arize-phoenix 4.4.4rc5__py3-none-any.whl → 4.4.4rc6__py3-none-any.whl

{arize_phoenix-4.4.4rc5.dist-info → arize_phoenix-4.4.4rc6.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: arize-phoenix
-Version: 4.4.4rc5
+Version: 4.4.4rc6
 Summary: AI Observability and Evaluation
 Project-URL: Documentation, https://docs.arize.com/phoenix/
 Project-URL: Issues, https://github.com/Arize-ai/phoenix/issues
@@ -20,7 +20,7 @@ Requires-Python: <3.13,>=3.8
 Requires-Dist: aioitertools
 Requires-Dist: aiosqlite
 Requires-Dist: alembic<2,>=1.3.0
-Requires-Dist: arize-phoenix-evals>=0.3.0
+Requires-Dist: arize-phoenix-evals>=0.13.1
 Requires-Dist: cachetools
 Requires-Dist: grpcio
 Requires-Dist: hdbscan>=0.8.33
@@ -36,7 +36,7 @@ Requires-Dist: opentelemetry-exporter-otlp
 Requires-Dist: opentelemetry-proto>=1.12.0
 Requires-Dist: opentelemetry-sdk
 Requires-Dist: opentelemetry-semantic-conventions
-Requires-Dist: pandas
+Requires-Dist: pandas>=1.0
 Requires-Dist: protobuf<6.0,>=3.20
 Requires-Dist: psutil
 Requires-Dist: pyarrow
@@ -79,6 +79,7 @@ Requires-Dist: llama-index>=0.10.3; extra == 'dev'
 Requires-Dist: nbqa; extra == 'dev'
 Requires-Dist: pandas-stubs==2.0.3.230814; (python_version < '3.9') and extra == 'dev'
 Requires-Dist: pandas-stubs==2.2.2.240603; (python_version >= '3.9') and extra == 'dev'
+Requires-Dist: pandas>=1.0; extra == 'dev'
 Requires-Dist: pre-commit; extra == 'dev'
 Requires-Dist: prometheus-client; extra == 'dev'
 Requires-Dist: psycopg[binary]; extra == 'dev'
@@ -88,10 +89,15 @@ Requires-Dist: pytest-postgresql; extra == 'dev'
 Requires-Dist: pytest==8.2.2; extra == 'dev'
 Requires-Dist: ruff==0.4.9; extra == 'dev'
 Requires-Dist: strawberry-graphql[debug-server,opentelemetry]==0.235.0; extra == 'dev'
+Requires-Dist: tabulate; extra == 'dev'
+Requires-Dist: types-tabulate; extra == 'dev'
 Provides-Extra: evals
 Provides-Extra: experimental
 Provides-Extra: llama-index
-Requires-Dist: llama-index==0.10.44; extra == 'llama-index'
+Requires-Dist: llama-index-embeddings-openai; extra == 'llama-index'
+Requires-Dist: llama-index-llms-openai; extra == 'llama-index'
+Requires-Dist: llama-index-readers-file; extra == 'llama-index'
+Requires-Dist: llama-index==0.10.51; extra == 'llama-index'
 Provides-Extra: pg
 Requires-Dist: asyncpg; extra == 'pg'
 Description-Content-Type: text/markdown
@@ -127,7 +133,7 @@ Description-Content-Type: text/markdown
 
 Phoenix is an open-source AI observability platform designed for experimentation, evaluation, and troubleshooting. It provides:
 
--   **_Tracing_** - Trace your LLM application's runtime using using OpenTelemetry-based instrumentation.
+-   **_Tracing_** - Trace your LLM application's runtime using OpenTelemetry-based instrumentation.
 -   **_Evaluation_** - Leverage LLMs to benchmark your application's performance using response and retrieval evals.
 -   **_Inference Analysis_** - Visualize inferences and embeddings using dimensionality reduction and clustering to identify drift and performance degradation.
 

{arize_phoenix-4.4.4rc5.dist-info → arize_phoenix-4.4.4rc6.dist-info}/RECORD

@@ -1,24 +1,16 @@
 phoenix/__init__.py,sha256=JMXBf8J0tAa5ycWDIn8QWYgQu6H2vL77RC4lful-YH8,1542
-phoenix/config.py,sha256=9xkQBn_Z-tsQct-zq0B80N4Xa2k2jfhivdc_qYCA5G8,7618
-phoenix/datetime_utils.py,sha256=oqkxJ5I7ggrCKYTEi8q-akC501calylD26NVOPQQcHw,3305
+phoenix/config.py,sha256=eXciIho_PDh4ZSmq4Gtuo7Qz__yTluDP3_WUwig5OiU,8141
+phoenix/datetime_utils.py,sha256=yDKjwX2Vtqw9h5F_ProtP-TsXidM43uIvmJ_pOzYc9A,3405
 phoenix/exceptions.py,sha256=n2L2KKuecrdflB9MsCdAYCiSEvGJptIsfRkXMoJle7A,169
 phoenix/py.typed,sha256=AbpHGcgLb-kRsJGnwFEktk7uzpZOCcBY74-YBdrKVGs,1
 phoenix/services.py,sha256=aTxhcOA1pZHB6U-B3TEcp6fqDF5oT0xCUvEUNMZVTUQ,5175
 phoenix/settings.py,sha256=cO-qgis_S27nHirTobYI9hHPfZH18R--WMmxNdsVUwc,273
-phoenix/version.py,sha256=-Vg_bLotyeJdv0gFqG5-A64nsG-6AR0xZSp3sDDsV_w,25
+phoenix/version.py,sha256=rZ0Z9PgUs79kMn4HpCH3vAEVOqqPCzzD7Xz8N5sa7qI,25
 phoenix/core/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 phoenix/core/embedding_dimension.py,sha256=zKGbcvwOXgLf-yrJBpQyKtd-LEOPRKHnUToyAU8Owis,87
 phoenix/core/model.py,sha256=km_a--PBHOuA337ClRw9xqhOHhrUT6Rl9pz_zV0JYkQ,4843
 phoenix/core/model_schema.py,sha256=F2dbbVnkDLsPYoyZDv1q03uhvP8LcU1wXp0g-exiWs0,50551
 phoenix/core/model_schema_adapter.py,sha256=0Tm_Y_gV-WED8fKBCaFXAEFwE3CTEZS1dowqnTZ7x7g,8426
-phoenix/datasets/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-phoenix/datasets/experiments.py,sha256=RzZezHQcTpPcr7gY9rGtoYlfoesFNhNV7EO5f_oHNFk,21198
-phoenix/datasets/tracing.py,sha256=wVpt8Ie9WNPoi1djJdcrkwCokHdTO0bicXViLg3O-1Y,2831
-phoenix/datasets/types.py,sha256=N17mnnVwmu1k3bnmbyROPt_6TxPaZY_QkOZmCOR5_jE,4835
-phoenix/datasets/evaluators/__init__.py,sha256=KSr9fNG4O93swYxNdPj_UihP9Itl_5mj0a492wi_4_0,465
-phoenix/datasets/evaluators/code_evaluators.py,sha256=DdCcAi274t_TLs_aARd-GmWWpJrxVeNEAegMFEAfe0E,3894
-phoenix/datasets/evaluators/llm_evaluators.py,sha256=aVfAHOWhskBiy0IVeq_ACTs7B37uXTTtDoNBS0XenIc,9165
-phoenix/datasets/evaluators/utils.py,sha256=S7OGrb1sBWg5l9K35X29OKJe5wZ3k7xMhxJBclzxta0,10452
 phoenix/db/README.md,sha256=IvKaZyf9ECbGBYYePaRhBveKZwDbxAc-c7BMxJYZh6Q,595
 phoenix/db/__init__.py,sha256=pDjEFXukHmJBM-1D8RjmXkvLsz85YWNxMQczt81ec3A,118
 phoenix/db/alembic.ini,sha256=p8DjVqGUs_tTx8oU56JP7qj-rMUebNFizItUSv_hPhs,3763
@@ -26,18 +18,28 @@ phoenix/db/bulk_inserter.py,sha256=zbZGWZFDybKaGLGzpxgLwxAS5sC0_wXcvM0be4kUhh8,1
 phoenix/db/engines.py,sha256=vLWaZlToMtDI7rJDxSidYkfOoojamxaZxaz8ND3zTus,4770
 phoenix/db/helpers.py,sha256=L2_jP1iIWpUREhKLYYb4_vf_6v_BiU1E73Z2PczGm6s,1589
 phoenix/db/migrate.py,sha256=MuhtNWnR24riROvarvKfbRb4_D5xuQi6P760vBUKl1E,2270
-phoenix/db/models.py,sha256=lYzI3tCDUl8njXb3Vf3R8e6y56-MErprjjfBE-o9Kao,20419
+phoenix/db/models.py,sha256=zFtdhVuQFOvquyKsto62aqAVaTRUlq9gxU0j1M1yLdg,20408
 phoenix/db/insertion/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 phoenix/db/insertion/dataset.py,sha256=_vxy5e6W5jEuvO2fMKbbNCn9JvHkwI4LRKk_10eKFVg,7171
-phoenix/db/insertion/evaluation.py,sha256=fAerUy3QGf2wID_tiVmPvzxBDFGiONPl3pmpZDgJDWQ,7183
-phoenix/db/insertion/helpers.py,sha256=7tf6qQyJ05nn3IXaZEpj2b4Jz5boGLWT8tzlMaJ9tQY,2337
-phoenix/db/insertion/span.py,sha256=DNBjSrx5g2W5KuTB1dkHwtkb0SFnMIxN1jB-BAdGKFY,5634
+phoenix/db/insertion/evaluation.py,sha256=HoUncZN9ZlIr1QO0uA37SbWhrjmwQVYVJlgFX2VefY8,7211
+phoenix/db/insertion/helpers.py,sha256=5AZQSyTGAthyaIl_l5jL4yva1IrTTBG9y2G7l1r2Yyk,1937
+phoenix/db/insertion/span.py,sha256=d85O3R_Cc3aVFDJSgLLX66qNCPNbKtDxug_dSKFDfew,5655
 phoenix/db/migrations/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 phoenix/db/migrations/env.py,sha256=QbzB5zrRs6XQQmrYeUpuzeilcMlM-MsbaAgHHYcIHTI,3626
 phoenix/db/migrations/script.py.mako,sha256=MEqL-2qATlST9TAOeYgscMn1uy6HUS9NFvDgl93dMj8,635
 phoenix/db/migrations/types.py,sha256=Frq1AKSyBKQQ0FLzON-EmgTqE4kNkOpHMsbWnI-WgCE,605
 phoenix/db/migrations/versions/10460e46d750_datasets.py,sha256=l69yZfScFrjfZZpY0gnqwhsDUEctLeo02qMgA_aOGDg,8155
 phoenix/db/migrations/versions/cf03bd6bae1d_init.py,sha256=CbWT3ZTR0CZqeT3zWLoTWhboFmnOy3Ju1z6Ztpq8WIM,8122
+phoenix/experiments/__init__.py,sha256=6JGwgUd7xCbGpuHqYZlsmErmYvVgv7N_j43bn3dUqsk,123
+phoenix/experiments/functions.py,sha256=w0A6BK80avoupxd3sPJZ_btftV1pXrkbZj4omR_H214,24723
+phoenix/experiments/tracing.py,sha256=wVpt8Ie9WNPoi1djJdcrkwCokHdTO0bicXViLg3O-1Y,2831
+phoenix/experiments/types.py,sha256=tj7DxfsU_nQP5bNe_h6p4KvRjkXKaaB3FeaIerAi_iA,22790
+phoenix/experiments/utils.py,sha256=ZZajvIrZTURhOX5Nx4nyogJEbI18sKCHYiYwOxz2vYU,340
+phoenix/experiments/evaluators/__init__.py,sha256=j63fi3fa3U7-itVPHa82GowhjQRU-wO6yhO34u_lhsA,714
+phoenix/experiments/evaluators/base.py,sha256=uhO4R06YWBbTxdpvXLldANnTxTA5r2h_Ktj-ZMLH57c,5305
+phoenix/experiments/evaluators/code_evaluators.py,sha256=0qIKQS14Knze50ziJEPVEnNeV3QIs4g1IXtCmaWZu7o,3923
+phoenix/experiments/evaluators/llm_evaluators.py,sha256=EFce6LKZwUZDBa5ZozvcdqeZpdWM6n6bmq7_oIzM2Nw,9211
+phoenix/experiments/evaluators/utils.py,sha256=o84UTWN7fzjCGZDTS-KpGZ2VBrk2iSuO3X2LoC7pr3Y,6966
 phoenix/inferences/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 phoenix/inferences/errors.py,sha256=cGp9vxnw4SewFoWBV3ZGMkhE0Kh73lPIv3Ppz_H_RoA,8261
 phoenix/inferences/fixtures.py,sha256=FC2eRL4dpobKQHYOilFtDexUWFkMZ_w6jun_4WkbMk0,20792
@@ -58,18 +60,19 @@ phoenix/pointcloud/pointcloud.py,sha256=4zAIkKs2xOUbchpj4XDAV-iPMXrfAJ15TG6rlIYG
 phoenix/pointcloud/projectors.py,sha256=zO_RrtDYSv2rqVOfIP2_9Cv11Dc8EmcZR94xhFcBYPU,1057
 phoenix/pointcloud/umap_parameters.py,sha256=lJsEOrbSuSiqI7g4Yt6xj7kgYxEqoep4ZHWLr6VWBqw,1760
 phoenix/server/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-phoenix/server/app.py,sha256=LQrHWt5HG_pWqnR9Ozb3-vnAGiiRGuZ3uV_9-886Yxw,18340
+phoenix/server/app.py,sha256=Ld9NvW7sBT4aJn8CXAB_PIw4c5wlkxCcjT_hzjtn7dM,18478
 phoenix/server/grpc_server.py,sha256=faktLxEtWGlCB1bPR4QwwTsRoQloahKMx0hAWqRGI5s,3379
 phoenix/server/main.py,sha256=mtzH_2Kyvuy3AHiiKfqiCdUQ6SGFzeT4q9fefbV6GLg,11114
 phoenix/server/prometheus.py,sha256=j9DHB2fERuq_ZKmwVaqR-9wx5WcPPuU1Cm5Bhg5241Y,2996
 phoenix/server/telemetry.py,sha256=T_2OKrxNViAeaANlNspEekg_Y5uZIFWvKAnpz8Aoqvk,2762
 phoenix/server/thread_server.py,sha256=dP6cm6Cf08jNhDA1TRlVZpziu1YgtPDmaeIJMm725eI,2154
 phoenix/server/api/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-phoenix/server/api/context.py,sha256=jb69SVdb5hpVbfM4U0pZi4sGa2a-0VKOJWcBjjS7l4s,2761
+phoenix/server/api/context.py,sha256=4jcy203Gtx38399FP21iU3HmFsq-50EKFJlX4IW2Los,2878
 phoenix/server/api/interceptor.py,sha256=ykDnoC_apUd-llVli3m1CW18kNSIgjz2qZ6m5JmPDu8,1294
 phoenix/server/api/queries.py,sha256=wp5BlapuxDIoaQJm7mzG0dURfVxR32vXSJVC0JqG4_Y,19845
 phoenix/server/api/schema.py,sha256=BcxdqO5CSGqpKd-AAJHMjFlzaK9oJA8GJuxmMfcdjn4,434
-phoenix/server/api/dataloaders/__init__.py,sha256=urbG3M-k2cpj2ymMLYQ28tzIXAG1edECxM-tJ22ylqE,4720
+phoenix/server/api/dataloaders/__init__.py,sha256=qehXL37vGdw7v5PFs3kbZVIuhuzrVNVeZACDQjYpwyo,4847
+phoenix/server/api/dataloaders/average_experiment_run_latency.py,sha256=RiO0AKC6Y5byafsV0zTJEIOt8Nudjte73f1T78cBe1k,1817
 phoenix/server/api/dataloaders/dataset_example_revisions.py,sha256=Vpr5IEKSR4QnAVxE5NM7u92fPNgeHQV2ieYc6JakCj0,3788
 phoenix/server/api/dataloaders/dataset_example_spans.py,sha256=_jLlo0KdUS65d4PNTtE9aXVyG_NZWgA7VcpNC9udQ8U,1484
 phoenix/server/api/dataloaders/document_evaluation_summaries.py,sha256=dgAAlD0n8X6oAPLaD-czoefNkDqP338MouWsKaW8bOY,5684
@@ -127,14 +130,14 @@ phoenix/server/api/openapi/main.py,sha256=WY0pj3B7siQyyYqKyhqnzWC7P8MtEtiukOBUjG
 phoenix/server/api/openapi/schema.py,sha256=uuSYe1Ecu72aXRgTNjyMu-9ZPE13DAHJPKtedS-MsSs,451
 phoenix/server/api/routers/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 phoenix/server/api/routers/utils.py,sha256=M41BoH-fl37izhRuN2aX7lWm7jOC20A_3uClv9TVUUY,583
-phoenix/server/api/routers/v1/__init__.py,sha256=B5eSaylPI7MoYia1-VgKrU8rDi-69r_hRwPU5yMLUTE,2808
-phoenix/server/api/routers/v1/dataset_examples.py,sha256=wtplRUv2ee9xGTrcEMgTn-7L4NX_73IcwUXkCMZEFc4,6726
-phoenix/server/api/routers/v1/datasets.py,sha256=ws2Guou9mspwFx3-cBFZoD2VuTwWGoFZmtt2Sr3zg6k,31516
-phoenix/server/api/routers/v1/evaluations.py,sha256=rwSVg-rpujhsMcDVFt-VAr0Ix9TgvLcY_bSxeh8PzJI,9241
-phoenix/server/api/routers/v1/experiment_evaluations.py,sha256=xhrkPUc_4ncIBm24aUyzu47UU0CN1tGlbisn-oLqt_Y,2702
-phoenix/server/api/routers/v1/experiment_runs.py,sha256=0AUNHA5nvpGDeoJUGK8VxP2TFN3iPwhMW3D9QmHstPk,4399
-phoenix/server/api/routers/v1/experiments.py,sha256=5Rh7q6sHswmk11PZSJ7KMrtqfIE16X_xSKkKSASK9-I,7251
-phoenix/server/api/routers/v1/spans.py,sha256=FEnmlRPBPl71BSGNBuPrz14fk8nmxJQYsKECdDbdUdw,3977
+phoenix/server/api/routers/v1/__init__.py,sha256=vvdpUa2LJPWEg8HbvDm_ANkBAwubPIFPbbHi7elOUws,2808
+phoenix/server/api/routers/v1/dataset_examples.py,sha256=XfqOvDKF1oxb0pkeYfBycwwGt3LnSyyGdMLKC5VKoGQ,6690
+phoenix/server/api/routers/v1/datasets.py,sha256=f2gLG-geu-_wtEw4mKSzNWK2cFb5TYOyRL3tQ7Fl7Es,31544
+phoenix/server/api/routers/v1/evaluations.py,sha256=8g6P_e2BweV3RDU0esFmpkb0L5fCwonQPXiJ0y6HLwg,9126
+phoenix/server/api/routers/v1/experiment_evaluations.py,sha256=HeyV3PXS1BxQpzNOUBpQlX_0JH_jbjZjTxrqy2ujwJQ,2746
+phoenix/server/api/routers/v1/experiment_runs.py,sha256=_c7qmPIja_gpvoVaf_t7KtNc9Zz-0m9da9MS-EcbPBo,3918
+phoenix/server/api/routers/v1/experiments.py,sha256=ntb0lRV2h90mFepWiZfQ1MIAJhOaK9tkWzTejmpwed0,7243
+phoenix/server/api/routers/v1/spans.py,sha256=PFeS3ayKj4cUle0CH-f-CpM1fRi-JicEG7BEtkANzAo,4074
 phoenix/server/api/routers/v1/traces.py,sha256=dYEf5pThenAQCgfQljHdrnwd4tC_tAXm6Kvk6GphPYs,2774
 phoenix/server/api/types/AnnotatorKind.py,sha256=UmYb2KG0JfxdX0mW1qrXrUoIgjMOncRJr1i8mJki1sE,141
 phoenix/server/api/types/Cluster.py,sha256=ac4YfT1OH3xLVmex7EUmB6b9IpULnhLTt554LR0jglE,5689
@@ -159,7 +162,7 @@ phoenix/server/api/types/EvaluationSummary.py,sha256=EFucuzAhcxR9sdEn6WNAtmAGJk-
 phoenix/server/api/types/Event.py,sha256=XdYgaIxcVIW-YFViCkxj5l9OaVNepyIrCtm5Iqg2le8,3989
 phoenix/server/api/types/EventMetadata.py,sha256=-J0tYF9eZTHwCjwxQHY7Gckr2_MNW5OoWT1mydweZNM,635
 phoenix/server/api/types/ExampleRevisionInterface.py,sha256=gV3Gt9-3Oi5wjaVtepC6nOt3FzTzZFD1KebNnqiw56E,294
-phoenix/server/api/types/Experiment.py,sha256=Cs0EKhVLI5l5LKFI0hQA-ekZuaiJcOHT88JGFBa2deU,4906
+phoenix/server/api/types/Experiment.py,sha256=ELYdYFKwgBllxx3cZ_X0XicHjLtshZl0bFqqJdVGXRQ,5177
 phoenix/server/api/types/ExperimentAnnotationSummary.py,sha256=Uk3JtxIrsMoZT5tqc4nJdUOM3XegVzjUyoV3pkjNotE,256
 phoenix/server/api/types/ExperimentComparison.py,sha256=0sFz6MoBDw39dds0qVyaqhVs9qqO5rkG1FMSjmfBeCc,441
 phoenix/server/api/types/ExperimentRun.py,sha256=8jUIi3ApVCqQHwnYe59CYhrmh5iZ6-QmlH5WpF7UWtM,2990
@@ -199,12 +202,12 @@ phoenix/server/static/apple-touch-icon-76x76.png,sha256=CT_xT12I0u2i0WU8JzBZBuOQ
 phoenix/server/static/apple-touch-icon.png,sha256=fOfpjqGpWYbJ0eAurKsyoZP1EAs6ZVooBJ_SGk2ZkDs,3801
 phoenix/server/static/favicon.ico,sha256=bY0vvCKRftemZfPShwZtE93DiiQdaYaozkPGwNFr6H8,34494
 phoenix/server/static/index.css,sha256=KKGpx4iwF91VGRm0YN-4cn8oC-oIqC6HecoPf0x3ZM8,1885
-phoenix/server/static/index.js,sha256=I9Y8svcPruUrXklKcZUxFz5HfLB0vOwczYLSwLAs_04,3500011
+phoenix/server/static/index.js,sha256=qAPO3xGdQ2mIA8TcIPNkIfLvoGeZ78fVKbGZevfJzoM,3518643
 phoenix/server/static/modernizr.js,sha256=mvK-XtkNqjOral-QvzoqsyOMECXIMu5BQwSVN_wcU9c,2564
 phoenix/server/templates/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 phoenix/server/templates/index.html,sha256=S4z7qSoNSwnKFAH9r96AR-YJEyoKMd-VMWVlJ_IdzME,2039
 phoenix/session/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-phoenix/session/client.py,sha256=R7dV38yjkIQa522nhG6jhDllWcXft2JJ7RlcPYpqiiQ,24846
+phoenix/session/client.py,sha256=5mnWVqMFbC8NYbX4m2oRla1VvlmrgabD1oT2UdwDRJ8,33201
 phoenix/session/data_extractor.py,sha256=dwhiDu-ISaXr8UI9I-CszZhB5BlUNmdDopjFZvMIXMw,2101
 phoenix/session/evaluation.py,sha256=aKeV8UVOyq3b7CYOwt3cWuLz0xzvMjX7vlEPILJ_fcs,5311
 phoenix/session/session.py,sha256=rjIuSSK2gAYIUPQTJc4E2ebew5o6I070FWRoFn4W3EI,26620
@@ -213,7 +216,7 @@ phoenix/trace/attributes.py,sha256=xSr2EvlkDS9Wiij7FywXzw7UmMAMLIARSEm_P686BKM,1
 phoenix/trace/errors.py,sha256=wB1z8qdPckngdfU-TORToekvg3344oNFAA83_hC2yFY,180
 phoenix/trace/evaluation_conventions.py,sha256=t8jydM3U0-T5YpiQKRJ3tWdWGlHtzKyttYdw-ddvPOk,1048
 phoenix/trace/exporter.py,sha256=eAYemdvDCHMugDJiaR29BFFMTQBdf3oerdkz34Cl3hE,4736
-phoenix/trace/fixtures.py,sha256=gBGFG2gkcBsSDzolzzR9AJDrB_fdOQfUaGgHV-EHdco,14204
+phoenix/trace/fixtures.py,sha256=tDso17oAYMBQ8GJsAlHFzrC0otXPzUOEh4TgIjEfB7M,14260
 phoenix/trace/otel.py,sha256=WA720jvRadiZBAKjsYoPyXzypHwbyEK2OZRVUwtbjB8,9976
 phoenix/trace/projects.py,sha256=2BwlNjFE-uwpqYtCu5YyBiYZk9wRPpM13vh3-Cv7GkA,2157
 phoenix/trace/schemas.py,sha256=Mjc6fD9OyeMnEk5wPPSbveqnNUYWK3p3BxpOvSGanHU,5950
@@ -239,13 +242,13 @@ phoenix/trace/v1/evaluation_pb2.pyi,sha256=cCbbx06gwQmaH14s3J1X25TtaARh-k1abbxQd
 phoenix/utilities/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 phoenix/utilities/deprecation.py,sha256=cFuTVvjSYyRlrdxdJewjJVieIEHPk30BukSRGRydQ3k,1046
 phoenix/utilities/error_handling.py,sha256=7b5rpGFj9EWZ8yrZK1IHvxB89suWk3lggDayUQcvZds,1946
-phoenix/utilities/json.py,sha256=nuD1SbPm871tbt9Cz7V2NUmS9m0jWgcpRUx1YOV3NYQ,1954
+phoenix/utilities/json.py,sha256=y_w-McDfvlTeGJT28sCtyjzVkwFicakxERG-sGRc8Ak,1948
 phoenix/utilities/logging.py,sha256=lDXd6EGaamBNcQxL4vP1au9-i_SXe0OraUDiJOcszSw,222
 phoenix/utilities/project.py,sha256=qWsvKnG1oKhOFUowXf9qiOL2ia7jaFe_ijFFHEt8GJo,431
 phoenix/utilities/re.py,sha256=PDve_OLjRTM8yQQJHC8-n3HdIONi7aNils3ZKRZ5uBM,2045
 phoenix/utilities/span_store.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-arize_phoenix-4.4.4rc5.dist-info/METADATA,sha256=yT0gbMlPkiRkZeC8Yj_eLyaufriREVn3jxz5-qTKDjI,11012
-arize_phoenix-4.4.4rc5.dist-info/WHEEL,sha256=1yFddiXMmvYK7QYTqtRNtX66WJ0Mz8PYEiEUoOUUxRY,87
-arize_phoenix-4.4.4rc5.dist-info/licenses/IP_NOTICE,sha256=JBqyyCYYxGDfzQ0TtsQgjts41IJoa-hiwDrBjCb9gHM,469
-arize_phoenix-4.4.4rc5.dist-info/licenses/LICENSE,sha256=HFkW9REuMOkvKRACuwLPT0hRydHb3zNg-fdFt94td18,3794
-arize_phoenix-4.4.4rc5.dist-info/RECORD,,
+arize_phoenix-4.4.4rc6.dist-info/METADATA,sha256=gyc5KyS4aFqefmGcezl1eC_8lCZ5DF0iHdSDh0V41f8,11337
+arize_phoenix-4.4.4rc6.dist-info/WHEEL,sha256=1yFddiXMmvYK7QYTqtRNtX66WJ0Mz8PYEiEUoOUUxRY,87
+arize_phoenix-4.4.4rc6.dist-info/licenses/IP_NOTICE,sha256=JBqyyCYYxGDfzQ0TtsQgjts41IJoa-hiwDrBjCb9gHM,469
+arize_phoenix-4.4.4rc6.dist-info/licenses/LICENSE,sha256=HFkW9REuMOkvKRACuwLPT0hRydHb3zNg-fdFt94td18,3794
+arize_phoenix-4.4.4rc6.dist-info/RECORD,,

phoenix/config.py

@@ -233,4 +233,25 @@ def get_env_client_headers() -> Optional[Dict[str, str]]:
     return None
 
 
+def get_base_url() -> str:
+    host = get_env_host()
+    if host == "0.0.0.0":
+        host = "127.0.0.1"
+    base_url = get_env_collector_endpoint() or f"http://{host}:{get_env_port()}"
+    return base_url if base_url.endswith("/") else base_url + "/"
+
+
+def get_web_base_url() -> str:
+    """Return the web UI base URL.
+
+    Returns:
+        str: the web UI base URL
+    """
+    from phoenix.session.session import active_session
+
+    if session := active_session():
+        return session.url
+    return get_base_url()
+
+
 DEFAULT_PROJECT_NAME = "default"

phoenix/datetime_utils.py

@@ -14,6 +14,10 @@ from pandas.core.dtypes.common import (
 _LOCAL_TIMEZONE = datetime.now(timezone.utc).astimezone().tzinfo
 
 
+def local_now() -> datetime:
+    return datetime.now(timezone.utc).astimezone(tz=_LOCAL_TIMEZONE)
+
+
 def normalize_datetime(
     dt: Optional[datetime],
     tz: Optional[tzinfo] = None,

phoenix/db/insertion/evaluation.py

@@ -6,7 +6,7 @@ from typing_extensions import assert_never
 
 from phoenix.db import models
 from phoenix.db.helpers import SupportedSQLDialect, num_docs_col
-from phoenix.db.insertion.helpers import OnConflict, insert_stmt
+from phoenix.db.insertion.helpers import OnConflict, insert_on_conflict
 from phoenix.exceptions import PhoenixException
 from phoenix.trace import v1 as pb
 
@@ -91,7 +91,7 @@ async def _insert_trace_evaluation(
     set_.pop("metadata_")
     set_["metadata"] = values["metadata_"]  # `metadata` must match database
     await session.execute(
-        insert_stmt(
+        insert_on_conflict(
             dialect=dialect,
             table=models.TraceAnnotation,
             values=values,
@@ -139,7 +139,7 @@ async def _insert_span_evaluation(
     set_.pop("metadata_")
     set_["metadata"] = values["metadata_"]  # `metadata` must match database
     await session.execute(
-        insert_stmt(
+        insert_on_conflict(
             dialect=dialect,
             table=models.SpanAnnotation,
             values=values,
@@ -196,7 +196,7 @@ async def _insert_document_evaluation(
     set_.pop("metadata_")
     set_["metadata"] = values["metadata_"]  # `metadata` must match database
     await session.execute(
-        insert_stmt(
+        insert_on_conflict(
             dialect=dialect,
             table=models.DocumentAnnotation,
             values=values,

phoenix/db/insertion/helpers.py

@@ -2,7 +2,7 @@ from abc import ABC
 from enum import Enum, auto
 from typing import Any, Awaitable, Callable, Mapping, Optional, Sequence
 
-from sqlalchemy import Insert, insert
+from sqlalchemy import Insert
 from sqlalchemy.dialects.postgresql import insert as insert_postgresql
 from sqlalchemy.dialects.sqlite import insert as insert_sqlite
 from sqlalchemy.ext.asyncio import AsyncSession
@@ -25,26 +25,18 @@ class OnConflict(Enum):
     DO_UPDATE = auto()
 
 
-def insert_stmt(
+def insert_on_conflict(
     dialect: SupportedSQLDialect,
     table: Any,
     values: Mapping[str, Any],
-    constraint: Optional[str] = None,
-    column_names: Sequence[str] = (),
+    constraint: str,
+    column_names: Sequence[str],
     on_conflict: OnConflict = OnConflict.DO_NOTHING,
     set_: Optional[Mapping[str, Any]] = None,
 ) -> Insert:
     """
     Dialect specific insertion statement using ON CONFLICT DO syntax.
     """
-    if bool(constraint) != bool(column_names):
-        raise ValueError(
-            "Both `constraint` and `column_names` must be provided or omitted at the same time."
-        )
-    if (dialect is SupportedSQLDialect.POSTGRESQL and constraint is None) or (
-        dialect is SupportedSQLDialect.SQLITE and not column_names
-    ):
-        return insert(table).values(values)
     if dialect is SupportedSQLDialect.POSTGRESQL:
         stmt_postgresql = insert_postgresql(table).values(values)
         if on_conflict is OnConflict.DO_NOTHING or not set_:

phoenix/db/insertion/span.py

@@ -7,7 +7,7 @@ from sqlalchemy.ext.asyncio import AsyncSession
 
 from phoenix.db import models
 from phoenix.db.helpers import SupportedSQLDialect
-from phoenix.db.insertion.helpers import OnConflict, insert_stmt
+from phoenix.db.insertion.helpers import OnConflict, insert_on_conflict
 from phoenix.trace.attributes import get_attribute_value
 from phoenix.trace.schemas import Span, SpanStatusCode
 
@@ -27,7 +27,7 @@ async def insert_span(
 ) -> Optional[SpanInsertionEvent]:
     dialect = SupportedSQLDialect(session.bind.dialect.name)
     project_rowid = await session.scalar(
-        insert_stmt(
+        insert_on_conflict(
             dialect=dialect,
             table=models.Project,
             constraint="uq_projects_name",
@@ -87,7 +87,7 @@ async def insert_span(
         cumulative_llm_token_count_prompt += cast(int, accumulation[1] or 0)
         cumulative_llm_token_count_completion += cast(int, accumulation[2] or 0)
     span_rowid = await session.scalar(
-        insert_stmt(
+        insert_on_conflict(
             dialect=dialect,
             table=models.Span,
             constraint="uq_spans_span_id",

phoenix/db/models.py

@@ -92,7 +92,7 @@ class UtcTimeStamp(TypeDecorator[datetime]):
 
 
 class ExperimentResult(TypedDict, total=False):
-    result: Dict[str, Any]
+    result: Any
 
 
 class Base(DeclarativeBase):

phoenix/experiments/__init__.py

@@ -0,0 +1,6 @@
+from .functions import evaluate_experiment, run_experiment
+
+__all__ = [
+    "evaluate_experiment",
+    "run_experiment",
+]

phoenix/experiments/evaluators/__init__.py

@@ -0,0 +1,29 @@
+from phoenix.experiments.evaluators.code_evaluators import (
+    ContainsAllKeywords,
+    ContainsAnyKeyword,
+    ContainsKeyword,
+    JSONParsable,
+    MatchesRegex,
+)
+from phoenix.experiments.evaluators.llm_evaluators import (
+    CoherenceEvaluator,
+    ConcisenessEvaluator,
+    HelpfulnessEvaluator,
+    LLMCriteriaEvaluator,
+    RelevanceEvaluator,
+)
+from phoenix.experiments.evaluators.utils import create_evaluator
+
+__all__ = [
+    "create_evaluator",
+    "ContainsAllKeywords",
+    "ContainsAnyKeyword",
+    "ContainsKeyword",
+    "JSONParsable",
+    "MatchesRegex",
+    "CoherenceEvaluator",
+    "ConcisenessEvaluator",
+    "LLMCriteriaEvaluator",
+    "HelpfulnessEvaluator",
+    "RelevanceEvaluator",
+]

phoenix/experiments/evaluators/base.py

@@ -0,0 +1,153 @@
+import functools
+import inspect
+from abc import ABC
+from types import MappingProxyType
+from typing import Any, Awaitable, Callable, Optional, Union
+
+from typing_extensions import TypeAlias
+
+from phoenix.experiments.evaluators.utils import validate_signature
+from phoenix.experiments.types import (
+    AnnotatorKind,
+    EvaluationResult,
+    EvaluatorKind,
+    EvaluatorName,
+    EvaluatorOutput,
+    ExampleInput,
+    ExampleMetadata,
+    ExampleOutput,
+    TaskOutput,
+)
+
+
+class Evaluator(ABC):
+    """
+    A helper super class to guide the implementation of an `Evaluator` object.
+    Subclasses must implement either the `evaluate` or `async_evaluate` method.
+    Implementing both methods is recommended, but not required.
+
+    This Class is intended to be subclassed, and should not be instantiated directly.
+    """
+
+    _kind: AnnotatorKind
+    _name: EvaluatorName
+
+    @functools.cached_property
+    def name(self) -> EvaluatorName:
+        if hasattr(self, "_name"):
+            return self._name
+        return self.__class__.__name__
+
+    @functools.cached_property
+    def kind(self) -> EvaluatorKind:
+        if hasattr(self, "_kind"):
+            return self._kind.value
+        return AnnotatorKind.CODE.value
+
+    def __new__(cls, *args: Any, **kwargs: Any) -> "Evaluator":
+        if cls is Evaluator:
+            raise TypeError(f"{cls.__name__} is an abstract class and should not be instantiated.")
+        return object.__new__(cls)
+
+    def evaluate(
+        self,
+        *,
+        output: Optional[TaskOutput] = None,
+        expected: Optional[ExampleOutput] = None,
+        metadata: ExampleMetadata = MappingProxyType({}),
+        input: ExampleInput = MappingProxyType({}),
+        **kwargs: Any,
+    ) -> EvaluationResult:
+        # For subclassing, one should implement either this sync method or the
+        # async version. Implementing both is recommended but not required.
+        raise NotImplementedError
+
+    async def async_evaluate(
+        self,
+        *,
+        output: Optional[TaskOutput] = None,
+        expected: Optional[ExampleOutput] = None,
+        metadata: ExampleMetadata = MappingProxyType({}),
+        input: ExampleInput = MappingProxyType({}),
+        **kwargs: Any,
+    ) -> EvaluationResult:
+        # For subclassing, one should implement either this async method or the
+        # sync version. Implementing both is recommended but not required.
+        return self.evaluate(
+            output=output,
+            expected=expected,
+            metadata=metadata,
+            input=input,
+            **kwargs,
+        )
+
+    def __init_subclass__(cls, is_abstract: bool = False, **kwargs: Any) -> None:
+        super().__init_subclass__(**kwargs)
+        if is_abstract:
+            return
+        evaluate_fn_signature = inspect.signature(Evaluator.evaluate)
+        for super_cls in inspect.getmro(cls):
+            if super_cls in (LLMEvaluator, Evaluator):
+                break
+            if evaluate := super_cls.__dict__.get(Evaluator.evaluate.__name__):
+                assert callable(evaluate), "`evaluate()` method should be callable"
+                # need to remove the first param, i.e. `self`
+                _validate_sig(functools.partial(evaluate, None), "evaluate")
+                return
+            if async_evaluate := super_cls.__dict__.get(Evaluator.async_evaluate.__name__):
+                assert callable(async_evaluate), "`async_evaluate()` method should be callable"
+                # need to remove the first param, i.e. `self`
+                _validate_sig(functools.partial(async_evaluate, None), "async_evaluate")
+                return
+        raise ValueError(
+            f"Evaluator must implement either "
+            f"`def evaluate{evaluate_fn_signature}` or "
+            f"`async def async_evaluate{evaluate_fn_signature}`"
+        )
+
+
+def _validate_sig(fn: Callable[..., Any], fn_name: str) -> None:
+    sig = inspect.signature(fn)
+    validate_signature(sig)
+    for param in sig.parameters.values():
+        if param.kind is inspect.Parameter.VAR_KEYWORD:
+            return
+    else:
+        raise ValueError(f"`{fn_name}` should allow variadic keyword arguments `**kwargs`")
+
+
+class CodeEvaluator(Evaluator, ABC, is_abstract=True):
+    """
+    A convenience super class for defining code evaluators.
+
+    This class is intended to be subclassed, and should not be instantiated directly.
+    """
+
+    _kind = AnnotatorKind.CODE
+
+    def __new__(cls, *args: Any, **kwargs: Any) -> "CodeEvaluator":
+        if cls is CodeEvaluator:
+            raise TypeError(f"{cls.__name__} is an abstract class and should not be instantiated.")
+        return object.__new__(cls)
+
+
+class LLMEvaluator(Evaluator, ABC, is_abstract=True):
+    """
+    A convenience super class for defining LLM evaluators.
+
+    This class is intended to be subclassed, and should not be instantiated directly.
+    """
+
+    _kind = AnnotatorKind.LLM
+
+    def __new__(cls, *args: Any, **kwargs: Any) -> "LLMEvaluator":
+        if cls is LLMEvaluator:
+            raise TypeError(f"{cls.__name__} is an abstract class and should not be instantiated.")
+        return object.__new__(cls)
+
+
+ExperimentEvaluator: TypeAlias = Union[
+    Evaluator,
+    Callable[..., EvaluatorOutput],
+    Callable[..., Awaitable[EvaluatorOutput]],
+]

phoenix/{datasets → experiments}/evaluators/code_evaluators.py

@@ -4,11 +4,11 @@ import json
 import re
 from typing import Any, List, Optional, Union
 
-from phoenix.datasets.evaluators.utils import Evaluator
-from phoenix.datasets.types import EvaluationResult, TaskOutput
+from phoenix.experiments.evaluators.base import CodeEvaluator
+from phoenix.experiments.types import EvaluationResult, TaskOutput
 
 
-class JSONParsable(Evaluator):
+class JSONParsable(CodeEvaluator):
     def evaluate(self, *, output: Optional[TaskOutput] = None, **_: Any) -> EvaluationResult:
         assert isinstance(output, str), "Experiment run output must be a string"
         try:
@@ -21,7 +21,7 @@ class JSONParsable(Evaluator):
         )
 
 
-class ContainsKeyword(Evaluator):
+class ContainsKeyword(CodeEvaluator):
     def __init__(self, keyword: str, name: Optional[str] = None) -> None:
         self.keyword = keyword
         self._name = name or f"Contains({repr(keyword)})"
@@ -38,7 +38,7 @@ class ContainsKeyword(Evaluator):
         )
 
 
-class ContainsAnyKeyword(Evaluator):
+class ContainsAnyKeyword(CodeEvaluator):
     def __init__(self, keywords: List[str], name: Optional[str] = None) -> None:
         self.keywords = keywords
         self._name = name or f"ContainsAny({keywords})"
@@ -56,7 +56,7 @@ class ContainsAnyKeyword(Evaluator):
         )
 
 
-class ContainsAllKeywords(Evaluator):
+class ContainsAllKeywords(CodeEvaluator):
     def __init__(self, keywords: List[str], name: Optional[str] = None) -> None:
         self.keywords = keywords
         self._name = name or f"ContainsAll({keywords})"
@@ -76,7 +76,7 @@ class ContainsAllKeywords(Evaluator):
         )
 
 
-class MatchesRegex(Evaluator):
+class MatchesRegex(CodeEvaluator):
     def __init__(self, pattern: Union[str, re.Pattern[str]], name: Optional[str] = None) -> None:
         if isinstance(pattern, str):
             pattern = re.compile(pattern)

phoenix/{datasets → experiments}/evaluators/llm_evaluators.py

@@ -2,19 +2,19 @@ import re
 from types import MappingProxyType
 from typing import Any, Callable, Optional, Type
 
-from phoenix.datasets.evaluators.utils import (
-    ExampleInput,
-    ExampleMetadata,
+from phoenix.evals.models.base import BaseModel as LLMBaseModel
+from phoenix.evals.utils import snap_to_rail
+from phoenix.experiments.evaluators.base import (
     ExperimentEvaluator,
     LLMEvaluator,
-    _unwrap_json,
 )
-from phoenix.datasets.types import (
+from phoenix.experiments.evaluators.utils import unwrap_json
+from phoenix.experiments.types import (
     EvaluationResult,
+    ExampleInput,
+    ExampleMetadata,
     TaskOutput,
 )
-from phoenix.evals.models.base import BaseModel as LLMBaseModel
-from phoenix.evals.utils import snap_to_rail
 
 
 class LLMCriteriaEvaluator(LLMEvaluator):
@@ -62,7 +62,7 @@ class LLMCriteriaEvaluator(LLMEvaluator):
 
     def _format_eval_template(self, output: TaskOutput) -> str:
         assert output is not None
-        result = _unwrap_json(output)
+        result = unwrap_json(output)
         return self.template.format(text=str(result))
 
     def _parse_eval_output(self, unparsed_response: str) -> EvaluationResult:
@@ -217,7 +217,7 @@ class RelevanceEvaluator(LLMEvaluator):
         self, output: Optional[TaskOutput] = None, *args: Any, **kwargs: Any
     ) -> str:
         assert output is not None
-        return str(_unwrap_json(output))
+        return str(unwrap_json(output))
 
     def evaluate(
         self,