[kms12425-ctrl]

Summary

Fix the medical diagnosis demo so data setup and tests work correctly in the existing configured environment.

Changes

• fix run_diagnosis.py to use the in-package default config path
• fix prepare_data.py to read/write data under the medical diagnosis app directory
• fix setup_data.sh to reuse the current Python interpreter instead of drifting to another environment
• update medical diagnosis embedding config from hf to hash
• ignore generated medical test artifacts and local dataset cache

Validation

Ran the following successfully in the existing conda environment:

• python -m pytest -q apps/tests/test_medical_diagnosis.py
• python -m pytest -q apps/tests/medical_diagnosis/test_knowledge_base.py
• python -m pytest -q apps/tests/medical_diagnosis/test_image_quality.py
• python -m pytest -q apps/tests/medical_diagnosis/test_image_analyzer.py
• SAGE_ENABLE_MEDICAL_DIAGNOSIS=1 python -m pytest -q apps/tests/medical_diagnosis/test_diagnosis.py

Also verified:

• python examples/run_medical_diagnosis.py

Notes

• The example still falls back when the local LLM service is unavailable, but the diagnosis flow completes correctly.

变更概述

修复 medical diagnosis 示例中的数据准备与路径问题，使其能够在现有已配置环境中正确完成数据生成、测试执行和示例运行。

主要修改

• 修复 run_diagnosis.py 的默认配置路径，改为使用应用目录内的配置文件
• 修复 prepare_data.py 的数据读写路径，统一落到 medical diagnosis 应用目录下
• 修复 setup_data.sh，改为复用当前 Python 解释器，避免切到其他环境
• 将 medical diagnosis 的 embedding 配置从 hf 调整为 hash
• 补充忽略规则，避免将本地生成的数据缓存和测试产物误提交

验证结果

已在当前 conda 环境中完成以下验证：

• python -m pytest -q apps/tests/test_medical_diagnosis.py
• python -m pytest -q apps/tests/medical_diagnosis/test_knowledge_base.py
• python -m pytest -q apps/tests/medical_diagnosis/test_image_quality.py
• python -m pytest -q apps/tests/medical_diagnosis/test_image_analyzer.py
• SAGE_ENABLE_MEDICAL_DIAGNOSIS=1 python -m pytest -q apps/tests/medical_diagnosis/test_diagnosis.py

另外已验证：

• python examples/run_medical_diagnosis.py

说明

• 当本地 LLM 服务不可用时，示例仍会回退到模板报告生成，但整体诊断流程可以正常完成。
• 本次修改未创建新的虚拟环境，全部验证均基于现有 conda 环境完成。

[Copilot]

Pull request overview

Fixes the medical diagnosis demo’s data/setup pathing so dataset preparation, tests, and the example entrypoint work reliably in the existing configured (conda) environment.

Changes:

• Update setup_data.sh to use python -m pip via a configurable interpreter and point to in-app scripts.
• Rebase medical diagnosis data read/write paths to live under the app directory (rather than repo root / examples paths).
• Switch embedding configuration defaults from hf to hash and ignore generated artifacts/caches.

Reviewed changes

Copilot reviewed 6 out of 7 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
apps/src/sage/apps/medical_diagnosis/setup_data.sh	Uses a configurable Python interpreter and updates script/data paths and printed usage hints.
apps/src/sage/apps/medical_diagnosis/scripts/prepare_data.py	Writes processed outputs under the medical diagnosis app directory.
apps/src/sage/apps/medical_diagnosis/run_diagnosis.py	Defaults config path to the in-package config/agent_config.yaml.
apps/src/sage/apps/medical_diagnosis/config/agent_config.yaml	Switches embedding method to hash.
apps/src/sage/apps/medical_diagnosis/agents/diagnostic_agent.py	Updates built-in default config to use hash embedding.
.vscode/settings.json	Updates VS Code Python env project configuration for conda.
.gitignore	Ignores additional generated test artifacts and local dataset cache files.

[Copilot]

setup_data.sh calls download_lumbar_dataset.py, which writes to a relative Path("data/..."), but this script never cds to $SCRIPT_DIR. If a user runs setup_data.sh from another working directory, the dataset will be downloaded to the wrong location and prepare_data.py (which reads under the app dir) won’t find it. Consider cd "$SCRIPT_DIR" early in the script (or invoke the Python scripts with an explicit working directory) to make paths robust.

Suggested change

	# 确保后续 Python 脚本中的相对路径始终相对于当前应用目录解析
	cd "$SCRIPT_DIR" || {
	echo "❌ 无法切换到脚本目录: $SCRIPT_DIR"
	exit 1
	}

[Copilot]

PYTHON_BIN is introduced to avoid drifting to another environment, but the early “dataset already exists” stats block still uses a hard-coded python3 -c (above). For consistency and to ensure the same interpreter is always used, switch that invocation to use $PYTHON_BIN as well.

[Copilot]

The post-setup usage hint suggests running a pytest test file directly (python apps/tests/medical_diagnosis/test_diagnosis.py ...). That file calls pytest.skip(...) and relies on pytest markers, which can behave unexpectedly when executed as a standalone script. Consider updating the message to a python -m pytest ... invocation (and/or keep the example entrypoint python examples/run_medical_diagnosis.py --interactive).

Suggested change

	echo " python apps/tests/medical_diagnosis/test_diagnosis.py --mode single"
	echo " python examples/run_medical_diagnosis.py --interactive"
	echo " $PYTHON_BIN -m pytest apps/tests/medical_diagnosis/test_diagnosis.py"
	echo " $PYTHON_BIN examples/run_medical_diagnosis.py --interactive"

[Copilot]

prepare_dataset() now loads from APP_DIR/data/medical/.../cache, but there’s no explicit existence check before load_from_disk(...). When the dataset isn’t present (or was downloaded to a different working directory), this will raise a stack trace that’s not very actionable. Consider checking dataset_path.exists() and raising a FileNotFoundError with a message pointing to the expected location and the setup script to run.

Suggested change

	if not dataset_path.exists() or not dataset_path.is_dir():
	raise FileNotFoundError(
	"未找到医疗数据集缓存目录。\n"
	f"预期位置: {dataset_path}\n"
	"请先运行 medical_diagnosis 应用的数据下载/初始化脚本，"
	"确保数据集已下载到上述目录后，再重新执行本脚本。"
	)