MCPcopy Create free account

hub / github.com/Jennyee1/AcademicAgent / functions

Functions1,438 in github.com/Jennyee1/AcademicAgent

↓ 1 callersFunction_format_metrics_inline
从 run_summary["by_metric"] 取每个指标的 mean,拼成单行文本。
src/evaluation/monitor.py:104
↓ 1 callersFunction_get_encoder
懒加载 sentence-transformer。失败返回 None,匹配阶段自动跳过 embedding 档。
src/evaluation/metrics/kg_semantic.py:46
↓ 1 callersFunction_get_linker_for
(graph_store)
src/mcp_servers/knowledge_graph.py:124
↓ 1 callersFunction_get_prompts
加载所有 analysis prompt 模板
src/core/multimodal.py:132
↓ 1 callersFunction_git_hash
()
src/evaluation/runner.py:33
↓ 1 callersFunction_go
()
scripts/experiments/chunk_model_sweep.py:203
↓ 1 callersFunction_gold_labels
(gold: dict)
scripts/validate_review.py:45
↓ 1 callersFunction_imported_modules
从源码里解析出 import 的模块名(去掉 `as` 别名 / 子模块、合并 `import a, b`)。
src/execution/sandbox.py:181
↓ 1 callersMethod_keyword_fallback
关键词匹配回退(无 scikit-learn 时使用)
src/knowledge/graph_store.py:678
↓ 1 callersFunction_kg_extraction_for
(slug: str, run_id: str | None)
scripts/gen_codex_review_prompt.py:46
↓ 1 callersFunction_llm_rca
调 MiniMax 生成根因假设。失败返回 None,让上层 fallback 到规则表。 用 OpenAI-compatible client(同 extractor.py 的路径):MINIMAX_API_KEY + MINIMAX_BASE_URL;模型用 MINIMAX_M
src/evaluation/auto_rca.py:132
↓ 1 callersMethod_load
从 JSON 加载注册表
src/core/paper_registry.py:568
↓ 1 callersMethod_load
(self)
src/knowledge/entity_linker.py:75
↓ 1 callersFunction_load_all_adapters
触发所有 adapter 模块的 import(从而完成注册)。 逐模块 try/except:某个 adapter 模块缺少传递依赖(如 httpx)时, 只让该模块的工具不可用,不拖垮其余离线 adapter。
src/evaluation/adapters/base.py:80
↓ 1 callersFunction_load_gold
(slug: str)
scripts/validate_review.py:38
↓ 1 callersFunction_load_progress
()
scripts/ingest_llm_agent_v1.py:109
↓ 1 callersFunction_load_progress
()
scripts/ingest_comms_v1.py:127
↓ 1 callersFunction_load_reviews
同一篇的所有 reviewer JSON,文件名形如 <slug>.<reviewer>.json。
scripts/apply_gold_reviews.py:43
↓ 1 callersMethod_load_state
(self)
src/recorder/refactor_recorder.py:443
↓ 1 callersFunction_load_thresholds_for_cards
Phase 6.1 后:thresholds.yaml 仅作为失败卡片"是否低于基线"的判定输入。 门禁退出码已移除(个人项目不再做 CI 风格 gate)。本函数取代 gate.load_thresholds, 保持失败卡片生成逻辑不变。
src/evaluation/runner.py:47
↓ 1 callersMethod_loose_parse
宽松 JSON 解析(从 LLM 输出中提取 JSON)
src/report/generator.py:465
↓ 1 callersFunction_m
(t)
lib/vis-9.1.2/vis-network.min.js:26
↓ 1 callersFunction_make_encoder
构造 encoder。返回 (encoder, is_strong)。 is_strong=True 表示是 sentence-transformer;False 表示 fallback。
src/knowledge/semantic_index.py:98
↓ 1 callersFunction_make_resource_limit_preexec
构造 subprocess 的 preexec_fn,用于 POSIX setrlimit。 Windows 不支持 setrlimit,返回 None(让 subprocess 用默认行为)。 生产场景 Windows 应该用 Job Object,但本 lite 沙箱不引入额外
src/execution/sandbox.py:251
↓ 1 callersMethod_maybe_recorder_sync
Phase 6.1: 把 run 摘要走 RefactorRecorder("eval-runs") 同步管道。
src/evaluation/runner.py:251
↓ 1 callersMethod_maybe_trim
文件超长时把最老的若干小节合并成"压缩摘要"占位。 策略简单:超 max_chars 时保留最后 N 个 `## [...]` 小节 + 在头部 写一行 "[truncated N entries before {date}]" 占位。 不调 LLM
src/runtime/memory.py:103
↓ 1 callersFunction_merge_aliases
合并 ADD_ALIAS 提议的 aliases,按归一化去重,保留原顺序。
scripts/apply_gold_reviews.py:93
↓ 1 callersMethod_merge_results
合并多段抽取结果(本地去重) 【工程思考】为什么本地合并而不用 LLM merge? 1. 节省 token 成本(merge prompt 也要消耗 token) 2. 本地去重逻辑确定性高(node_id 相同则合并)
src/knowledge/extractor.py:551
↓ 1 callersMethod_normalize_label
标签标准化:小写 + 去除特殊字符 + 空格转下划线 用于保证 "ReAct" == "react" == "REACT" 映射到同一节点
src/knowledge/schema.py:298
↓ 1 callersFunction_parse_kvs
把 --key value --key2 value2 转成 {key: value}。
src/runtime/cli.py:43
↓ 1 callersMethod_parse_result
将 LLM 原始输出解析为结构化结果
src/core/multimodal.py:354
↓ 1 callersFunction_path_for
每个 domain 一个 JSON 文件。
src/mcp_servers/knowledge_graph.py:62
↓ 1 callersFunction_pdf_text
完整 PDF text。复用 src.core.pdf_parser。
scripts/experiments/chunk_model_sweep.py:50
↓ 1 callersFunction_rate_limit_wait
主动限速:确保两次请求间隔不低于 MIN_REQUEST_INTERVAL(并发安全)
src/mcp_servers/paper_search.py:112
↓ 1 callersFunction_read_json
(path: Path, default)
src/evaluation/auto_rca.py:51
↓ 1 callersFunction_read_jsonl
(path: Path)
src/evaluation/monitor.py:89
↓ 1 callersFunction_read_output
(args: argparse.Namespace)
src/recorder/cli.py:43
↓ 1 callersFunction_relations_incompatible
两个 relation 是否在 _INCOMPATIBLE_PAIRS 内同组。
src/knowledge/evidence.py:165
↓ 1 callersMethod_render
(self, last: str = "")
src/evaluation/monitor.py:61
↓ 1 callersFunction_render_dashboard
渲染仪表盘 HTML
src/report/dashboard.py:170
↓ 1 callersMethod_render_html
渲染 HTML 报告 【工程选择】先用字符串拼接实现,后续迁移到 Jinja2 模板。 原因:Phase 1A 先跑通完整链路,Phase 4 再用 Jinja2 做高质量模板。
src/report/generator.py:247
↓ 1 callersFunction_render_learning_path_html
渲染学习路径 / 盲区检测的 HTML 报告
skills/learning_path/scripts/analyze_knowledge.py:268
↓ 1 callersMethod_reports
(self)
src/evaluation/runner.py:237
↓ 1 callersFunction_resolve_tag
(tag: str)
scripts/experiments/summarize_chunk_model_sweep.py:20
↓ 1 callersMethod_rewrite_md_header
更新 md 头部的 (pending) 占位为真实 page_id。
src/recorder/refactor_recorder.py:352
↓ 1 callersFunction_rule_based_rca
(cat: str, items: list[dict])
src/evaluation/auto_rca.py:117
↓ 1 callersFunction_run_one
(job: PaperJob)
scripts/ingest_llm_agent_v1.py:134
↓ 1 callersFunction_run_one
(job: PaperJob)
scripts/ingest_comms_v1.py:145
↓ 1 callersMethod_run_step
( self, step: WorkflowStep, ctx: WorkflowContext, reflector, )
src/runtime/workflow.py:209
↓ 1 callersFunction_save_progress
(state: dict[str, JobResult])
scripts/ingest_llm_agent_v1.py:121
↓ 1 callersFunction_save_progress
(state: dict[str, JobResult])
scripts/ingest_comms_v1.py:137
↓ 1 callersFunction_save_run_log
(run_log: dict)
src/runtime/cli.py:62
↓ 1 callersFunction_seed_run_dir
(base: Path)
scripts/smoke_evaluation_phase6.py:31
↓ 1 callersFunction_setup_logging
(verbose: bool)
src/runtime/cli.py:35
↓ 1 callersFunction_sha256
文件 sha256;文件不存在返回空串。
src/evaluation/isolation.py:70
↓ 1 callersFunction_sha256_text
(text: str)
scripts/build_personal_v1_tasks.py:57
↓ 1 callersFunction_short
(v: Any, max_len: int = 80)
src/runtime/experience.py:75
↓ 1 callersFunction_should_skip
返回 skip 原因;不应 skip 时返回 None。 --offline 同时跳过 requires_api 与 requires_llm(LLM 也是网络调用), 使离线 smoke run 完全确定、不触达任何外部服务。
src/evaluation/layers/layer1_component.py:27
↓ 1 callersFunction_should_skip
(task: TaskSpec, offline: bool)
src/evaluation/layers/layer2_workflow.py:34
↓ 1 callersMethod_slugify
将标题转为文件名安全的 slug
src/report/generator.py:446
↓ 1 callersFunction_valid_schema_types
()
src/evaluation/metrics/kg.py:123
↓ 1 callersFunction_write_empty_kg
(path: Path, domain: str, dry_run: bool)
scripts/kg_reset_for_v1.py:45
↓ 1 callersMethod_write_md_header
(self)
src/recorder/refactor_recorder.py:295
↓ 1 callersFunctionaC
(t)
lib/vis-9.1.2/vis-network.min.js:26
↓ 1 callersFunctionaO
(t)
lib/vis-9.1.2/vis-network.min.js:26
↓ 1 callersFunctionaT
(t,e)
lib/vis-9.1.2/vis-network.min.js:26
↓ 1 callersFunctionabort_workflow
()
scripts/smoke_runtime_phase5.py:92
↓ 1 callersFunctionadd_paper_to_graph
从论文文本中抽取知识实体和关系,添加到知识图谱中。 适合使用的场景: - 用户说"把这篇论文加入知识图谱" - 需要积累论文中的关键概念、方法、评估指标 - 构建个人学术知识网络 两种输入方式(任选其一): 1. pdf_path: 提供
src/mcp_servers/knowledge_graph.py:142
↓ 1 callersFunctionaggregate_run
生成 run_summary.json 的内容。
src/evaluation/aggregate.py:65
↓ 1 callersFunctionall_workflows
()
src/evaluation/workflows/registry.py:53
↓ 1 callersFunctionam
(t)
lib/vis-9.1.2/vis-network.min.js:26
↓ 1 callersMethodanalyze
分析一张论文图表 Args: image_base64: 图片的 base64 编码 context: 论文上下文(如该图所在段落的文字) figure_type: 已知的图表类型(UNKNO
src/core/multimodal.py:199
↓ 1 callersMethodappend_blocks
分批追加 blocks(每批 ≤100)。返回是否全部成功。
src/recorder/notion_client.py:267
↓ 1 callersFunctionappend_version_log
把一个 run 的摘要追加到 eval_runs.md。失败返回 None(非致命)。
src/evaluation/reporting/version_log.py:36
↓ 1 callersFunctionartifact_produced_rate
是否按期望产出了文件(图表等)。expect_artifact=False 时恒为 1.0。
src/evaluation/metrics/capability.py:154
↓ 1 callersFunctionbC
(t)
lib/vis-9.1.2/vis-network.min.js:26
↓ 1 callersMethodbatch_semantic_dedup
对一批 KGNode/KGEdge 做一次性语义去重。 策略:按 node_type 分桶 → 桶内 encode 全部 label(含 store 同 type 节点) → 计算 cos_sim 矩阵 → 贪心合并:cos_sim ≥ threshold 的对,把
src/knowledge/entity_linker.py:337
↓ 1 callersFunctionbf
(t)
lib/vis-9.1.2/vis-network.min.js:26
↓ 1 callersFunctionbuild_gap_evolution
从历史学习路径报告中提取盲区演进数据
src/report/dashboard.py:108
↓ 1 callersFunctionbuild_kg_growth
利用 created_at 双时态数据构建图谱增长时间线
src/report/dashboard.py:91
↓ 1 callersFunctionbuild_kg_tasks
读 gold/*.json,输出 (kg tasks, kg gold rows)。 任务规格:layer1_component / kg_extraction / smoke / add_paper_to_graph, args.text_file 引向 fixtures/pap
scripts/build_personal_v1_tasks.py:61
↓ 1 callersFunctionbuild_parser
()
src/recorder/cli.py:128
↓ 1 callersFunctionbuild_parser
()
src/runtime/cli.py:121
↓ 1 callersFunctionbuild_prompt
(slug: str, run_id: str | None = None)
scripts/gen_codex_review_prompt.py:72
↓ 1 callersFunctionbuild_report
(classifications: list[NodeClassification])
scripts/migrate_kg_to_dual.py:247
↓ 1 callersFunctionbuild_retrieval_tasks
读 gold/retrieval.jsonl,输出 retrieval tasks + retrieval gold rows。
scripts/build_personal_v1_tasks.py:123
↓ 1 callersFunctionbuild_test_image
Generate a 256x128 PNG with 'VLM-OK 42' text + a red filled circle.
scripts/probe_vision.py:55
↓ 1 callersFunctionbullet
(text: str)
src/recorder/block_builders.py:70
↓ 1 callersFunctioncall_model
(client: OpenAI, model: str, paper_title: str, text: str)
scripts/spot_check_kg.py:101
↓ 1 callersFunctioncall_one
(client: OpenAI, model: str, paper_title: str, text: str)
scripts/spot_check_v2.py:242
↓ 1 callersMethodcanLoad
(e)
lib/tom-select/tom-select.complete.min.js:184
↓ 1 callersMethodclassify_batch
(self, nodes: list[KGNode])
scripts/migrate_kg_to_dual.py:136
↓ 1 callersMethodclearOptions
()
lib/tom-select/tom-select.complete.min.js:235
↓ 1 callersFunctioncm
(t,e,i,n)
lib/vis-9.1.2/vis-network.min.js:26
↓ 1 callersFunctioncmd_deep
深度多模态解析:一次性提取文本 + 章节结构 + 所有嵌入图片。 设计原理: - 文本用文本提取(零 Vision token 消耗) - 图片只提取嵌入的 figure(不做全页渲染,节省 token) - 自动过滤过小的图片(icon/lo
skills/paper_reader/scripts/parse_pdf.py:209
↓ 1 callersFunctioncmd_detect_gaps
Detect knowledge gaps.
skills/learning_path/scripts/analyze_knowledge.py:62
↓ 1 callersFunctioncmd_download
Download a PDF from arXiv ID or direct URL. Supports: - arXiv ID: "2210.03629" → https://arxiv.org/pdf/2210.03629.pdf - Direct U
skills/paper_reader/scripts/parse_pdf.py:48
↓ 1 callersFunctioncmd_images
Extract images from a specific page and save to disk.
skills/paper_reader/scripts/parse_pdf.py:172
↓ 1 callersFunctioncmd_importance
Get concept importance ranking.
skills/learning_path/scripts/analyze_knowledge.py:116
↓ 1 callersFunctioncmd_learning_path
Generate learning path.
skills/learning_path/scripts/analyze_knowledge.py:45
← previousnext →601–700 of 1,438, ranked by callers