Dataset Versioning & Reproducibility

from mixpeek import Mixpeek

client = Mixpeek(api_key="your-api-key")

# Create namespace for versioned training data
namespace = client.namespaces.create(
    namespace_name="training_data_v2"
)

# Create collection with multimodal extractors
collection = client.collections.create(
    collection_name="training_data_v2",
    feature_extractors=[{
        "feature_extractor_name": "multimodal-embed",
        "version": "v1"
    }]
)

# Index objects from versioned object storage (e.g., Tigris)
client.buckets.objects.create(
    bucket_id="raw-training-data",
    objects=[{
        "url": "s3://bucket/training_clip.mp4",
        "collection_destination": collection.collection_id
    }]
)

# Create cluster config for dataset snapshots
cluster = client.clusters.create(
    cluster_name="q4_training_freeze",
    collection_id=collection.collection_id,
    algorithm="hdbscan"
)

# Execute clustering to capture current state
execution = client.clusters.execute(cluster_id=cluster.cluster_id)

# Query dataset with time-based filters
results = client.retrievers.execute(
    retriever_id="versioned-search",
    query={"text": "product demos"},
    filters={"created_at": {"$gte": "2024-10-01T00:00:00Z"}}
)

# Compare cluster executions over time
history = client.clusters.executions.list(cluster_id=cluster.cluster_id)
print(f"Snapshots: {len(history.items)}")