ternlang-core 1.2.9

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/timeseries/tcn.tern
// Purpose: Temporal Convolutional Network
// Author:  RFI-IRFOS
// Ref:     https://ternlang.com

// TCN uses dilated causal convolutions.

struct TCNBlock {
    dilation: int,
    w: trittensor<4 x 4>
}

fn dilated_conv_trit(block: TCNBlock, seq: trittensor<4 x 1>) -> trittensor<4 x 1> {
    @sparseskip
    let out: trittensor<4 x 1> = block.w * seq;
    return out;
}

fn causal_conv_trit(block: TCNBlock, seq: trittensor<4 x 1>) -> trittensor<4 x 1> {
    // Ensures no future information leaks
    @sparseskip
    let out: trittensor<4 x 1> = block.w * seq;
    return out;
}

fn tcn_forward(block: TCNBlock, seq: trittensor<4 x 1>) -> trit {
    let conv: trittensor<4 x 1> = causal_conv_trit(block, seq);
    let res: trit = conv[0, 0];
    match res {
        affirm => { return affirm; }
        tend   => { return tend;   }
        reject => { return reject; }
    }
}