@lancedb/lancedb 0.41.0 → 0.42.0-beta.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/dist/blob.d.ts CHANGED
@@ -173,7 +173,7 @@ export declare class BlobFile {
173
173
  export type BlobReadOptions = {
174
174
  /**
175
175
  * Cancels the read. The call rejects with `signal.reason`, and the native
176
- * read stops, including in-flight requests to a remote table.
176
+ * read stops, including in-flight requests to a LanceDB Distributed table.
177
177
  */
178
178
  signal?: AbortSignal;
179
179
  };
@@ -213,9 +213,9 @@ export declare abstract class Connection {
213
213
  /**
214
214
  * Define a materialized view named `name` over the table `source`.
215
215
  *
216
- * Local creation and refresh require a source build with the
216
+ * LanceDB Embedded creation and refresh require a source build with the
217
217
  * `experimental-functions` Cargo feature, disabled in published packages.
218
- * Remote Cloud/Enterprise operations do not require this feature.
218
+ * LanceDB Distributed operations do not require this feature.
219
219
  *
220
220
  * The view is populated before creation returns. Set `withNoData` to create
221
221
  * only its definition and empty backing table. The view is a normal table:
@@ -275,18 +275,17 @@ export declare abstract class Connection {
275
275
  */
276
276
  abstract describeView(name: string, namespacePath?: string[]): Promise<ViewDescription>;
277
277
  /**
278
- * Drop the view named `name` and wait for its definition to be deleted.
278
+ * Drop the view named `name` and wait for catalog removal.
279
279
  *
280
280
  * The tables it reads are untouched: a view holds no rows of its own. Use
281
- * {@link dropViewAsync} to retain the cleanup job instead of waiting on it.
281
+ * {@link dropViewAsync} to retain its SQL query job. Completion confirms
282
+ * catalog removal; physical cleanup continues in the background.
282
283
  */
283
284
  abstract dropView(name: string, namespacePath?: string[]): Promise<void>;
284
285
  /**
285
- * Start dropping the view named `name` and return the job deleting its
286
- * definition, without waiting for completion.
287
- *
288
- * The name is free before this resolves. When nothing was bound to it, the
289
- * returned job is already finished and has no id.
286
+ * Drop a view and return its SQL query job.
287
+ * The job has type `query` and its ID is the SQL query UUID. Waiting confirms
288
+ * catalog removal; physical cleanup continues in the background.
290
289
  */
291
290
  abstract dropViewAsync(name: string, namespacePath?: string[]): Promise<Job>;
292
291
  /**
@@ -421,9 +420,10 @@ export declare abstract class Connection {
421
420
  /**
422
421
  * Rename a table.
423
422
  *
424
- * Currently only supported by LanceDB Cloud. Local OSS connections and
425
- * namespace-backed connections (via {@link connectNamespace}) reject with
426
- * a "not supported" error.
423
+ * Supported in LanceDB Distributed and by namespace backends that implement
424
+ * renaming. Directory namespaces support renaming tables resolved through their
425
+ * manifest, including tables within or across child namespaces. LanceDB Embedded
426
+ * connections and directory-listing root tables reject with a "not supported" error.
427
427
  *
428
428
  * @param {string} currentName - The current name of the table.
429
429
  * @param {string} newName - The new name for the table.
package/dist/expr.d.ts ADDED
@@ -0,0 +1,64 @@
1
+ import { Expr as NativeExpr } from "./native.js";
2
+ export { Expr as NativeExpr } from "./native";
3
+ /** Values accepted by expression operators. */
4
+ export type ExprLike = Expr | string | number | boolean;
5
+ /**
6
+ * A SQL expression backed by the shared LanceDB expression builder.
7
+ *
8
+ * @example
9
+ * ```ts
10
+ * table.query().where(col("firstName").eq("Alice")).toArray();
11
+ * ```
12
+ */
13
+ export declare class Expr {
14
+ inner: NativeExpr;
15
+ constructor(inner: NativeExpr);
16
+ /** Compare for equality. */
17
+ eq(other: ExprLike): Expr;
18
+ /** Compare for inequality. */
19
+ ne(other: ExprLike): Expr;
20
+ /** Compare for less than. */
21
+ lt(other: ExprLike): Expr;
22
+ /** Compare for less than or equal to. */
23
+ lte(other: ExprLike): Expr;
24
+ /** Compare for greater than. */
25
+ gt(other: ExprLike): Expr;
26
+ /** Compare for greater than or equal to. */
27
+ gte(other: ExprLike): Expr;
28
+ /** Combine two predicates with AND. */
29
+ and(other: Expr): Expr;
30
+ /** Combine two predicates with OR. */
31
+ or(other: Expr): Expr;
32
+ /** Negate this predicate. */
33
+ not(): Expr;
34
+ /** Add a value. */
35
+ add(other: ExprLike): Expr;
36
+ /** Subtract a value. */
37
+ sub(other: ExprLike): Expr;
38
+ /** Multiply by a value. */
39
+ mul(other: ExprLike): Expr;
40
+ /** Divide by a value. */
41
+ div(other: ExprLike): Expr;
42
+ /** Convert a string to lowercase. */
43
+ lower(): Expr;
44
+ /** Convert a string to uppercase. */
45
+ upper(): Expr;
46
+ /** Test whether a string contains a substring. */
47
+ contains(substr: ExprLike): Expr;
48
+ /** Test membership in a list of values. */
49
+ isIn(values: ExprLike[]): Expr;
50
+ /** Cast to an Arrow data type, such as "Int64". */
51
+ cast(dataType: string): Expr;
52
+ /** Serialize with the LanceDB SQL dialect, preserving column case. */
53
+ toSql(): string;
54
+ }
55
+ /** Reference a column by its exact, case-sensitive name. */
56
+ export declare function col(name: string): Expr;
57
+ /** Create a string or numeric literal. Unsafe integers are represented as Float64. */
58
+ export declare function lit(value: string | number): Expr;
59
+ /** Create a boolean literal. */
60
+ export declare function litBool(value: boolean): Expr;
61
+ /** Create an Int64 literal from a JavaScript safe integer. Other numbers throw RangeError. */
62
+ export declare function litInt(value: number): Expr;
63
+ /** Call a registered SQL function, such as `func("lower", col("firstName"))`. */
64
+ export declare function func(name: string, ...args: ExprLike[]): Expr;
package/dist/expr.js ADDED
@@ -0,0 +1,148 @@
1
+ "use strict";
2
+ // SPDX-License-Identifier: Apache-2.0
3
+ // SPDX-FileCopyrightText: Copyright The LanceDB Authors
4
+ Object.defineProperty(exports, "__esModule", { value: true });
5
+ exports.Expr = exports.NativeExpr = void 0;
6
+ exports.col = col;
7
+ exports.lit = lit;
8
+ exports.litBool = litBool;
9
+ exports.litInt = litInt;
10
+ exports.func = func;
11
+ const native_js_1 = require("./native.js");
12
+ var native_1 = require("./native");
13
+ Object.defineProperty(exports, "NativeExpr", { enumerable: true, get: function () { return native_1.Expr; } });
14
+ function coerce(value) {
15
+ if (value instanceof Expr) {
16
+ return value;
17
+ }
18
+ if (typeof value === "string") {
19
+ return lit(value);
20
+ }
21
+ if (typeof value === "boolean") {
22
+ return litBool(value);
23
+ }
24
+ if (typeof value === "number") {
25
+ return lit(value);
26
+ }
27
+ throw new TypeError(`Unsupported literal type: ${typeof value}`);
28
+ }
29
+ /**
30
+ * A SQL expression backed by the shared LanceDB expression builder.
31
+ *
32
+ * @example
33
+ * ```ts
34
+ * table.query().where(col("firstName").eq("Alice")).toArray();
35
+ * ```
36
+ */
37
+ class Expr {
38
+ inner;
39
+ constructor(inner) {
40
+ this.inner = inner;
41
+ }
42
+ /** Compare for equality. */
43
+ eq(other) {
44
+ return new Expr(this.inner.eq(coerce(other).inner));
45
+ }
46
+ /** Compare for inequality. */
47
+ ne(other) {
48
+ return new Expr(this.inner.ne(coerce(other).inner));
49
+ }
50
+ /** Compare for less than. */
51
+ lt(other) {
52
+ return new Expr(this.inner.lt(coerce(other).inner));
53
+ }
54
+ /** Compare for less than or equal to. */
55
+ lte(other) {
56
+ return new Expr(this.inner.lte(coerce(other).inner));
57
+ }
58
+ /** Compare for greater than. */
59
+ gt(other) {
60
+ return new Expr(this.inner.gt(coerce(other).inner));
61
+ }
62
+ /** Compare for greater than or equal to. */
63
+ gte(other) {
64
+ return new Expr(this.inner.gte(coerce(other).inner));
65
+ }
66
+ /** Combine two predicates with AND. */
67
+ and(other) {
68
+ return new Expr(this.inner.and(other.inner));
69
+ }
70
+ /** Combine two predicates with OR. */
71
+ or(other) {
72
+ return new Expr(this.inner.or(other.inner));
73
+ }
74
+ /** Negate this predicate. */
75
+ not() {
76
+ return new Expr(this.inner.not());
77
+ }
78
+ /** Add a value. */
79
+ add(other) {
80
+ return new Expr(this.inner.add(coerce(other).inner));
81
+ }
82
+ /** Subtract a value. */
83
+ sub(other) {
84
+ return new Expr(this.inner.sub(coerce(other).inner));
85
+ }
86
+ /** Multiply by a value. */
87
+ mul(other) {
88
+ return new Expr(this.inner.mul(coerce(other).inner));
89
+ }
90
+ /** Divide by a value. */
91
+ div(other) {
92
+ return new Expr(this.inner.div(coerce(other).inner));
93
+ }
94
+ /** Convert a string to lowercase. */
95
+ lower() {
96
+ return new Expr(this.inner.lower());
97
+ }
98
+ /** Convert a string to uppercase. */
99
+ upper() {
100
+ return new Expr(this.inner.upper());
101
+ }
102
+ /** Test whether a string contains a substring. */
103
+ contains(substr) {
104
+ return new Expr(this.inner.contains(coerce(substr).inner));
105
+ }
106
+ /** Test membership in a list of values. */
107
+ isIn(values) {
108
+ return new Expr(this.inner.isIn(values.map((v) => coerce(v).inner)));
109
+ }
110
+ /** Cast to an Arrow data type, such as "Int64". */
111
+ cast(dataType) {
112
+ return new Expr(this.inner.cast(dataType));
113
+ }
114
+ /** Serialize with the LanceDB SQL dialect, preserving column case. */
115
+ toSql() {
116
+ return this.inner.toSql();
117
+ }
118
+ }
119
+ exports.Expr = Expr;
120
+ /** Reference a column by its exact, case-sensitive name. */
121
+ function col(name) {
122
+ return new Expr((0, native_js_1.col)(name));
123
+ }
124
+ /** Create a string or numeric literal. Unsafe integers are represented as Float64. */
125
+ function lit(value) {
126
+ if (typeof value === "string") {
127
+ return new Expr((0, native_js_1.lit)(value));
128
+ }
129
+ if (Number.isSafeInteger(value)) {
130
+ return new Expr((0, native_js_1.litInt)(value));
131
+ }
132
+ return new Expr((0, native_js_1.lit)(value));
133
+ }
134
+ /** Create a boolean literal. */
135
+ function litBool(value) {
136
+ return new Expr((0, native_js_1.litBool)(value));
137
+ }
138
+ /** Create an Int64 literal from a JavaScript safe integer. Other numbers throw RangeError. */
139
+ function litInt(value) {
140
+ if (!Number.isSafeInteger(value)) {
141
+ throw new RangeError("Integer literals must be safe integers");
142
+ }
143
+ return new Expr((0, native_js_1.litInt)(value));
144
+ }
145
+ /** Call a registered SQL function, such as `func("lower", col("firstName"))`. */
146
+ function func(name, ...args) {
147
+ return new Expr((0, native_js_1.func)(name, args.map((a) => coerce(a).inner)));
148
+ }
package/dist/header.d.ts CHANGED
@@ -1,5 +1,5 @@
1
1
  /**
2
- * Header providers for LanceDB remote connections.
2
+ * Header providers for LanceDB Enterprise connections.
3
3
  *
4
4
  * This module provides a flexible header management framework for LanceDB remote
5
5
  * connections, allowing users to implement custom header strategies for
package/dist/header.js CHANGED
@@ -4,7 +4,7 @@
4
4
  Object.defineProperty(exports, "__esModule", { value: true });
5
5
  exports.OAuthHeaderProvider = exports.StaticHeaderProvider = exports.HeaderProvider = void 0;
6
6
  /**
7
- * Header providers for LanceDB remote connections.
7
+ * Header providers for LanceDB Enterprise connections.
8
8
  *
9
9
  * This module provides a flexible header management framework for LanceDB remote
10
10
  * connections, allowing users to implement custom header strategies for
package/dist/index.d.ts CHANGED
@@ -26,6 +26,8 @@ export { Scannable, ScannableOptions } from "./scannable";
26
26
  export * as rerankers from "./rerankers";
27
27
  export { SchemaLike, TableLike, DataTypeLike, FieldLike, RecordBatchLike, DataLike, IntoVector, MultiVector, } from "./arrow";
28
28
  export { IntoSql, packBits } from "./util";
29
+ export { Expr, col, lit, litBool, litInt, func } from "./expr";
30
+ export type { ExprLike } from "./expr";
29
31
  /**
30
32
  * Options for tokenizing a full-text search query without a table index.
31
33
  */
@@ -74,11 +76,11 @@ export declare function tokenize(query: string, options?: Partial<TokenizeOption
74
76
  *
75
77
  * Accepted formats:
76
78
  *
77
- * - `/path/to/database` - local database
79
+ * - `/path/to/database` - database on the local file system
78
80
  * - `s3://bucket/path/to/database` or `gs://bucket/path/to/database` - database on cloud storage
79
- * - `db://host:port` - remote database (LanceDB cloud)
81
+ * - `db://host:port` - LanceDB Enterprise or LanceDB Cloud (legacy) database
80
82
  * @param {string} uri - The uri of the database. If the database uri starts
81
- * with `db://` then it connects to a remote database.
83
+ * with `db://` then it connects to LanceDB Enterprise or LanceDB Cloud (legacy).
82
84
  * @see {@link ConnectionOptions} for more details on the URI format.
83
85
  * @param options - The options to use when connecting to the database
84
86
  * @example
@@ -111,9 +113,9 @@ export declare function connect(uri: string, options?: Partial<ConnectionOptions
111
113
  *
112
114
  * Accepted formats:
113
115
  *
114
- * - `/path/to/database` - local database
116
+ * - `/path/to/database` - database on the local file system
115
117
  * - `s3://bucket/path/to/database` or `gs://bucket/path/to/database` - database on cloud storage
116
- * - `db://host:port` - remote database (LanceDB cloud)
118
+ * - `db://host:port` - LanceDB Enterprise or LanceDB Cloud (legacy) database
117
119
  * @param options - The options to use when connecting to the database
118
120
  * @see {@link ConnectionOptions} for more details on the URI format.
119
121
  * @example
@@ -186,7 +188,7 @@ export interface RestNamespaceConfig {
186
188
  * Connect to a LanceDB database through a namespace.
187
189
  *
188
190
  * Unlike {@link connect}, which routes by URI scheme (local path vs.
189
- * `db://` cloud), `connectNamespace` always returns a namespace-backed
191
+ * `db://`), `connectNamespace` always returns a namespace-backed
190
192
  * connection. The `implName` selects the namespace implementation:
191
193
  *
192
194
  * - `"dir"` — directory namespace, configured with {@link DirNamespaceConfig}.
package/dist/index.js CHANGED
@@ -2,7 +2,8 @@
2
2
  // SPDX-License-Identifier: Apache-2.0
3
3
  // SPDX-FileCopyrightText: Copyright The LanceDB Authors
4
4
  Object.defineProperty(exports, "__esModule", { value: true });
5
- exports.Listing = exports.connectCatalog = exports.DatabaseNames = exports.Catalog = exports.packBits = exports.rerankers = exports.Scannable = exports.PermutationBuilder = exports.permutationBuilder = exports.embedding = exports.MergeInsertBuilder = exports.OAuthSession = exports.OAuthFlowType = exports.ClientAuthMethod = exports.OAuthHeaderProvider = exports.StaticHeaderProvider = exports.HeaderProvider = exports.Branches = exports.Table = exports.Index = exports.Occur = exports.Operator = exports.FullTextQueryType = exports.BooleanQuery = exports.MultiMatchQuery = exports.BoostQuery = exports.PhraseQuery = exports.MatchQuery = exports.RecordBatchIterator = exports.TakeQuery = exports.VectorQuery = exports.QueryBase = exports.Query = exports.AutoQuery = exports.Job = exports.Session = exports.Connection = exports.BlobFile = exports.isBlobField = exports.blob = exports.VectorColumnOptions = exports.MakeArrowTableOptions = exports.makeJsonField = exports.makeArrowTable = exports.BranchContents = exports.TagContents = exports.Tags = exports.instrumentLanceDbMetrics = exports.NativeJsHeaderProvider = exports.MaterializedView = void 0;
5
+ exports.litBool = exports.lit = exports.col = exports.Expr = exports.packBits = exports.rerankers = exports.Scannable = exports.PermutationBuilder = exports.permutationBuilder = exports.embedding = exports.MergeInsertBuilder = exports.OAuthSession = exports.OAuthFlowType = exports.ClientAuthMethod = exports.OAuthHeaderProvider = exports.StaticHeaderProvider = exports.HeaderProvider = exports.Branches = exports.Table = exports.Index = exports.Occur = exports.Operator = exports.FullTextQueryType = exports.BooleanQuery = exports.MultiMatchQuery = exports.BoostQuery = exports.PhraseQuery = exports.MatchQuery = exports.RecordBatchIterator = exports.TakeQuery = exports.VectorQuery = exports.QueryBase = exports.Query = exports.AutoQuery = exports.Job = exports.Session = exports.Connection = exports.BlobFile = exports.isBlobField = exports.blob = exports.VectorColumnOptions = exports.MakeArrowTableOptions = exports.makeJsonField = exports.makeArrowTable = exports.BranchContents = exports.TagContents = exports.Tags = exports.instrumentLanceDbMetrics = exports.NativeJsHeaderProvider = exports.MaterializedView = void 0;
6
+ exports.Listing = exports.connectCatalog = exports.DatabaseNames = exports.Catalog = exports.func = exports.litInt = void 0;
6
7
  exports.tokenize = tokenize;
7
8
  exports.connect = connect;
8
9
  exports.connectNamespace = connectNamespace;
@@ -76,6 +77,13 @@ Object.defineProperty(exports, "Scannable", { enumerable: true, get: function ()
76
77
  exports.rerankers = require("./rerankers");
77
78
  var util_1 = require("./util");
78
79
  Object.defineProperty(exports, "packBits", { enumerable: true, get: function () { return util_1.packBits; } });
80
+ var expr_1 = require("./expr");
81
+ Object.defineProperty(exports, "Expr", { enumerable: true, get: function () { return expr_1.Expr; } });
82
+ Object.defineProperty(exports, "col", { enumerable: true, get: function () { return expr_1.col; } });
83
+ Object.defineProperty(exports, "lit", { enumerable: true, get: function () { return expr_1.lit; } });
84
+ Object.defineProperty(exports, "litBool", { enumerable: true, get: function () { return expr_1.litBool; } });
85
+ Object.defineProperty(exports, "litInt", { enumerable: true, get: function () { return expr_1.litInt; } });
86
+ Object.defineProperty(exports, "func", { enumerable: true, get: function () { return expr_1.func; } });
79
87
  /**
80
88
  * Tokenize a full-text search query using an explicit tokenizer.
81
89
  *
package/dist/indices.d.ts CHANGED
@@ -6,14 +6,29 @@ export interface IvfPqOptions {
6
6
  * The number of IVF partitions to create.
7
7
  *
8
8
  * This value should generally scale with the number of rows in the dataset.
9
- * By default the number of partitions is the square root of the number of
10
- * rows.
9
+ * When set, this takes precedence over `targetPartitionSize`. When omitted,
10
+ * the partition count is derived from `targetPartitionSize`, which defaults
11
+ * to 8192 rows for this index type.
11
12
  *
12
13
  * If this value is too large then the first part of the search (picking the
13
14
  * right partition) will be slow. If this value is too small then the second
14
15
  * part of the search (searching within a partition) will be slow.
15
16
  */
16
17
  numPartitions?: number;
18
+ /**
19
+ * The target number of rows in each IVF partition.
20
+ *
21
+ * This is an alternative to `numPartitions`. When it is set and
22
+ * `numPartitions` is not, the number of partitions is derived by dividing
23
+ * the number of rows by this value. When both are set, `numPartitions` is
24
+ * used and this value is ignored. When neither is set, the target is
25
+ * 8192 rows.
26
+ *
27
+ * Larger partitions mean fewer of them, which makes the first part of the
28
+ * search (picking the right partitions) faster and the second part
29
+ * (searching within a partition) slower.
30
+ */
31
+ targetPartitionSize?: number;
17
32
  /**
18
33
  * Number of sub-vectors of PQ.
19
34
  *
@@ -106,14 +121,29 @@ export interface IvfRqOptions {
106
121
  * The number of IVF partitions to create.
107
122
  *
108
123
  * This value should generally scale with the number of rows in the dataset.
109
- * By default the number of partitions is the square root of the number of
110
- * rows.
124
+ * When set, this takes precedence over `targetPartitionSize`. When omitted,
125
+ * the partition count is derived from `targetPartitionSize`, which defaults
126
+ * to 4096 rows for this index type.
111
127
  *
112
128
  * If this value is too large then the first part of the search (picking the
113
129
  * right partition) will be slow. If this value is too small then the second
114
130
  * part of the search (searching within a partition) will be slow.
115
131
  */
116
132
  numPartitions?: number;
133
+ /**
134
+ * The target number of rows in each IVF partition.
135
+ *
136
+ * This is an alternative to `numPartitions`. When it is set and
137
+ * `numPartitions` is not, the number of partitions is derived by dividing
138
+ * the number of rows by this value. When both are set, `numPartitions` is
139
+ * used and this value is ignored. When neither is set, the target is
140
+ * 4096 rows.
141
+ *
142
+ * Larger partitions mean fewer of them, which makes the first part of the
143
+ * search (picking the right partitions) faster and the second part
144
+ * (searching within a partition) slower.
145
+ */
146
+ targetPartitionSize?: number;
117
147
  /**
118
148
  * Number of bits per dimension for residual quantization.
119
149
  *
@@ -198,6 +228,8 @@ export interface HnswPqOptions {
198
228
  /**
199
229
  * The number of IVF partitions to create.
200
230
  *
231
+ * When set, this takes precedence over `targetPartitionSize`.
232
+ *
201
233
  * For HNSW, we recommend a small number of partitions. Setting this to 1 works
202
234
  * well for most tables. For very large tables, training just one HNSW graph
203
235
  * will require too much memory. Each partition becomes its own HNSW graph, so
@@ -205,6 +237,20 @@ export interface HnswPqOptions {
205
237
  *
206
238
  */
207
239
  numPartitions?: number;
240
+ /**
241
+ * The target number of rows in each IVF partition.
242
+ *
243
+ * This is an alternative to `numPartitions`. When it is set and
244
+ * `numPartitions` is not, the number of partitions is derived by dividing
245
+ * the number of rows by this value. When both are set, `numPartitions` is
246
+ * used and this value is ignored. When neither is set, the target is
247
+ * 1,048,576 rows.
248
+ *
249
+ * Larger partitions mean fewer of them, which makes the first part of the
250
+ * search (picking the right partitions) faster and the second part
251
+ * (searching within a partition) slower.
252
+ */
253
+ targetPartitionSize?: number;
208
254
  /**
209
255
  * Number of sub-vectors of PQ.
210
256
  *
@@ -307,6 +353,8 @@ export interface HnswSqOptions {
307
353
  /**
308
354
  * The number of IVF partitions to create.
309
355
  *
356
+ * When set, this takes precedence over `targetPartitionSize`.
357
+ *
310
358
  * For HNSW, we recommend a small number of partitions. Setting this to 1 works
311
359
  * well for most tables. For very large tables, training just one HNSW graph
312
360
  * will require too much memory. Each partition becomes its own HNSW graph, so
@@ -314,6 +362,20 @@ export interface HnswSqOptions {
314
362
  *
315
363
  */
316
364
  numPartitions?: number;
365
+ /**
366
+ * The target number of rows in each IVF partition.
367
+ *
368
+ * This is an alternative to `numPartitions`. When it is set and
369
+ * `numPartitions` is not, the number of partitions is derived by dividing
370
+ * the number of rows by this value. When both are set, `numPartitions` is
371
+ * used and this value is ignored. When neither is set, the target is
372
+ * 1,048,576 rows.
373
+ *
374
+ * Larger partitions mean fewer of them, which makes the first part of the
375
+ * search (picking the right partitions) faster and the second part
376
+ * (searching within a partition) slower.
377
+ */
378
+ targetPartitionSize?: number;
317
379
  /**
318
380
  * Max iterations to train kmeans.
319
381
  *
@@ -378,14 +440,29 @@ export interface IvfFlatOptions {
378
440
  * The number of IVF partitions to create.
379
441
  *
380
442
  * This value should generally scale with the number of rows in the dataset.
381
- * By default the number of partitions is the square root of the number of
382
- * rows.
443
+ * When set, this takes precedence over `targetPartitionSize`. When omitted,
444
+ * the partition count is derived from `targetPartitionSize`, which defaults
445
+ * to 4096 rows for this index type.
383
446
  *
384
447
  * If this value is too large then the first part of the search (picking the
385
448
  * right partition) will be slow. If this value is too small then the second
386
449
  * part of the search (searching within a partition) will be slow.
387
450
  */
388
451
  numPartitions?: number;
452
+ /**
453
+ * The target number of rows in each IVF partition.
454
+ *
455
+ * This is an alternative to `numPartitions`. When it is set and
456
+ * `numPartitions` is not, the number of partitions is derived by dividing
457
+ * the number of rows by this value. When both are set, `numPartitions` is
458
+ * used and this value is ignored. When neither is set, the target is
459
+ * 4096 rows.
460
+ *
461
+ * Larger partitions mean fewer of them, which makes the first part of the
462
+ * search (picking the right partitions) faster and the second part
463
+ * (searching within a partition) slower.
464
+ */
465
+ targetPartitionSize?: number;
389
466
  /**
390
467
  * Distance type to use to build the index.
391
468
  *
@@ -633,7 +710,7 @@ export declare class Index {
633
710
  * Create a label list index.
634
711
  *
635
712
  * LabelList index is a scalar index that can be used on `List<T>` columns to
636
- * support queries with `array_contains_all` and `array_contains_any`
713
+ * support queries with `array_has_all` and `array_has_any`
637
714
  * using an underlying bitmap index.
638
715
  */
639
716
  static labelList(): Index;
package/dist/indices.js CHANGED
@@ -32,7 +32,7 @@ class Index {
32
32
  * currently is also a memory intensive operation.
33
33
  */
34
34
  static ivfPq(options) {
35
- return new Index(native_1.Index.ivfPq(options?.distanceType, options?.numPartitions, options?.numSubVectors, options?.numBits, options?.maxIterations, options?.sampleRate));
35
+ return new Index(native_1.Index.ivfPq(options?.distanceType, options?.numPartitions, options?.numSubVectors, options?.numBits, options?.maxIterations, options?.sampleRate, options?.targetPartitionSize));
36
36
  }
37
37
  /**
38
38
  * Create an IvfRq index
@@ -51,7 +51,7 @@ class Index {
51
51
  * currently is also a memory intensive operation.
52
52
  */
53
53
  static ivfRq(options) {
54
- return new Index(native_1.Index.ivfRq(options?.distanceType, options?.numPartitions, options?.numBits, options?.maxIterations, options?.sampleRate));
54
+ return new Index(native_1.Index.ivfRq(options?.distanceType, options?.numPartitions, options?.numBits, options?.maxIterations, options?.sampleRate, options?.targetPartitionSize));
55
55
  }
56
56
  /**
57
57
  * Create an IvfFlat index
@@ -70,7 +70,7 @@ class Index {
70
70
  * currently is also a memory intensive operation.
71
71
  */
72
72
  static ivfFlat(options) {
73
- return new Index(native_1.Index.ivfFlat(options?.distanceType, options?.numPartitions, options?.maxIterations, options?.sampleRate));
73
+ return new Index(native_1.Index.ivfFlat(options?.distanceType, options?.numPartitions, options?.maxIterations, options?.sampleRate, options?.targetPartitionSize));
74
74
  }
75
75
  /**
76
76
  * Create a btree index
@@ -109,7 +109,7 @@ class Index {
109
109
  * Create a label list index.
110
110
  *
111
111
  * LabelList index is a scalar index that can be used on `List<T>` columns to
112
- * support queries with `array_contains_all` and `array_contains_any`
112
+ * support queries with `array_has_all` and `array_has_any`
113
113
  * using an underlying bitmap index.
114
114
  */
115
115
  static labelList() {
@@ -148,7 +148,7 @@ class Index {
148
148
  *
149
149
  */
150
150
  static hnswPq(options) {
151
- return new Index(native_1.Index.hnswPq(options?.distanceType, options?.numPartitions, options?.numSubVectors, options?.maxIterations, options?.sampleRate, options?.m, options?.efConstruction));
151
+ return new Index(native_1.Index.hnswPq(options?.distanceType, options?.numPartitions, options?.numSubVectors, options?.maxIterations, options?.sampleRate, options?.m, options?.efConstruction, options?.targetPartitionSize));
152
152
  }
153
153
  /**
154
154
  *
@@ -160,7 +160,7 @@ class Index {
160
160
  *
161
161
  */
162
162
  static hnswSq(options) {
163
- return new Index(native_1.Index.hnswSq(options?.distanceType, options?.numPartitions, options?.maxIterations, options?.sampleRate, options?.m, options?.efConstruction));
163
+ return new Index(native_1.Index.hnswSq(options?.distanceType, options?.numPartitions, options?.maxIterations, options?.sampleRate, options?.m, options?.efConstruction, options?.targetPartitionSize));
164
164
  }
165
165
  }
166
166
  exports.Index = Index;