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)
• add spec for hyperparameters in task design and coder (#995) (10246fd)
• align scenario descriptions and include debug timeout (#1079) (13b6663)
• allow prev_out keys to be None in workspace cleanup assertion (#1214) (1f4d190)
• based on response schema; not function calling (#1038) (99da8c5)
• cancel tasks on resume and kill subprocesses on termination (#1166) (cf6e418)
• change runner prompts (#1223) (6d3e73d)
• clear ws_ckp after extraction to reduce workspace object size (#1137) (783affe)
• correct DS_LOCAL_DATA_PATH error in devcontainer (#1063) (588fcfa)
• dockerfile: install coreutils to resolve timeout command error (#1260) (07f89b0)
• docs: update rdagent ui with correct params (#1249) (f360d0a)
• enable embedding truncation (#1188) (2421fa4)
• enhance feedback handling in MultiProcessEvolvingStrategy for improved task evolution (#1274) (961e561)
• error in prompt template (#1065) (a90e598)
• filter log folders bug in ui (#1073) (d0f33c5)
• fix a bug in return curve display (#1042) (249f661)
• fix a small bug in json_mode (#1041) (8bc12ea)
• fix a small bug in response_schema (#1043) (66cadcd)
• fix bug for hypo_select_with_llm when not support response_schema (#1208) (54cc2c4)
• fix chat_max_tokens calculation method to show true input_max_tokens (#1241) (7d749b8)
• fix code diff bug (#1115) (4603e88)
• fix mcts (#1270) (c73f67a)
• fix some bugs in RD-Agent(Q) (#1143) (44fd2ee)
• graph: using assignment expression to avoid repeated function call (#1174) (b4f57ce)
• handle mixed str and dict types in code_list (#1279) (63ecb3b)
• handle None output and conditional step dump in LoopBase execution (#1212) (68b6985)
• handle the no-update case of root node in uncommited_rec_status (#1062) (ead8dce)
• handle ValueError in stdout shrinking and refactor shrink logic (#1228) (bc7a3b4)
• ignore case when checking metric name (#1160) (fc0df6e)
• ignore class types when filtering workflow steps (#1085) (64e3ec8)
• ignore RuntimeError for shared workspace double recovery (#1140) (8fc1e9b)
• improve the logic of json_schema and refine the reasoning extraction logic for reasoning model (#1044) (12060b1)
• increase retry count in hypothesis_gen decorator to 10 (#1230) (c4b8baa)
• increase time default not controlled by LLM (#1196) (8c62561)
• insert await asyncio.sleep(0) to yield control in loop (#1186) (5705be0)
• jinja problem of enumerate (#1216) (af9068c)
• kaggle competition metric direction (#1195) (a933b6c)
• merge candidates (#1254) (5a78c89)
• minor conflict in prompts (#1081) (f821e4c)
• minor fix to runtime_environment (#1089) (bff82ef)
• model/factor experiment filtering in Qlib proposals (#1257) (0f722e1)
• move snapshot saving after step index update in loop execution (#1206) (0e3a9af)
• move task cancellation to finally block and fix subprocess kill typo (#1234) (fb628e3)
• package and timer bug (#1092) (7faf6d9)
• path traversal risk (#1050) (2f78216)
• prevent JSON content from being added multiple times during retries (#1255) (9d46a68)
• prevent parallelism in feedback and record steps (#1046) (d0272a9)
• prompt yaml (#1112) (1f2c9b1)
• properly assign sota_exp_fb before None comparison (#1037) (5d6a927)
• refine DSCoSTEER_eval prompts (#1157) (c62e5fc)
• refine prompt, equal lightgbm, discourage over hypertuning (#1072) (56ba15a)
• refine prompt; runner focus on low hanging fruit (#1076) (1778b8c)
• refine prompts and add additional package info (#1179) (22428a4)
• refine task scheduling logic in MultiProcessEvolvingStrategy for… (#1275) (417766e)
• refine the prompt to force complete code & refine the logic of running (#1069) (1e61de3)
• remove refine decision & bug fix (#1031) (0059a6a)
• remove unused imports in data science scenario module (#1136) (2307237)
• replace hardcoded ChromeDriver path with webdriver-manager (#1271) (40876e2)
• revert 2 commits (#1239) (1265ae9)
• revert to v10 setting (#1220) (d868188)
• scheduler next selection parallel disorder (#1028) (f468595)
• set requires_documentation_search to None to disable feature in eval (#1245) (e117234)
• skip res_ratio check if timer or res_time is None (#1189) (17400a3)
• split then sample & remove simple model guide in ds proposal (#1034) (2dde8b8)
• stop evolve if global timer is timeout (#1039) (ad37417)
• summary page bug (#1219) (36fec9a)
• TypeError: cannot unpack non-iterable bool object (#1036) (f4370a4)
• ui bug (#1192) (ad901aa)
• update fallback criterion (#1210) (05fca1a)
• update requirements.txt's streamlit (#1133) (512d08f)
• use CoSTEERSettings for DSRunnerCoSTEERSettings (#1096) (152a70f)