Spaces:
Sleeping
Sleeping
Commit History
-
Sync from GitHub Actions
Verified signatureb68936d
-
Sync from GitHub Actions
Verified signature202497a -
Sync from GitHub Actions
Verified signaturef7dec05 -
Sync from GitHub Actions
Verified signature36edab1 -
Sync from GitHub Actions
Verified signature70d8e1b -
Sync from GitHub Actions
Verified signature3599f8f
-
fix(hybrid): apply_adjustment에서 역사 행(yhat=None) 산술 오류 수정
역사 구간 행에 조정 계수 곱셈 시도로 발생하는 NoneType 연산 오류 수정. 시나리오도 예측 구간만 포함. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>f4e3cc6 -
fix(hybrid): 차트에 역사 데이터(실측값) 표시 안 되는 문제 수정
앙상블 결과에 역사 구간 데이터를 prepend하여 실측값 선이 차트에 렌더링되도록 수정. 시나리오 및 엑셀 내보내기의 인덱스 오정렬도 함께 수정. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>d402898 -
feat(hybrid): LLM 보정 범위를 앙상블 불확실성 기반 동적 클램프로 교체
고정 ±30% 대신, 모델 간 변동계수(uncertainty)를 기반으로 보정 허용 범위를 데이터에 맞게 자동 산정한다. margin = max(10%, min(30%, uncertainty × 2)) Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>d706d31 -
feat: Add 80% confidence interval support for Prophet, ARIMA, and Theta models
d86d6bd
-
fix(data_service): EUC-KR 인코딩 및 한국어 날짜 형식 지원 추가
- CSV 파일 읽기 시 utf-8-sig 실패 시 cp949/euc-kr 폴백 시도 - DATE_FORMATS에 '%Y년%m월' 형식 추가 (주유소 유가이력 등) Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>57dd81c -
fix(backend): strip markdown code blocks from Claude response in timesfm_service
149e23a
-
fix: install timesfm from GitHub source for 2.5 support, restore correct API
69e9837 -
fix: use timesfm-1.0-200m-pytorch (PyPI compatible), cap horizon at 128
0e66adb -
fix: use correct timesfm.TimesFm API instead of non-existent TimesFM_2p5_200M_torch
2666821 -
fix: AI interpret response check, remove 95% CI option, serialize dict to JSON for Claude prompt
4e391bd -
fix: address code review issues in timesfm_service.py
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>7f4692d -
feat: add timesfm_service.py with singleton model loading
835ad76
-
feat(hybrid): Add data polarity option and Excel export functionality
da4a1a1 -
chore: fix python 3.9 type hints, integrate venv in start.sh, and add dotenv support
0e7340f -
fix: resolve macOS OpenMP conflict between PyTorch and LightGBM in ensemble
- Set KMP_DUPLICATE_LIB_OK=TRUE and OMP_NUM_THREADS=1 in main.py before ML library imports to prevent process crash when N-BEATS (PyTorch) and LightGBM run concurrently (they use different OpenMP runtimes) - Add asyncio.Semaphore(3) in run_ensemble_async to cap concurrent threads - Add asyncio.wait_for(timeout=180) per model to guard against hangs Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>a12fbd3 -
fix: add KeyError guard and ImportError handling in llm_adjustment_service
b0ed680 -
feat: add llm_adjustment_service with Claude API integration and tests
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>da6142c -
fix: use get_running_loop, add length guard and DataFrame type check in ensemble_service
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>9a1c9e5 -
feat: add ensemble_service with async orchestrator and pure-function tests
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>556ff34
-
fix: resolve validation split causing insufficient training samples in N-BEATS
- Cap val_size so training set always has at least icl+ocl+30 samples (previously max() caused val_size to dominate, leaving ~9-21 training samples which prevented meaningful learning → MASE>2, R²<0) - Tighten can_validate to require min_fit_size training points - Fix infer_freq fallback: return "YS" instead of "MS" for yearly data Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>0b5ef5a
-
Add models:LightGBM,Theta, ETS
김재겸6d538de -
Creating a new common data profiling model, etc.
김재겸b646b17
-
Add ARIMA/SARIMA models
김재겸a9a54a5
-
Add Prophet, XGBoost models
김재겸0237ba0 -
first commit
김재겸d939ae5