colm-openbench 3.0.0__tar.gz

colm_openbench-3.0.0/.gitignore

@@ -0,0 +1,75 @@
+# Python
+__pycache__/
+*.py[cod]
+*$py.class
+*.egg-info/
+*.egg
+dist/
+build/
+.eggs/
+
+# Virtual environments
+.venv/
+venv/
+env/
+
+# IDE
+.vscode/
+.idea/
+*.swp
+*.swo
+
+# OS
+.DS_Store
+Thumbs.db
+
+# Project
+output/
+output_station_smoke/
+logs/
+*.nc
+*.nc4
+# Exception: ship the standard classification masks (IGBP/PFT/Köppen) with the
+# package. They are compressed integer masks (~1.1 MB total) consumed by the
+# *_groupby evaluation paths via openbench.util.static_datasets. Users can still
+# override them via $OPENBENCH_DATASET_DIR or ./dataset/.
+!src/openbench/dataset/
+!src/openbench/dataset/*.nc
+cache/
+.pytest_cache/
+.ruff_cache/
+htmlcov/
+.coverage
+
+# Registry runtime scratch (atomic-write backup + advisory lock files,
+# created next to reference_catalog.yaml / model_catalog.yaml).
+*.bak
+*.lock
+*.tmp
+
+OpenBench-wei/
+openbench-wizard/
+bk/
+
+test_data/
+.worktrees/
+
+# Local/generated tests and planning artifacts
+docs/performance-benchmark.md
+tests/manual/test_script_entrypoints.py
+
+# Personal/local test configs
+test_real.yaml
+test_wei.yaml
+test_wei2.yaml
+test_wei_full.yaml
+test_cross_dtype.yaml
+test_river.yaml
+test_lu_stn.yaml
+openbench_full_options.yaml
+outputs/
+openbench_init_*.yaml
+/test_*.yaml
+
+docs/superpowers/
+/培训课件

colm_openbench-3.0.0/CHANGELOG.md

@@ -0,0 +1,334 @@
+# Changelog
+
+## [3.0.0] - 2026-09-11
+
+First stable 3.0 release, including the main-branch fixes validated across Linux,
+macOS and Windows on Python 3.10–3.12. The opt-in uncertainty-aware pipeline
+remains on its separate development branch.
+
+### Fixed
+- Shared grid preprocessing no longer depends on which reference is evaluated
+  first; per-pair masked references survive cache reuse and post-processing errors.
+- Statistics respect preprocessed units, finite observations and coordinate
+  alignment; ANOVA, PLSR and Three-Cornered Hat receive the correct source layout.
+- Non-Gregorian monthly conversions, computed reference variables and mixed
+  flat/resolution-specific reference directories retain their intended data.
+- Remote configuration changes apply atomically without retaining another
+  host's passwords; recursive deletes clear matching cached and pending files.
+
+### Maintenance
+- Remove obsolete processing scaffolding and consolidate calendar and saved
+  credential handling without removing public compatibility entry points.
+- Share wheel/sdist resource checks between CI and publishing, require explicitly
+  selected artifacts to exist, and document fresh-build release verification.
+- Remove redundant comments and duplicate test scaffolding while retaining
+  meaningful scientific documentation and deterministic regression coverage.
+- Align the Conda recipe with the stable PyPI version and verify its dependency,
+  command-line and bundled-data installation contracts.
+
+## [3.0.0b16] - 2026-08-31
+
+Beta release over 3.0.0b15 for remote workflow reliability, expanded evaluation
+methods, and complete output packaging.
+
+### Added
+- Evaluation-guide appendix metrics and methods, including deterministic,
+  categorical, uncertainty, trend, and distribution diagnostics.
+- Live remote CPU and memory monitoring tied to the active OpenBench process.
+
+### Changed
+- Index of Agreement is available through the metric workflow.
+- Generated figures are published under the advertised top-level `figures/`
+  directory while existing metric, comparison, and report paths remain valid.
+
+### Fixed
+- Remote installation, Conda environment creation, SSH host selection, output
+  paths, project confirmation, and local/remote dataset discovery.
+- Remote progress now waits for the real evaluation process across jump hosts
+  instead of reporting premature success or completion.
+- Legacy flat reference layouts, CLM multi-stream file prefixes, and simulation
+  data scanning now preserve the files selected by the user.
+- Appendix NetCDF execution, GUI/Qt CI stability, and repository formatting.
+
+## [3.0.0b15] - 2026-08-26
+
+Beta release over 3.0.0b14 for GUI/runtime reliability, truthful reporting,
+station evaluation fixes, and preprocessing performance.
+
+### Added
+- Content-addressed xESMF weight caching to reuse identical regridding weights.
+- GUI progress events and bounded report summaries for safer long-running local
+  and remote evaluations.
+
+### Changed
+- Local Dask evaluation avoids distributed NetCDF/HDF5 worker writes and keeps
+  scanned simulation case cards readable.
+- Shared preprocessing, reference masking, MFM component work, and multi-model
+  comparisons reuse expensive intermediate work where results are equivalent.
+
+### Fixed
+- GUI local and remote workflows now fail truthfully instead of silently falling
+  back, stalling, or running on the wrong target.
+- Reference and simulation scanning preserve requested datasets, station lists,
+  paths, and inline metadata more reliably.
+- Time alignment, finite-observation masks, singleton conservative regrids, and
+  station missing-value handling no longer silently alter scientific results.
+- Large NetCDF/CSV reports remain bounded and avoid fabricated or incomplete
+  summaries.
+- Windows path normalization, environment discovery, CPU quota handling, and CI
+  formatting stability.
+- Station evaluation process updates from PRs #167 and #170.
+
+## [3.0.0b14] - 2026-08-22
+
+Beta release over 3.0.0b13 for actionable GUI preflight diagnostics.
+
+### Fixed
+- GUI configuration-check failures now include the complete CLI check output,
+  so validation details are not hidden by the final summary lines.
+
+## [3.0.0b13] - 2026-08-22
+
+Beta release over 3.0.0b12 for safer configuration, registration, and GUI
+workflows.
+
+### Added
+- Chinese/English GUI switching with immediate page retranslation.
+- Back navigation throughout interactive CLI setup and registration wizards.
+
+### Changed
+- GUI evaluations now run the same CLI preflight check before execution.
+- Reference rescans distinguish new and registered datasets, default to new
+  entries, and enrich only explicitly selected registered datasets.
+- CLI registration accepts quoted whitespace-containing units in named
+  `key=value` variable specifications and provides actionable shell guidance.
+
+### Fixed
+- CLI and GUI preflight reject evaluations whose configured years do not
+  overlap the available data period.
+- Reference rescans preserve manual catalog corrections and keep remote refresh
+  work responsive and narrowly targeted.
+- Multi-step CLI back navigation retains previously entered values across
+  variables and empty intermediate steps.
+- Cross-platform CI behavior for paths, generated artifacts, and GUI tests.
+
+## [3.0.0b12] - 2026-08-18
+
+Beta release over 3.0.0b11 for reliable generated station configuration.
+
+### Fixed
+- `openbench init` and `openbench sim scan` now write absolute station
+  `fulllist` paths so generated YAML works from any current directory.
+
+## [3.0.0b11] - 2026-08-13
+
+Beta release over 3.0.0b10 for GUI dataset discovery, registry coverage, and
+statistical evaluation correctness.
+
+### Added
+- Expanded Crop reference catalog coverage and CoLM2024 model metadata.
+- Run-manifest and configuration-key validation for more reproducible runs.
+
+### Changed
+- GUI simulation scans retain per-case grid/station metadata and prepare station
+  lists before local preview/export.
+- SMPI remains available through its dedicated comparison workflow instead of
+  the single-value metric selector.
+
+### Fixed
+- GUI navigation now persists edits, registered reference paths are no longer
+  overwritten by scan roots, and simulation discovery follows the selected run.
+- Reference station-list resolution, Streamflow catalog entries, units, and
+  registry behavior were corrected.
+- Mann-Kendall handles tied short series, categorical kappa rejects continuous
+  inputs, and SMPI bootstrap intervals are consistent for NumPy and Dask data.
+
+## [3.0.0b10] - 2026-07-18
+
+Beta release over 3.0.0b9 for station-reference routing correctness.
+
+### Changed
+- Renamed the active configuration runtime module from `legacy_processors.py`
+  to `runtime_info.py` to reflect that it remains on the main execution path.
+
+### Fixed
+- Non-Streamflow station references no longer enter the discharge/CaMA-specific
+  `station_matching` engine.
+- Reference scans remove stale non-Streamflow matching blocks and generate
+  portable station-list catalogs instead.
+
+## [3.0.0b9] - 2026-07-16
+
+Beta release over 3.0.0b8 for registry coverage and CI stability.
+
+### Changed
+- Station fulllists generated by scans are stored under `OPENBENCH_HOME`,
+  allowing shared reference roots to remain read-only.
+- Expanded bundled model and reference metadata.
+
+### Fixed
+- Compute validation now accepts keyword arguments such as `isel(lake=0)`.
+- Restored required station-matching metadata and removed incomplete generated
+  station catalog entries.
+- Simulation scanning retains the TE `Albedo` compatibility key.
+- NetCDF evaluation writes no longer retain lazy references to closed source
+  datasets.
+- Ruff formatting for the FFT component metric.
+
+## [3.0.0b8] - 2026-07-08
+
+Patch release over 3.0.0b7 for model coverage, audit fixes, and CI stability.
+
+### Added
+- Bundled ECLand and LM4 model profiles and aliases.
+
+### Fixed
+- Station simulation scanning now loads merged NetCDF sources before writing the
+  target file, avoiding platform-sensitive HDF5/netCDF CI hangs.
+- Hardened audit fixes for compute sandboxing, remote uploads, registry locking,
+  station lists, time decoding, station matching, and unit conversions.
+
+## [3.0.0b7] - 2026-06-11
+
+Patch release over 3.0.0b6 for GUI selection behavior.
+
+### Fixed
+- Multi-resolution reference selection now allows lower-frequency variants such
+  as LowRes/Month instead of forcing the highest-frequency available variant.
+- Simulation dataset scanning no longer auto-selects every model-profile
+  variable or expands the user's Evaluation Variables selection in the preview.
+
+## [3.0.0b6] - 2026-06-11
+
+Patch release over 3.0.0b5 for GUI wizard flow alignment.
+
+### Changed
+- GUI setup now follows the CLI/config order: choose Evaluation Variables first,
+  then configure reference datasets and simulation datasets for those variables.
+
+## [3.0.0b5] - 2026-06-11
+
+Patch release over 3.0.0b4 for GUI reference selection and remote SSH hardening.
+
+### Fixed
+- Multi-resolution reference selection no longer crashes when the dialog is
+  populated from registry `ReferenceDataset` entries, which do not carry the
+  scan-only `file_count` field.
+- SSH output streaming falls back to short polling when `select.select()`
+  rejects Paramiko-like channels on platforms such as Windows.
+
+## [3.0.0b4] - 2026-06-11
+
+Remote-workflow hardening release: the GUI's remote mode was reworked end to
+end (responsive SSH layer + three review rounds over it).
+
+### Fixed
+- Conda activation on remote hosts uses the POSIX `.` command — the `source`
+  bashism broke every conda-env remote flow on hosts whose `/bin/sh` is
+  dash/ash (commands are wrapped in `sh -c` for csh/tcsh login shells)
+- Remote paths starting with `~` (including the documented `~/OpenBench`
+  default) now expand to `$HOME` everywhere a path reaches the remote shell:
+  the evaluation run command, pip dependency step, conda create, the
+  OpenBench install check, the sync engine, scanners, validators, NML/model
+  editors and file browsers. Bare `shlex.quote` turned them into literal-`~`
+  paths that no remote shell resolves
+- Remote dataset scanning trusts the inspection results shipped from the
+  remote host instead of coincidentally same-named local directories
+  (wrong-machine inspection on shared mounts), and station fulllists are
+  generated remotely for `stn` references
+- A dropped SSH session is reported as "Connection Lost" instead of the
+  misleading "Remote directory not found"
+- The remote scan bootstrap `expanduser()`s the checkout path so a plain-git
+  `~/OpenBench` install is importable even without the pip step
+
+### Changed
+- Every SSH call made from the GUI thread now routes through a responsive
+  worker layer: the window keeps painting during connects, scans and
+  installs; long installs stream output into Esc-guarded progress dialogs;
+  scans and installs are cancellable mid-command
+- Remote installs run `pip install -e` for dependencies (the old
+  requirements.yml flow targeted a file that no longer exists)
+- Browse start-path resolution and symlink-target classification each probe
+  the remote host in a single compound round trip (was up to N serial
+  round trips per click)
+- The Runtime page's local git install/update streams through a worker
+  thread with a stoppable subprocess (no more event-loop pumping)
+
+## [3.0.0b3] - 2026-06-09
+
+### Changed
+- `pip install colm-openbench` is now fully featured by default: statistics
+  (scikit-learn, statsmodels), plotting (seaborn), legacy migration (f90nml)
+  and HTML reports ship in the base install. Only `[gui]` (PySide6), `[remote]`
+  (paramiko) and `[report]` (xhtml2pdf — needs system cairo) remain optional;
+  `[all]` = gui + remote + report.
+
+### Added
+- Reference datasets: `Rodai2025_NPP` (station NPP) and lake stations
+  `G_REALM_LakeLevel`, `GLAST_LakeSurfaceWaterTemperature`, `ReaLSAT_LakeArea`
+- 30-minute (half-hour) time-resolution detection
+
+### Docs
+- User's Guide (CN+EN): climatology added to the `tim_res` field table; a
+  multi-model `simulation` example; `timezone` flagged as not-yet-implemented;
+  default `io`/`dask` sub-blocks shown; detailed cases for unresolvable
+  reference profiles and `sim scan` model inference
+
+## [3.0.0b2] - 2026-06-09
+
+Bug-fix and hardening release over 3.0.0b1 (deep code review + full cross-platform CI).
+
+### Fixed
+- Target diagram plotted total RMSE on the uRMSD axis; now passes centered CRMSD so
+  points satisfy RMSD² = bias² + uRMSD² (with an invariant guard)
+- `absolute_percent_bias` normalizes by `|Σo|` so a negative observed sum no longer yields a negative APB
+- ANOVA default `analysis_type` corrected to `oneway` (was the non-accepted `one-way`)
+- `br2` uses `|slope|` (Krause 2005), keeping it in `[0, r²]` for negative slopes
+- SMPI grid path applies area weights and uses a consistent bootstrap dimension
+- Taylor grid summary computes std/correlation/CRMSD over one pairwise finite mask
+- NetCDF writes are serialized with a lock (netCDF4/HDF5 is not thread-safe) — fixes a segfault
+- Registry catalog writes hold a cross-process lock; `delete_reference` writes a tombstone
+- Config rejects non-mapping `project`/`evaluation`/`simulation` sections with a clean error
+- Cross-platform path handling (POSIX paths in catalogs/configs; Windows file reads)
+
+### Changed
+- Version is a single source of truth (`__init__.__version__`, hatchling dynamic)
+- The full test suite (84 files) now runs on CI across Linux/macOS/Windows
+
+### Added
+- 30-minute (half-hour) time-resolution detection
+- Expanded reference dataset metadata; `CITATION.cff`
+
+## [3.0.0b1] - 2026-06-07
+
+First public beta of the 3.0 line. APIs and config schema may still change
+before the 3.0.0 final release.
+
+### Added
+- Unified package structure (`src/openbench/`) merging OpenBench-wei and openbench-wizard
+- Single YAML configuration file (`openbench.yaml`) replacing 4-6 file setup
+- Data registry with 101 reference datasets (69 grid + 32 station)
+- 22 model profiles (CoLM2024, CLM5, NoahMP5, ERA5-Land, …)
+- Bundled IGBP / PFT / Köppen classification masks for group-by analysis
+- CLI commands: `run`, `check`, `init`, `ref`, `sim`, `model`, `migrate`, `cache`, `gui`, `version`
+- Interactive config generator (`openbench init`)
+- Config migration tool (`openbench migrate`) for old JSON/NML formats
+- `_defaults` merge support for simulation configs
+- `!include` tag support in YAML configs
+- Three time alignment strategies: intersection, per_pair, strict
+- Config adapter bridging new and legacy evaluation engine formats
+- SSH remote execution infrastructure (requires `openbench[remote]`)
+- GUI wizard (requires `openbench[gui]`)
+- Optional extras: `[gui]`, `[remote]`, `[report]`, `[all]`
+- GitHub Actions CI (lint + test matrix)
+- 70+ tests covering config, registry, metrics, scores, CLI
+
+### Changed
+- Build system: hatchling (was setuptools/manual)
+- Config format: YAML only (JSON and Fortran NML deprecated)
+- Package name: `colm-openbench` on PyPI
+- Python requirement: >=3.10
+
+### Migration
+- Use `openbench migrate old-config.json -o openbench.yaml` to convert existing configs
+- Evaluation results are numerically identical to v2.0

colm_openbench-3.0.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2023-2026 Zhongwang Wei, CoLM-SYSU
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.