Commit message (Collapse) | Author | Age | Files | Lines | |
---|---|---|---|---|---|
* | remove parameter count from interpreted functions | Håvard Pettersen | 2020-04-03 | 1 | -1/+1 |
| | |||||
* | count reachables for statistics | Arne Juul | 2020-04-01 | 2 | -3/+38 |
| | |||||
* | use snake_case | Arne Juul | 2020-04-01 | 2 | -7/+7 |
| | |||||
* | add some simple statistics when exploring state | Arne Juul | 2020-04-01 | 3 | -0/+28 |
| | |||||
* | Merge pull request #12779 from ↵ | Geir Storli | 2020-03-31 | 9 | -77/+217 |
|\ | | | | | | | | | vespa-engine/geirst/nearest-neighbor-index-saving-and-loading Nearest neighbor index saving and loading | ||||
| * | Test that hnsw index is integrated in dense tensor attribute. | Geir Storli | 2020-03-31 | 1 | -11/+49 |
| | | |||||
| * | Implement saving and loading of nearest neighbor index. | Geir Storli | 2020-03-31 | 9 | -66/+168 |
| | | |||||
* | | - Add testing of enumerated weighted sets. | Henning Baldersheim | 2020-03-31 | 2 | -14/+19 |
|/ | | | | | - Add testing that query can use floating point numbers for the weight. - Stay with the type feature_t for query. Do not assume th ethe weight type used in the attribute. | ||||
* | Merge pull request #12765 from vespa-engine/geirst/attribute-loading-cleanup | Geir Storli | 2020-03-30 | 16 | -162/+141 |
|\ | | | | | Attribute loading cleanup | ||||
| * | Use const reference to attribute vector. | Geir Storli | 2020-03-30 | 2 | -19/+19 |
| | | |||||
| * | Remove opening of udat file which is never used. | Geir Storli | 2020-03-30 | 4 | -41/+10 |
| | | | | | | | | All loading of udat file is using LoadUtils::loadUDAT() instead. | ||||
| * | Move functions used to open / load attribute vector data files to separate ↵ | Geir Storli | 2020-03-30 | 15 | -129/+139 |
| | | | | | | | | class. | ||||
* | | Merge pull request #12753 from vespa-engine/toregge/handle-newer-gtest-version | Tor Egge | 2020-03-30 | 3 | -5/+5 |
|\ \ | |/ |/| | Handle newer gtest versions where the legacy API is deprecated. | ||||
| * | Reduce code duplication in test code. | Tor Egge | 2020-03-30 | 3 | -25/+5 |
| | | |||||
| * | Handle newer gtest versions where the legacy API is deprecated. | Tor Egge | 2020-03-29 | 3 | -0/+20 |
| | | |||||
* | | Merge pull request #12733 from vespa-engine/arnej/add-hnsw-index-save-and-load | Geir Storli | 2020-03-30 | 16 | -154/+553 |
|\ \ | |/ |/| | Add save and load of HNSW index | ||||
| * | review follow-up: | Arne Juul | 2020-03-27 | 9 | -38/+55 |
| | | | | | | | | | | | | * add documentation comments * style fixes * return bool from load() for error handling | ||||
| * | Add save and load of HNSW index | Arne Juul | 2020-03-26 | 14 | -150/+532 |
| | | | | | | | | | | | | | | | | * split graph data out into a simpler struct * implement HnswIndexSaver * implement HnswIndexLoader * add unit test checking that saving and loading works as expected | ||||
* | | Extend attribute save targets to support custom file writers. | Geir Storli | 2020-03-27 | 11 | -30/+304 |
| | | | | | | | | This is a preparation for saving and loading nearest neighbor index in the tensor attribute code. | ||||
* | | adaptive sequenced executor (experimental) | Håvard Pettersen | 2020-03-26 | 6 | -18/+768 |
|/ | |||||
* | Merge pull request #12713 from ↵ | Arne H Juul | 2020-03-25 | 5 | -4/+67 |
|\ | | | | | | | | | vespa-engine/geirst/nearest-neighbor-index-saver-skeleton Add skeleton for nearest neighbor index saver. | ||||
| * | Add skeleton for nearest neighbor index saver. | Geir Storli | 2020-03-25 | 5 | -4/+67 |
| | | |||||
* | | Unify the metrics for queuesize similar to what we have for the spi queues. | Henning Baldersheim | 2020-03-25 | 1 | -1/+1 |
|/ | |||||
* | Merge pull request #12691 from ↵ | Geir Storli | 2020-03-24 | 14 | -11/+114 |
|\ | | | | | | | | | vespa-engine/geirst/tensor-attribute-in-state-explorer Expose information from tensor attribute and hnsw index in state expl… | ||||
| * | Expose information from tensor attribute and hnsw index in state explorer ↵ | Geir Storli | 2020-03-24 | 14 | -11/+114 |
| | | | | | | | | for attribute. | ||||
* | | Merge pull request #12690 from vespa-engine/havardpe/rank-setup-stack-usage | Håvard Pettersen | 2020-03-24 | 1 | -9/+35 |
|\ \ | | | | | | | rank setup stack usage | ||||
| * | | rank setup stack usage | Håvard Pettersen | 2020-03-24 | 1 | -9/+35 |
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | With large models comes large stack usage. - compile in separate thread for deterministic stack size (verify rank setup vs. proton config setup) - warn about stack usage when over 128K (note that recursive decent parsing of rank expressions comes on top of the usage reported by the blueprint resolver) | ||||
* | | | Merge pull request #12671 from vespa-engine/arnej/configurable-distance-metrics | Arne H Juul | 2020-03-24 | 22 | -87/+534 |
|\ \ \ | |/ / |/| | | Arnej/configurable distance metrics | ||||
| * | | update copyright headers | Arne Juul | 2020-03-24 | 2 | -2/+2 |
| | | | |||||
| * | | extend unit test | Arne Juul | 2020-03-24 | 1 | -0/+4 |
| | | | |||||
| * | | add documentation comment | Arne Juul | 2020-03-24 | 1 | -0/+4 |
| | | | |||||
| * | | constify DistanceFunction | Arne Juul | 2020-03-24 | 9 | -13/+13 |
| | | | |||||
| * | | add reference and refactor | Arne Juul | 2020-03-24 | 1 | -7/+10 |
| | | | |||||
| * | | remove debug printing | Arne Juul | 2020-03-24 | 1 | -2/+0 |
| | | | |||||
| * | | unit test distance functions | Arne Juul | 2020-03-24 | 4 | -1/+208 |
| | | | |||||
| * | | output distance in meters | Arne Juul | 2020-03-24 | 1 | -7/+12 |
| | | | |||||
| * | | simplify iterator | Arne Juul | 2020-03-23 | 1 | -27/+8 |
| | | | | | | | | | | | | | | | * now distance function handles cell type, so iterator does not need to be templated on that. | ||||
| * | | track API changes | Arne Juul | 2020-03-23 | 1 | -7/+18 |
| | | | | | | | | | | | | * also, NNS iterators no longer handles different cell types | ||||
| * | | check that query tensor is converted correctly | Arne Juul | 2020-03-23 | 1 | -6/+10 |
| | | | |||||
| * | | use distance function from index if available | Arne Juul | 2020-03-23 | 6 | -34/+87 |
| | | | | | | | | | | | | | | | | | | | | | * convert query tensor to same cell type as attribute * use DistanceFunction to calculate abstract distances for NNS * use DistanceFunction to convert abstract distances to rawscore * if no index is available, use a fallback DistanceFunction | ||||
| * | | use external factory for distance function | Arne Juul | 2020-03-23 | 1 | -11/+5 |
| | | | |||||
| * | | track API changes | Arne Juul | 2020-03-23 | 1 | -1/+4 |
| | | | |||||
| * | | extend API with distance_function access | Arne Juul | 2020-03-23 | 2 | -0/+3 |
| | | | |||||
| * | | add distance metric to HnswIndexParams | Arne Juul | 2020-03-23 | 1 | -1/+15 |
| | | | |||||
| * | | split out DistanceFunction factory | Arne Juul | 2020-03-23 | 3 | -0/+57 |
| | | | |||||
| * | | more distance functions | Arne Juul | 2020-03-23 | 2 | -0/+106 |
| |/ | | | | | | | | | | | * extend DistanceFunction API * add Angular metric * add GeoDegrees metric | ||||
* / | Avoid expensive systemcall. We already track the writeposition in byteSize. | Henning Baldersheim | 2020-03-23 | 1 | -2/+4 |
|/ | |||||
* | Include limits header file when needed. | Tor Egge | 2020-03-21 | 2 | -0/+2 |
| | |||||
* | Revert "Revert "Since the most important aspect our use of hash maps/sets is ↵ | Henning Baldersheim | 2020-03-20 | 1 | -2/+2 |
| | | | | for spee…"" | ||||
* | Revert "Since the most important aspect our use of hash maps/sets is for ↵ | Harald Musum | 2020-03-20 | 1 | -2/+2 |
| | | | | spee…" |