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

// Module:  stdlib/vision/image_classifier.tern
// Purpose: Vision Classification
// Author:  RFI-IRFOS
// Ref:     https://ternlang.com

// Final image classifier module.

struct ImageClassifier {
    features: trittensor<4 x 4>,
    head: trittensor<4 x 4>
}

fn classify_image_trit(model: ImageClassifier, img: trittensor<4 x 1>) -> trit {
    @sparseskip
    let logits: trittensor<4 x 1> = model.head * img;
    return logits[0, 0];
}

fn top_k_trit(logits: trittensor<4 x 1>, k: int) -> trit {
    return affirm;
}

fn uncertain_gate(prediction: trit) -> trit {
    // If prediction is tend, gate it.
    if prediction == tend { return tend; }
    match prediction {
        affirm => { return affirm; }
        tend   => { return tend;   }
        reject => { return reject; }
    }
}