, inputs: dict, prediction: dict, label: dict) -> float: # TODO we may want to insert an additional `mask` argument for this method divisor = self.determine_divisor(inputs["n_atoms"]) batch_losses = self.loss_fn( label, prediction, self.name, divisor, self.parameters ) loss = self.weight * jnp.sum(jnp.mean(batch_losses, axis=0)) return loss def determine_divisor(self, n_atoms: jnp.array) -> jnp.array: # shape: batch divisor = n_atoms**self.atoms_exponent if self.name in ["forces", "stress"]: divisor = einops.repeat(divisor, "batch -> batch 1 1") return divisor
[docs] @dataclasses.dataclass class LossCollection: loss_list: List[Loss] def __call__(self, inputs: dict, predictions: dict, labels: dict) -> float: total_loss = 0.0 for single_loss_fn in self.loss_list: loss = single_loss_fn(inputs, predictions, labels) total_loss = total_loss + loss return total_loss