ternlang-core 0.3.3

Compiler and VM for Ternlang — balanced ternary language with affirm/tend/reject trit semantics, @sparseskip codegen, and BET bytecode execution.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33

// Module:  stdlib/rl/policy.tern
// Purpose: RL Policy Gradient
// Author:  RFI-IRFOS
// Ref:     https://ternlang.com

// In ternary policies, 'tend' represents a hold or exploration action.

struct TritPolicy {
    model: trittensor<4 x 4>
}

fn select_action_trit(policy: TritPolicy, state: trittensor<4 x 1>) -> trit {
    @sparseskip
    let logits: trittensor<4 x 1> = policy.model * state;
    
    let best_action: trit = logits[0, 0];
    match best_action {
        affirm => { return affirm; } // Take action A
        tend   => { return tend;   } // Wait / Explore
        reject => { return reject; } // Take action B
    }
}

fn policy_gradient_trit(log_prob: trit, advantage: trit) -> trit {
    // Loss is scaled by advantage
    if advantage == affirm { return affirm; } // Reinforce
    return reject; // Discourage
}

fn entropy_regularize_trit(policy_out: trit) -> trit {
    if policy_out == tend { return affirm; } // High entropy
    return tend; // Low entropy
}