Module treez::autograd
[−]
[src]
Structs
Context |
implementation of reverse automatic differentiation |
Link | |
OpAdd |
y = a + b; dy/da = 1; dy/db = 1; |
OpCos |
y = cos(x); dy/dx = -sin(x) |
OpExponential |
y = ax; dy/dx = ln(a) * ax |
OpLog |
y = log_base(x); dy/dx = 1/(x*ln(base)) |
OpMul |
y = a*b; dy/da = b; dy/db = a |
OpSin |
y = sin(x); dy/dx = cos(x) |
OpTan |
y = tan(x); dy/dx = 1/(cos(x))2 |
Enums
OpType |
Traits
Op |
forward Op and gradient interface |
Functions
compute_grad |
computes dy/dx and other variables as well back propagating from y |
fwd_pass |
checker for link validity, computes forward values, saves eval order for backward pass in context, and returns ids for input links |
init | |
init_op | |
init_var |