ngclearn.utils.optim package

Submodules

ngclearn.utils.optim.adam module

class ngclearn.utils.optim.adam.Adam(learning_rate=0.001, beta1=0.9, beta2=0.999, epsilon=1e-08)[source]

Bases: Opt

Implements the adaptive moment estimation (Adam) algorithm as a decoupled update rule given adjustments produced by a credit assignment algorithm/process.

Parameters:

  • learning_rate – step size coefficient for Adam update

  • beta1 – 1st moment control factor

  • beta2 – 2nd moment control factor

  • epsilon – numberical stability coefficient (for calculating final update)

update(theta, updates)[source]

Apply an update tensor to the current “theta” (parameter) tensor according to an internally specified optimization/change rule.

Parameters:

  • theta – parameter value tensor to change

  • updates – externally produced updates to apply to “theta” (note that updates should be same shape as “theta” to ensure expected behavior)

ngclearn.utils.optim.adam.step_update(param, update, g1, g2, lr, beta1, beta2, time, eps)

Runs one step of Adam over a set of parameters given updates. The dynamics for any set of parameters is as follows:

g1 = beta1 * g1 + (1 - beta1) * update
g2 = beta2 * g2 + (1 - beta2) * (update)^2
g1_unbiased = g1 / (1 - beta1**time)
g2_unbiased = g2 / (1 - beta2**time)
param = param - lr * g1_unbiased / (sqrt(g2_unbiased) + epsilon)
Parameters:

  • param – parameter tensor to change/adjust

  • update – update tensor to be applied to parameter tensor (must be same shape as “param”)

  • g1 – first moment factor/correction factor to use in parameter update (must be same shape as “update”)

  • g2 – second moment factor/correction factor to use in parameter update (must be same shape as “update”)

  • lr – global step size value to be applied to updates to parameters

  • beta1 – 1st moment control factor

  • beta2 – 2nd moment control factor

  • time – current time t or iteration step/call to this Adam update

  • eps – numberical stability coefficient (for calculating final update)

Returns:

adjusted parameter tensor (same shape as “param”)

ngclearn.utils.optim.opt module

class ngclearn.utils.optim.opt.Opt(name)[source]

Bases: object

A generic base-class for an optimizer.

Parameters:

name – string name of optimizer

update(theta, updates)[source]

Apply an update tensor to the current “theta” (parameter) tensor according to an internally specified optimization/change rule.

Parameters:

  • theta – parameter value tensor to change

  • updates – externally produced updates to apply to “theta” (note that updates should be same shape as “theta” to ensure expected behavior)

ngclearn.utils.optim.sgd module

class ngclearn.utils.optim.sgd.SGD(learning_rate=0.001)[source]

Bases: Opt

Implements stochastic gradient descent (SGD) as a decoupled update rule given adjustments produced by a credit assignment algorithm/process.

Parameters:

learning_rate – step size coefficient for SGD update

update(theta, updates)[source]

Apply an update tensor to the current “theta” (parameter) tensor according to an internally specified optimization/change rule.

Parameters:

  • theta – parameter value tensor to change

  • updates – externally produced updates to apply to “theta” (note that updates should be same shape as “theta” to ensure expected behavior)

ngclearn.utils.optim.sgd.step_update(param, update, lr)

Runs one step of SGD over a set of parameters given updates.

Parameters:

lr – global step size to apply when adjusting parameters

Returns:

adjusted parameter tensor (same shape as “param”)

Module contents