nip.utils.maths

Utilities for useful mathematical operations.

Functions

_zero_grad(params)	Zero the gradients of the parameters in a dictionary.
aggregate_mean_grouped_by_class(values, classes)	Compute the mean of values grouped by class.
compute_conjugate_gradient_ihvp(...)	Approximate the inverse Hessian vector product with conjugate gradient.
compute_neumann_ihvp(follower_loss, ...)	Approximate the inverse Hessian vector product with the Neumann method.
compute_nystrom_ihvp(follower_loss, ...[, ...])	Approximate the inverse Hessian vector product with the Nystrom method.
compute_sos_update(simultaneous_grad, ...)	Compute the update for the Stable Opponent Shaping (SOS) algorithm.
dict_dot_product(dict_1, dict_2)	Calculate the dot product between two dictionaries of tensors.
dict_scalar_multiple(dictionary, scalar)	Calculate a scalar multiple of a dictionary of tensors.
dict_sum(dict_1, dict_2)	Calculate the sum of two dictionaries of tensors, element-wise.
inverse_hessian_vector_product(...[, rank, ...])	Compute the inverse Hessian-vector product using specified approximation method.
is_broadcastable(shape_1, shape_2)	Check if two shapes are broadcastable.
logit_entropy(logits)	Compute the entropy of a set of logits.
logit_or(logits[, dim])	Compute the logit of the OR of n events given their logits.
logit_or_dual(a, b)	Compute the logit OR operation for two input tensors with the log-sum-exp trick.
mean_episode_reward(reward, done_mask)	Compute the mean total episode reward for a batch of concatenated episodes.
mean_for_unique_keys(data, key[, axis])	Compute the mean of values grouped by unique keys.
minstd_generate_pseudo_random_sequence(seed, ...)	Generate a pseudo-random sequence of numbers using the MINSTD algorithm.
set_seed(seed)	Set the seed in Python, NumPy, and PyTorch.