pulsatrix/api/learnable__scalar_8hpp_source.html

#pragma once


namespace pulsatrix {


class LearnableScalar {

public:

    explicit LearnableScalar(float initial_value = 0.0f);


    [[nodiscard]] float value() const { return value_; }


    [[nodiscard]] float grad() const { return grad_; }


    void accumulate_grad(float grad);


    void zero_grad();


    void step(float learning_rate);


private:

    float value_;

    float grad_ = 0.0f;

};


}  // namespace pulsatrix

pulsatrix::LearnableScalar
A bare learnable scalar (e.g. a GFlowNet loss's log Z), outside the Module/LRP hierarchy entirely.
Definition learnable_scalar.hpp:23

pulsatrix::LearnableScalar::step
void step(float learning_rate)
Plain SGD update: value_ -= learning_rate * grad_.

pulsatrix::LearnableScalar::value
float value() const
Current value.
Definition learnable_scalar.hpp:29

pulsatrix::LearnableScalar::LearnableScalar
LearnableScalar(float initial_value=0.0f)
Constructs a scalar with the given initial value and zero gradient.

pulsatrix::LearnableScalar::zero_grad
void zero_grad()
Resets the accumulated gradient to zero. Does not change value().

pulsatrix::LearnableScalar::grad
float grad() const
Accumulated gradient since the last zero_grad().
Definition learnable_scalar.hpp:32

pulsatrix::LearnableScalar::accumulate_grad
void accumulate_grad(float grad)
Adds to the accumulated gradient – mirrors Tensor::accumulate()'s across-multiple-contributions conve...

pulsatrix
Definition acquisition_functions.hpp:16