summaryrefslogtreecommitdiffstats
path: root/searchlib
Commit message (Collapse)AuthorAgeFilesLines
* Correct copyright headersJon Bratseth2018-02-012-2/+4
|
* Canonicalize featuresJon Bratseth2018-01-316-13/+138
| | | | | This allows us to find the type of features referenced in ranking expressions regardless of the form they are written in.
* Nonfunctional changes onlyJon Bratseth2018-01-311-0/+1
|
* Verify macrosJon Bratseth2018-01-3132-135/+162
|
* Propagate type information through ranking expressionsJon Bratseth2018-01-2919-37/+141
|
* Propagate type informationJon Bratseth2018-01-2911-33/+117
|
* Don't use uint16_t when UChar is needed.Tor Egge2018-01-273-2/+3
| | | | Include needed unicode header file.
* Merge pull request #4789 from ↵Henning Baldersheim2018-01-2611-69/+138
|\ | | | | | | | | vespa-engine/geirst/improve-alloc-strategy-in-data-store Geirst/improve alloc strategy in data store
| * Use definition of huge page size from memory allocator.Geir Storli2018-01-261-2/+1
| |
| * Improve buffer allocation strategy in data store by matching underlying ↵Geir Storli2018-01-267-42/+95
| | | | | | | | | | | | | | | | allocators. This should reduce the amount of memory wasted in allocations. 1) heap allocation: buffer size is power of 2 to match vespamalloc. 2) mmap allocation: buffer size is multiple of huge page size (2MB) to match mmap allocator.
| * Refactor calculation of buffer allocation.Geir Storli2018-01-254-27/+44
| |
* | Refactor tensorflow import and add dropout testLester Solbakken2018-01-2611-342/+3602
| |
* | Correct copyright headerJon Bratseth2018-01-251-1/+0
| |
* | Add copyright headersJon Bratseth2018-01-254-0/+4
| |
* | Remove loggingJon Bratseth2018-01-252-1/+5
| |
* | Refactor: RenameJon Bratseth2018-01-253-25/+24
| |
* | Refactor: Move state to helperJon Bratseth2018-01-253-32/+29
| |
* | Refactor: Extract test helper logicJon Bratseth2018-01-254-125/+166
| |
* | Minor improvementsJon Bratseth2018-01-251-12/+11
|/
* Keep imported tensor attribute vector read guard live during access.Tor Egge2018-01-231-1/+2
|
* Revert "Revert "Add imported tensor attribute vector""Tor Egge2018-01-2312-2/+415
|
* Revert "Add imported tensor attribute vector"Håkon Hallingstad2018-01-2312-415/+2
|
* Merge pull request #4733 from ↵Tor Brede Vekterli2018-01-2312-2/+415
|\ | | | | | | | | vespa-engine/toregge/add-imported-tensor-attribute-vector Add imported tensor attribute vector
| * Use ASSERT_TRUE instead of EXPECT_TRUE.Tor Egge2018-01-231-1/+1
| | | | | | | | Continuing on failure would lead to test program crashing.
| * Fix typo in tensor spec.Tor Egge2018-01-221-1/+1
| |
| * Add class comment.Tor Egge2018-01-221-0/+4
| |
| * Add imported tensor attribute vector read guard.Tor Egge2018-01-229-1/+217
| | | | | | | | | | Extend unit tests for imported attribute vector to check imported tensor attribute.
| * Add imported tensor attribute vector.Tor Egge2018-01-227-1/+194
| |
* | Merge branch 'master' into lesters/add-batch-norm-tf-caseLester Solbakken2018-01-2223-105/+267
|\ \
| * | Use double quotesJon Bratseth2018-01-222-4/+4
| | | | | | | | | | | | Single quotes are not supported in the backend.
| * | Merge pull request #4734 from ↵Henning Baldersheim2018-01-2221-101/+262
| |\ \ | | |/ | |/| | | | | | | vespa-engine/geirst/more-conservative-buffer-allocation-strategy-in-datastore-buffers Geirst/more conservative buffer allocation strategy in datastore buffers
| | * Add config for the alloc grow factor used when allocating buffers in ↵Geir Storli2018-01-223-9/+8
| | | | | | | | | | | | multi-value attributes.
| | * Add more conservative buffer allocation strategy in datastore buffers.Geir Storli2018-01-1919-78/+243
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | This should greatly reduce the memory footprint of multi-value attribute vectors which uses an array store for underlying values. The size of a new buffer is calculated as (simplified): *) before: (size of previous buffer) * 2 *) now: (size of all previous buffers) * allocation grow factor (default = 0.2) In addition we now support resizing of all buffers, not only the first one.
| | * Style changes.Geir Storli2018-01-182-16/+13
| | |
* | | Support negative values dimension sizes in tensorflow reshape importLester Solbakken2018-01-221-2/+4
| | |
* | | Support negative dimesions in import of expanddimsLester Solbakken2018-01-221-3/+7
| | |
* | | Fix string representations of tensor expressionsLester Solbakken2018-01-221-1/+1
| | |
* | | Add batch normalization test caseLester Solbakken2018-01-2212-8699/+33116
|/ /
* | Parse generated tensor function treesJon Bratseth2018-01-228-130/+79
| | | | | | | | | | | | | | | | | | To make generated tensor function trees transparent to the config model we need to convert each tensor function node to the corresponding ranking expression node. This is most easily done by parsing the tensor function tree string output as a ranking expression (something which is required to always work in any case).
* | Add read interface for tensor attribute, used by tensor attribute featureTor Egge2018-01-1811-27/+66
| | | | | | | | executors.
* | Use protobuf 3.4.0 by defaultJon Bratseth2018-01-181-1/+0
|/
* Merge pull request #4692 from ↵Jon Bratseth2018-01-172-1/+8
|\ | | | | | | | | vespa-engine/bratseth/store-converted-expressions-in-zk Bratseth/store converted expressions in zk
| * Access files through application packageJon Bratseth2018-01-161-1/+6
| |
| * Write generated expressionsJon Bratseth2018-01-151-0/+2
| |
* | Merge pull request #4681 from ↵Geir Storli2018-01-163-113/+66
|\ \ | | | | | | | | | | | | vespa-engine/geirst/minor-cleanup-in-searchlib-datastore Geirst/minor cleanup in searchlib datastore
| * | Rewrite to new syntax for test function setup.Geir Storli2018-01-161-74/+23
| | |
| * | Use FreeListAllocator in free list unit test and remove unused function.Geir Storli2018-01-163-39/+43
| |/
* / move more param stuff to lazy_params.hHåvard Pettersen2018-01-162-2/+3
|/
* Revert "Revert "Revert "Balder/group multiple commits rebased 1"""Henning Baldersheim2018-01-1419-1180/+621
|
* Revert accidental commit.Henning Baldersheim2018-01-121-0/+1
|