Skip to main content
v0.17 replaces the v0.16 getters (index_type, index_config) with explicit properties for the single DiskIVF index type: dimension, metric, and n_lists. Most are lazily fetched and cached on first read.

​
index_name

@property
index_name: str
Returns the name of the encrypted index. Always available locally — no network call.

​
Example Usage

print(f'Index name: {index.index_name}')
# Output: "my-vectors"

​
dimension

@property
dimension: int
Vector dimensionality.
  • Returns the real dimension when create_index was called with an explicit dimension, or after the first upsert (auto-detect).
  • Returns 0 if the index was created with auto-detect AND no upsert has happened yet.
Cached on first read; subsequent reads do not hit the server.

​
Example Usage

print(f'Vector dimensionality: {index.dimension}')
# Output: 384

​
metric

@property
metric: str
Distance metric: "euclidean", "squared_euclidean", or "cosine". Cached on first read alongside dimension.

​
Example Usage

print(f'Metric: {index.metric}')
# Output: "euclidean"

​
n_lists

@property
n_lists: int
Number of inverted lists in the IVF index.
  • Returns 1 for untrained indexes.
  • Returns the trained cluster count after train() completes.
Unlike dimension and metric, n_lists is fetched fresh on every read so callers reading immediately after training see the new value.

​
Example Usage

print(f'Lists: {index.n_lists}')
# Output: 1024  (after training)

​
is_trained

def is_trained(self) -> bool
Returns whether the index has been trained. Training is required for optimal query performance on large indexes.

​
Returns

bool: True if the index has been trained, False otherwise (or if the underlying describe call fails).

​
Example Usage

if not index.is_trained():
    print('Index needs training for optimal performance')
    index.train()

​
is_training

def is_training(self) -> bool
Returns whether this index is currently in the process of training (either actively training or queued).

​
Returns

bool: True if the index is currently training or queued, False otherwise.

​
Example Usage

import time

index.train()  # Kicks off training

while index.is_training():
    print('Index is still training...')
    time.sleep(5)
print('Index training complete')