Function encode

nodedb-vector/src/quantize/binary.rs:17–26

Binary Quantization (BQ): sign-bit encoding + Hamming distance. Each dimension is encoded as a single bit: 1 if positive, 0 if negative. 32x compression (D/8 bytes vs 4D bytes for FP32). Best used as a coarse pre-filter: compute Hamming distance to quickly eliminate far candidates before computing exact distances on survivors. Recall loss: 5-10% as a standalone index, but combined with a reranki

(vector: &[f32])

/// Bit layout: bit 0 of byte 0 = dimension 0, bit 1 = dimension 1, etc.
/// `output.len() = ceil(dim / 8)`.
pub fn encode(vector: &[f32]) -> Vec<u8> {
    let num_bytes = vector.len().div_ceil(8);
    let mut bits = vec![0u8; num_bytes];
    for (i, &val) in vector.iter().enumerate() {
        if val > 0.0 {
            bits[i / 8] |= 1 << (i % 8);
        }
    }
    bits
}

/// Batch encode: encode all vectors into contiguous binary representation.
///