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
34

// Module:  stdlib/models/cnn.tern
// Purpose: Convolutional Neural Network in Ternary
// Author:  RFI-IRFOS
// Ref:     https://ternlang.com

// CNN leveraging ternary convolutions and pooling.

struct TritCNN {
    conv1: trittensor<4 x 4>,
    dense1: trittensor<4 x 4>
}

fn feature_extract_trit(model: TritCNN, image: trittensor<4 x 1>) -> trittensor<4 x 1> {
    @sparseskip
    let features: trittensor<4 x 1> = model.conv1 * image;
    return features;
}

fn classify_trit(model: TritCNN, features: trittensor<4 x 1>) -> trit {
    @sparseskip
    let logits: trittensor<4 x 1> = model.dense1 * features;
    
    let pred: trit = logits[0, 0];
    match pred {
        affirm => { return affirm; }
        tend   => { return tend;   }
        reject => { return reject; }
    }
}

fn cnn_forward(model: TritCNN, image: trittensor<4 x 1>) -> trit {
    let feat: trittensor<4 x 1> = feature_extract_trit(model, image);
    return classify_trit(model, feat);
}