aarondb

-module(aarondb@cognitive@scoring).
-compile([no_auto_import, nowarn_unused_vars, nowarn_unused_function, nowarn_nomatch, inline]).
-define(FILEPATH, "src/aarondb/cognitive/scoring.gleam").
-export([exp/1, log/1, log10/1, log2/1, tanh/1, ebbinghaus_with_floor/3, compute_stability/2, hebbian_update/2, softmax/1, bayesian_update/2]).

-if(?OTP_RELEASE >= 27).
-define(MODULEDOC(Str), -moduledoc(Str)).
-define(DOC(Str), -doc(Str)).
-else.
-define(MODULEDOC(Str), -compile([])).
-define(DOC(Str), -compile([])).
-endif.

-file("src/aarondb/cognitive/scoring.gleam", 5).
-spec exp(float()) -> float().
exp(X) ->
    math:exp(X).

-file("src/aarondb/cognitive/scoring.gleam", 8).
-spec log(float()) -> float().
log(X) ->
    math:log(X).

-file("src/aarondb/cognitive/scoring.gleam", 11).
-spec log10(float()) -> float().
log10(X) ->
    math:log10(X).

-file("src/aarondb/cognitive/scoring.gleam", 14).
-spec log2(float()) -> float().
log2(X) ->
    math:log2(X).

-file("src/aarondb/cognitive/scoring.gleam", 17).
-spec tanh(float()) -> float().
tanh(X) ->
    math:tanh(X).

-file("src/aarondb/cognitive/scoring.gleam", 34).
?DOC(" EbbinghausWithFloor computes the Ebbinghaus retention with a floor value.\n").
-spec ebbinghaus_with_floor(float(), float(), float()) -> float().
ebbinghaus_with_floor(Days_since_access, Stability, Floor) ->
    Stab = case Stability =< +0.0 of
        true ->
            14.0;

        false ->
            Stability
    end,
    R = math:exp(+0.0 - (case Stab of
            +0.0 -> +0.0;
            -0.0 -> -0.0;
            Gleam@denominator -> Days_since_access / Gleam@denominator
        end)),
    case R < Floor of
        true ->
            Floor;

        false ->
            R
    end.

-file("src/aarondb/cognitive/scoring.gleam", 51).
?DOC(" ComputeStability computes new stability from access count and spacing.\n").
-spec compute_stability(integer(), float()) -> float().
compute_stability(Access_count, Avg_days_between_accesses) ->
    Count_float = erlang:float(Access_count),
    Base = math:log(1.0 + Count_float) * 20.0,
    Spacing = math:tanh(case 7.0 of
            +0.0 -> +0.0;
            -0.0 -> -0.0;
            Gleam@denominator -> Avg_days_between_accesses / Gleam@denominator
        end),
    Stability = Base * (1.0 + (0.5 * Spacing)),
    Stability1 = case Stability > 365.0 of
        true ->
            365.0;

        false ->
            Stability
    end,
    case Stability1 < 14.0 of
        true ->
            14.0;

        false ->
            Stability1
    end.

-file("src/aarondb/cognitive/scoring.gleam", 72).
?DOC(" Hebbian update for association weight\n").
-spec hebbian_update(float(), float()) -> float().
hebbian_update(Current_weight, Effective_signal) ->
    Clamped_current = case Current_weight =< +0.0 of
        true ->
            0.01;

        false ->
            Current_weight
    end,
    Log_new = math:log(Clamped_current) + (Effective_signal * math:log(
        1.0 + 0.01
    )),
    New_weight = math:exp(Log_new),
    case New_weight > 1.0 of
        true ->
            1.0;

        false ->
            New_weight
    end.

-file("src/aarondb/cognitive/scoring.gleam", 90).
?DOC(" Softmax normalizes the weight vector so values sum to 1.\n").
-spec softmax(list(float())) -> list(float()).
softmax(Weights) ->
    Max_w = case gleam@list:reduce(
        Weights,
        fun(Acc, W) -> gleam@float:max(Acc, W) end
    ) of
        {ok, M} ->
            M;

        {error, _} ->
            +0.0
    end,
    Shifted_exps = gleam@list:map(
        Weights,
        fun(W@1) -> math:exp(W@1 - Max_w) end
    ),
    Sum = case gleam@list:reduce(Shifted_exps, fun(Acc@1, E) -> Acc@1 + E end) of
        {ok, S} ->
            S;

        {error, _} ->
            1.0
    end,
    gleam@list:map(Shifted_exps, fun(E@1) -> case Sum of
                +0.0 -> +0.0;
                -0.0 -> -0.0;
                Gleam@denominator -> E@1 / Gleam@denominator
            end end).

-file("src/aarondb/cognitive/scoring.gleam", 122).
?DOC(" Bayesian update applies a Bayesian update to the prior confidence.\n").
-spec bayesian_update(float(), float()) -> float().
bayesian_update(Prior, Evidence) ->
    P = case Prior < +0.0 of
        true ->
            +0.0;

        false ->
            case Prior > 1.0 of
                true ->
                    1.0;

                false ->
                    Prior
            end
    end,
    E = case Evidence < +0.0 of
        true ->
            +0.0;

        false ->
            case Evidence > 1.0 of
                true ->
                    1.0;

                false ->
                    Evidence
            end
    end,
    Numerator = P * E,
    Denominator = Numerator + ((1.0 - P) * (1.0 - E)),
    D = case Denominator < 0.000000001 of
        true ->
            0.000000001;

        false ->
            Denominator
    end,
    Posterior = case D of
        +0.0 -> +0.0;
        -0.0 -> -0.0;
        Gleam@denominator -> Numerator / Gleam@denominator
    end,
    (0.95 * Posterior) + 0.025.