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

// Module:  stdlib/nn/train/augmentation.tern
// Purpose: Data Augmentation for Trit Tensors
// Author:  RFI-IRFOS
// Ref:     https://ternlang.com

// Applying noise or transforms directly in trit space.

fn noise_trit(val: trit, probability: float) -> trit {
    // Corrupts a trit. Instead of flipping affirm to reject,
    // we often flip it to 'tend' (neutral/blurred).
    let should_corrupt: trit = reject; // Simulated randomness
    if should_corrupt == affirm {
        return tend;
    }
    return val;
}

fn jitter_trit(val: trit) -> trit {
    // Shifts sequence data
    return val;
}

fn flip_trit(tensor: trittensor<4 x 4>) -> trittensor<4 x 4> {
    // Flips a tensor horizontally/vertically
    return tensor;
}

fn mixup_trit(val_a: trit, val_b: trit) -> trit {
    // Averages two labels.
    if val_a == val_b { return val_a; }
    return tend; // Blended label is uncertain
}