Compression Strategies

Gilial offers different compression strategies to match your needs.

Balanced (Default)

The safest option for most use cases.

Characteristics:

• Retention rate: 72.8%
• Vectors removed: ~27.2%
• Ideal for: Production environments, safety-first approach

When to use:

• You want predictable, moderate compression
• Data integrity is important
• You need good balance between space savings and quality

client.compress(strategy="balanced", dry_run=False)

Aggressive

Maximum compression for space-critical scenarios.

Characteristics:

• Retention rate: 67%
• Vectors removed: ~33%
• Ideal for: Cost optimization, non-critical indexes

When to use:

• Storage costs are critical
• You can afford minor quality loss
• You have backups and can re-compress if needed

client.compress(strategy="aggressive", dry_run=False)

How Scoring Works

Both strategies use L2 norm (Euclidean magnitude) to score vectors:

score = sqrt(v₁² + v₂² + ... + v₇₆₈²)

Higher norms = higher quality vectors Lower norms = candidates for deletion

Dry-Run First!

Always preview compression with dry_run=True:

# Preview changes
preview = client.compress(strategy="balanced", dry_run=True)
print(f"Would delete: {preview.metadata['deleted_vectors']} vectors")

# If happy with preview, apply:
result = client.compress(strategy="balanced", dry_run=False)

Recommendations

1. Start with Balanced - It’s proven and safe
2. Test on non-critical index first - Understand the impact
3. Always use dry-run - Preview before applying
4. Maintain backups - Have a recovery plan
5. Monitor metrics - Track query latency before/after