Releases: microsoft/RD-Agent
Releases · microsoft/RD-Agent
v0.8.0
0.8.0 (2025-11-03)
Features
- add a rag mcp in proposal (#1267) (a0cd102)
- add coder check and give more time (#1127) (e32d229)
- add enable_cache toggle for UI data caching (#1075) (0c9f193)
- add extra_eval config and import_class for custom evaluators (#1097) (5accec3)
- add hypo_critic and hypo_rewrite in proposal (#1106) (71440f6)
- add improve_mode to MultiProcessEvolvingStrategy for selective task implementation (#1273) (9344635)
- add loop ID mapping to trace nodes and update UI labels (#1098) (5437851)
- add mask inference in debug mode (#1154) (ef749ab)
- add only success filter toggle for traces (#1047) (5e582cc)
- add option to enable hyperparameter tuning only in first eval loop (#1211) (bc3fa17)
- add previous runner loops to runner history (#1142) (8de9f75)
- add reasoning attribute to DSRunnerFeedback for enhanced evaluation context (#1162) (4e41c97)
- add sample submission file check (#1053) (6a840d8)
- add show_hard_limit option and update time limit handling in DataScience settings (#1144) (fe762cd)
- add stdout into workspace for easier debugging (#1236) (d3d4967)
- add time ratio limit for hyperparameter tuning in Kaggle settin… (#1135) (e44bc83)
- add user interaction in data science scenario (#1251) (2afef70)
- add ws CLI and support optional timeout/cache (#1066) (fae3def)
- analyze feedback based on sota numbers (#1116) (167f5e2)
- create Jupyter notebook pipeline file based on main.py file (#1134) (2fa1790)
- enable drafting with knowledge (#998) (8e385eb)
- enable finetune llm (#1055) (909c7d6)
- enable LLM‑based hypothesis selection with time‑aware prompt & colored logging (#1122) (1c4ab89)
- enable meta planner (#1103) (c208209)
- enable to inject diversity cross async multi-trace (#1173) (bcdd957)
- enhance timeout handling in CoSTEER and DataScience scenarios (#1150) (06233cb)
- enhance timeout management and knowledge base handling in CoSTEER components (#1130) (963d260)
- fallback to acceptable results (#1129) (3ce2bd4)
- improve fallback handling in CoSTEER and add GPU usage guidelin… (#1165) (cec4240)
- init pydantic ai agent & context 7 mcp (#1240) (59af538)
- mcp: cache with one-click toggle (#1269) (6f86863)
- mcts policy based on trace scheduler (#1203) (13890e0)
- new prompt for auto-sota-selector (#1109) (13c92a9)
- offline selector (#1231) (76b2e87)
- prob-based trace scheduler (#1131) (970561a)
- query & cache package_info (#1083) (19869ea)
- refactor CoSTEER classes to use DSCoSTEER and update max seconds handling (#1156) (6d01e3e)
- refine the logic of enabling hyperparameter tuning and add criteira (#1175) (af071f5)
- show the summarized final difference between the final workspace and the base workspace (#1281) (2bf8345)
- streamline hyperparameter tuning checks and update evaluation g… (#1167) (383e5ed)
- ui, support disable cache (#1217) (92efe33)
- update README with latest paper acceptance to NeurIPS 2025 (#1252) (8332960)
Bug Fixes
- add a switch for ensemble_time_upper_bound and fix some bug in main (#1226) (f00a538)
- add gpu_info in research phase (#1094) (58c9c1b)
- add json format response fallback to prompt templates (#1246) (4dfb8a1)
- add metric in scores.csv and avoid reading sample_submission.csv (#1152) (fd039f1)
- add missing self parameter to instance methods in DSProposalV2ExpGen (#1213) ([68af035](ht...
v0.7.0
0.7.0 (2025-07-08)
Features
- add code change summary (#1000) (937ec26)
- add hide_base_name option and update data folder prompts (#1004) (2f61fa8)
- added running time statistics for the DS scenario experiment (#1007) (030abd8)
- merge code summary and support more traces (#1025) (48201e7)
- show first evo round codes diff (#1009) (4844622)
- try coder on whole data (#1017) (4973e05)
Bug Fixes
v0.6.1
v0.6.0
0.6.0 (2025-06-26)
Features
Bug Fixes
- add async to direct_exp_gen avoid infinite loop (#992) (78c203d)
- docker container cleanup to prevent accumulation and system slowdown (#975) (05cf094)
- fix a bug and update the docs (#978) (d1ae9e1)
- merge datascience v3 and v2 (#974) (1ba7548)
- refine details (#979) (25caa3d)
- refine prompt (#987) (76df96e)
v0.5.0
0.5.0 (2025-06-18)
Features
- add a check for whether values in score_df are NaN (#756) (d9cc780)
- add competition level filter and extract constants to utils (#869) (b40b605)
- add DocDev for auto-generating workspace documentation (#781) (bcba6ea)
- add drafting pipeline (#832) (efedddf)
- add last_exp_fb to DSTrace and update feedback retrieval usage (#910) (10531fd)
- add mlflow logger in RD loop to log (#815) (b91b54f)
- add naive experiment generator and update proposal configurations (#759) (75494f4)
- add RD-Agent-Quant scenario (#838) (6e42d52)
- add reasoning_effort parameter to LiteLLMAPIBackend and LLMSett… (#754) (113889f)
- add reviewer in feedback (#765) (1a95bee)
- advanced checkpoint selectors (#790) (50ea033)
- archive python and csv files in workspace to maintain results (#814) (67d0e01)
- checkpoint selection (#744) (a15a06a)
- custom data (#810) (6322916)
- dump model (#776) (b49481e)
- enable to set different version of idea-proposal for multi traces (#895) (236c28f)
- enhance compatibility with more LLM models (#905) (8800624)
- idea pool integrated to exp_gen & add timer to RD-Agent & pause-resume to RD-loops (#795) (e62aefa)
- joblib cache (#749) (83a0411)
- log api status to mlflow (#860) (049921b)
- log reaching max time limit before breaking CoSTEER evolution (#921) (837fff2)
- merge failed and successful traces together (#766) (3a2aa8c)
- merge selectively (#888) (06ba314)
- multi-trace online merge (#886) (2112d67)
- new proposal (structured outputs) prompts (#887) (150796a)
- parallel loop running based on asyncio (#932) (c63e207)
- propose hypothesis across multiple parts in pipeline (#827) (acb0e21)
- pull image with progress (#777) (5cad086)
- raise error when timeout in api call (#793) (eafd4df)
- raise policy violation (#894) (5b9d007)
- reanalyze competition info & pipeline coding evaluator prompt (#837) (f7b5258)
- refine merge (#842) (99463b4)
- refine prompt (#760) (a91b182)
- replace hard-coded cache paths with dynamic cache_path config (#952) (db56894)
- revert draft stage into a soft decay in hypothesis selection (#849) (d41db0c)
- trace merging (#836) (a3d5473)
- truncate by time (#863) (2b9427a)
- update prompt to improve json respond format of some LLM models (#928) (0b84709)
- using different chat model in different part (#822) (c052ea6)
Bug Fixes
- 'DSProposalV2ExpGen' object has no attribute 'COMPONENT_TASK_MAP… (#950) (e353895)
- adapting UI to mock trace (#841) (8a5754c)
- add missing semicolon after chmod in env shell command (#955) (1128eaa)
- add time to timer when api timeout bug (#826) (f45d6ae)
- add wait_retry to exp_gen v2 (#783) (b9fb7cf)
- adjust ds_trace lookup and add stderr redirect to mlebench command (#853) (4e53108)
- align competion_full_desc and scenario_all_desc, remove redundant info in problems proposal (#808) (76d8536)
- bug fix in timer start (#807) (9af7161)
- bug in problem identification (#806) (e1d5a29)
- conda error information (#941) (fd39a94)
- default cost to NaN when ca...
v0.4.0
0.4.0 (2025-04-04)
Features
- (Kaggle) add base template for competition: tabular-playground-series-may-2022 (#481) (f3405ca)
- a unified CoSTEER to fit more scenarios (#491) (cddbd02)
- add a new competition (#474) (2fc0d77)
- add a tool to enable saving workspace files into a specific folder (#728) (bca864b)
- add baseline score stat (#590) (2948026)
- add configurable volume mode for Docker volumes in env.py (#537) (642a022)
- add constraint labels for semantic search (#680) (0584cfc)
- add cross validation to workflow (#700) (82e9b00)
- add describe_data_folder_v2 (#738) (bc8e846)
- add do_truncate control for the load function (#656) (2b960a5)
- add do_truncate control for the load function (#656) (2b960a5)
- add eda to data science scenario (#639) (35aa479)
- add hypothesis guidelines and rule-based ranking (#746) (c077b82)
- Add line length limit to shrink_text function and settings (#715) (75ed5e1)
- add loop_n parameter to the main loop (#611) (778c166)
- add max time config to costeer in data science (#645) (534686c)
- add mlebench submission validitor (#545) (712d94a)
- add model removal and adjust some framework logic (#681) (1edf881)
- add output_path to load function of LoopBase (#628) (dd33726)
- add pipeline coder (#742) (759f295)
- add rank into report (mle_summary) (#665) (13f7922)
- add restart and fix unzip (#538) (ed2c7d1)
- add retry mechanism with wait_retry decorator and refactor diff generation (#572) (de1cd72)
- add the shape of the CSV to the dataset description (#561) (a10c881)
- add timeout settings and cleanup step in data science runner (#539) (295abd5)
- add type checker to api backend & align litellm and old backend (#647) (d38eae9)
- align mlebench data and evaluation & several fix on kaggle workflow (#477) (f6c522b)
- backend: integrate LiteLLM API Backend (#564) (f477687)
- base data science scenario UI (#525) (39917b3)
- condaenv & full docker env (#668) (084dd6d)
- diff mode fix (#569) (0c509f5)
- display LLM prompt (#676) (8c93bba)
- Dynamically find and use sample submission file in eval tests (#542) (5f12b44)
- end-to-end optimization (#473) (d41343a)
- Enhance eval script with file cleanup and detailed submission checks (#529) (cf2ff92)
- exclude invalid session log folder (#554) (fa86e4d)
- improve the framework's ability to adaptively adjust the model (#629) (93806f3)
- independent use_azure_token_provider on chat and embedding (#452) (d223004)
- integrate azure deepseek r1 (#591) (e79ce5c)
- kaggle refactor (#489) (1b057d0)
- kaggle: several update in kaggle scenarios (#476) (245d211)
- loader prompt & simplify YAML loading and update data loader specifications (#736) (86f8bbf)
- make spec optional (#719) (a16b70f)
- Make system prompt role customizable in LLM settings (#632) (e4acd92)
- multi log folder, replace "epxx" in workspace path (#555) (8a69c9c)
- new exp gen v2 implementation (#725) (5dcc2d5)
- new-york-city-taxi-fare-prediction_template (#488) (a9caab7)
- out spec change for o1-preview (#666) (22894bd)
- refactor for general data science (#498) ([7002dc4](https://github.com/microsof...
v0.3.0
0.3.0 (2024-10-21)
Features
- add a new template for kaggle (#289) (eee3ab5)
- add download submission.csv button for kaggle scenario (#317) (dcdcbe4)
- add kaggle command (#271) (0938394)
- add kaggle tpl: feedback-prize (#331) (a288e39)
- add more templates for kaggle (#291) (da752ec)
- add normal rag into framework (#360) (91b0b1f)
- add qlib_factor_strategy (#307) (f8f59ff)
- Add ranking in kaggle scenario (#401) (b16b4be)
- Add runtime measurement for each step and loop in RDLoop. (#281) (83058c8)
- add s3e11 kaggle template (#324) (8c57524)
- Added RepoAnalyzer to empower auto-summary of a workspace (#264) (0bd349a)
- Added support for loading and storing RAG in Kaggle scenarios. (#269) (c4895de)
- announce Discord and WeChat (#367) (acac507)
- auto submit result after one kaggle RDLoop (#345) (ab55d70)
- better feedback & evaluation (#346) (cc9a8c1)
- Dynamic scenario based on task (#392) (665a037)
- Factor Implement Search Enhancement (#294) (4ecf25f)
- Feature selection v3 to support all actions (#280) (0047641)
- fix some bugs and add original features' description (#259) (1a5f45a)
- get kaggle notebooks & disscussion text for RAG (#371) (cead345)
- Iceberge competition (#372) (c10ea4f)
- implement isolated model feature selection loop (#370) (cf1292d)
- Initial version if Graph RAG in KAGGLE scenario (#301) (fd3c0fd)
- Integrate RAG into the Kaggle scenarios. (#262) (be0e48a)
- Kaggle loop update (Feature & Model) (#241) (4cf22a6)
- kaggle templates related (#287) (785fdc1)
- Model context for tuning and selection (#284) (f2831e7)
- Modify FactorRowCountEvaluator and FactorIndexEvaluator to return the ratio (#328) (8f43f8e)
- New competition - Optiver (#356) (3705efe)
- random forest for s3e11 (#347) (b57846d)
- refine the code in model description and fix some bugs in feedback.py (#288) (5b124d7)
- refine the template in several Kaggle competitions (#343) (034f238)
- Revise to support better hypothesis proposal (#390) (c55ec0a)
- show workspace in demo (#348) (ddf567c)
- support Multi output (#330) (3d36c45)
- Supporting COVID-19 competition (#374) (a1b63db)
- supporting Mnist competition (#375) (e958a34)
- Supporting Model Specifications (#319) (e126471)
- supporting various Kaggle competitions & scenarios for RD-Agent (#409) (75eea22)
- template for kaggle (#308) (ff97cf0)
- use auto gen seed when using LLM cache (#441) (ca15365)
- use unified pickle cacher & move llm config into a isolated config (#424) (2879ecf)
- xgboost gpu accelerate (#359) (56a5b8f)
Bug Fixes
- a bug of developer& edit s4e8 template (#338) (f12ce72)
- actively raised errors aer also considered as negative feedback. (#268) (46ec908)
- bug of saving preprocess cache files (#310) (5fb0608)
- cache (#383) (f2a6e75)
- change css tag of kaggle competition info crawler (#306) (1e3d38b)
- debug dsagent (#387) (8fe9511)
- eval_method cannot catch run factor error (#260) ([2aaab31](https://github.com/microsoft/RD-Agent/com...
v0.2.1
v0.2.0
0.2.0 (2024-09-07)
Features
- add collect info (#233) (89f4af9)
- add cross validation for kaggle scenario (#236) (e0b03ba)
- add progress status for docker env (#215) (538d4ef)
- Added loop code for Kaggle scene. (#211) (975c327)
- Demo display effect and usage (#162) (8cf122a)
- piloting of the framework (#227) (e9b103e)
- support more models for kaggle scenario (#223) (e3a9659)
- update model_experiment.py to support basic EDA (#220) (bf2684c)