rd-agent-guide
安装方式
先用 Prompt 核对来源,也可切换到安装命令或下载可用的安装包。
用 AI 助手安装
复制提示词,让助手先核对固定版本和依赖。
查看 Prompt收起 Prompt
先检查当前 Agent 是否已安装 SkillPI Skill Finder。如果尚未安装,请按官方仓库的说明安装并启用:https://github.com/Findskills/skillpi。然后使用 Skill Finder 打开此 SkillPI 详情页并定位 "rd-agent-guide":https://skillpi.com/zh/skills/deepskill-de6751f0949819c4d44345b4dcd244486e9955be120db16e6c563c9b18e5877f。安装前核对上游来源和版本:https://github.com/brycewang-stanford/Auto-Empirical-Research-Skills/blob/bfa0a6e5a843babbf45a1ef1663fd5dc4f07c0ba/skills/43-wentorai-research-plugins/skills/research/automation/rd-agent-guide/SKILL.md。来源不一致时停止并说明差异。核对通过后,用 Skill Finder 安装目标 Skill,并报告安装路径和来源版本。
此命令从 GitHub 来源仓库获取指定技能。运行前请先查看源文件。
npx skills add https://github.com/brycewang-stanford/Auto-Empirical-Research-Skills/tree/bfa0a6e5a843babbf45a1ef1663fd5dc4f07c0ba/skills/43-wentorai-research-plugins/skills/research/automation/rd-agent-guide --skill rd-agent-guide来源与配置
查看 GitHub 源码brycewang-stanford/Auto-Empirical-Research-Skills固定提交 bfa0a6e5a843babbf45a1ef1663fd5dc4f07c0ba
配置方式待核实
请查看原始来源中的适用工具、依赖和权限要求。
安装实测待补充
本条目面向搭 Agent 与 AI 应用场景,提供“rd-agent-guide”技能的目录入口。公开标签:learning、AI研发、自动化实验、模型开发。请通过固定上游版本核对实际功能、依赖与权限;此页为 Skillπ 独立编写的用途介绍和原站链接,不代表运行安全保证。原文与附件依各自许可开放。
文件浏览
---
name: rd-agent-guide
description: "Microsoft AI-driven R&D agent for automated data and model development"
metadata:
openclaw:
emoji: "🤖"
category: "research"
subcategory: "automation"
keywords: ["r-and-d", "microsoft", "automation", "model-development", "data-science", "experiment-automation"]
source: "https://github.com/microsoft/RD-Agent"
---
# RD-Agent Guide
## Overview
RD-Agent is an open-source AI-powered research and development automation framework developed by Microsoft Research, with over 12,000 stars on GitHub. It automates key steps in the R&D lifecycle -- including hypothesis generation, experiment design, code implementation, and result analysis -- enabling researchers and data scientists to accelerate their development cycles significantly.
The framework implements a closed-loop R&D automation pipeline where an AI agent iteratively proposes hypotheses, implements experiments, evaluates results, and refines its approach based on feedback. This mirrors the scientific method but operates at machine speed, allowing researchers to explore a much larger space of ideas and configurations than would be feasible manually.
RD-Agent is particularly valuable for researchers working in quantitative finance, data science, and machine learning, where the development process involves iterating on feature engineering, model architectures, and hyperparameter configurations. The framework has demonstrated the ability to autonomously develop competitive machine learning models and trading strategies, achieving results comparable to experienced human practitioners.
## Installation and Setup
```bash
# Clone the repository
git clone https://github.com/microsoft/RD-Agent.git
cd RD-Agent
# Install dependencies
pip install -e .
# Or install from PyPI
pip install rdagent
```
### Environment Configuration
```bash
# LLM configuration (required)
export OPENAI_API_KEY=$OPENAI_API_KEY
export CHAT_MODEL=gpt-4o
# Or use Azure OpenAI
export AZURE_OPENAI_API_KEY=$AZURE_OPENAI_API_KEY
export AZURE_OPENAI_ENDPOINT=$AZURE_OPENAI_ENDPOINT
export AZURE_OPENAI_DEPLOYMENT=$AZURE_OPENAI_DEPLOYMENT
# Docker is required for sandboxed code execution
# Ensure Docker is installed and running
docker --version
```
RD-Agent uses Docker containers to execute generated code safely, ensuring that automatically generated experiments cannot affect the host system. This sandboxed execution is critical for an autonomous agent that writes and runs arbitrary code.
## Core Concepts
### The R&D Loop
RD-Agent implements a continuous improvement loop with four phases:
1. **Proposal**: The agent analyzes the current state and proposes new hypotheses or improvements
2. **Implementation**: Hypotheses are translated into executable code (feature engineering, model changes, etc.)
3. **Evaluation**: The implemented changes are executed in a sandbox and results are measured against defined metrics
4. **Feedback**: Results are analyzed and used to inform the next round of proposals
```python
from rdagent.core.runner import RDRunner
from rdagent.scenarios.data_science import DataScienceScenario
# Define the research scenario
scenario = DataScienceScenario(
task="tabular_classification",
dataset_path="path/to/dataset.csv",
target_column="label",
metric="auc",
)
# Create and run the R&D agent
runner = RDRunner(
scenario=scenario,
max_iterations=50,
llm_model="gpt-4o",
)
# Start the autonomous R&D loop
results = runner.run()
# Review the best solution found
print(f"Best metric: {results.best_score}")
print(f"Iterations: {results.total_iterations}")
print(f"Solutions explored: {results.num_solutions}")
```
### Scenario Types
RD-Agent supports multiple R&D scenarios out of the box:
#### Data Science / Kaggle Competitions
Automatically engineer features, select models, and tune hyperparameters for tabular data tasks:
```python
from rdagent.scenarios.data_science import DataScienceScenario
scenario = DataScienceScenario(
task="tabular_regression",
dataset_path="data/housing.csv",
target_column="price",
metric="rmse",
time_budget_hours=4,
)
```
#### Quantitative Finance
Develop and backtest trading factors and strategies:
```python
from rdagent.scenarios.qlib import QlibScenario
scenario = QlibScenario(
market="csi300",
task="alpha_factor_mining",
backtest_start="2020-01-01",
backtest_end="2024-12-31",
metric="information_coefficient",
)
```
#### Model Development
Iterate on model architectures and training procedures:
```python
from rdagent.scenarios.model_dev import ModelDevScenario
scenario = ModelDevScenario(
task="image_classification",
base_model="resnet50",
dataset="cifar100",
optimization_target="accuracy",
)
```
## Advanced Features
### Experiment Tracking and Analysis
RD-Agent maintains detailed logs of all experiments, enabling post-hoc analysis of the R&D process:
```python
# Access experiment history
for experiment in results.history:
print(f"Iteration {experiment.iteration}:")
print(f" Hypothesis: {experiment.hypothesis}")
print(f" Changes: {experiment.code_changes}")
print(f" Metric: {experiment.score}")
print(f" Analysis: {experiment.feedback}")
```
### Custom Evaluation Functions
Define custom evaluation metrics for domain-specific research:
```python
from rdagent.core.evaluation import EvaluationFunction
class CustomMetric(EvaluationFunction):
def evaluate(self, predictions, ground_truth, **kwargs):
# Your custom metric computation
score = compute_domain_specific_metric(predictions, ground_truth)
return {
"primary_metric": score,
"secondary_metrics": {
"precision": compute_precision(predictions, ground_truth),
"recall": compute_recall(predictions, ground_truth),
}
}
scenario = DataScienceScenario(
evaluation_function=CustomMetric(),
# ... other config
)
```
### Human-in-the-Loop Mode
Guide the agent with human feedback at key decision points:
```python
runner = RDRunner(
scenario=scenario,
human_in_the_loop=True,
review_frequency=5, # Review every 5 iterations
)
# The agent will pause for human review at specified intervals
# You can approve, reject, or modify proposed experiments
```
## Research Applications
### Ablation Studies at Scale
Use RD-Agent to systematically explore which components contribute most to model performance:
```python
# Define ablation study
ablation_config = {
"base_model": "your_full_model",
"components_to_ablate": [
"attention_mechanism",
"residual_connections",
"layer_normalization",
"data_augmentation",
],
"metric": "accuracy",
"num_seeds": 5, # Run each configuration with 5 seeds
}
```
### Automated Feature Engineering
Let the agent discover and implement novel features for your dataset:
```python
scenario = DataScienceScenario(
task="feature_engineering",
dataset_path="data/research_data.csv",
existing_features=["feature_a", "feature_b", "feature_c"],
target="outcome",
max_new_features=20,
)
```
### Reproducibility
Every experiment run by RD-Agent is fully reproducible. The framework saves the complete experiment specification including code, data transformations, random seeds, and environment details, enabling other researchers to reproduce and build upon the results.
## References
- Repository: https://github.com/microsoft/RD-Agent
- Microsoft Research blog post: https://www.microsoft.com/en-us/research/project/rd-agent/
- Qlib quantitative platform: https://github.com/microsoft/qlib
- Documentation: https://microsoft.github.io/RD-Agent/
版本历史
-
v2026.07
GitHub ↗
原版更新说明
## Highlights - **First tagged release.** Every gate is green: `make check` (validate + 183 unit tests + eval-harness + numeric benchmarks), validate-catalog, quality-evals, and OpenSSF Scorecard on `main`. - **Rigor double loop at 13 method families.** Each family closes taxonomy tag → eval scenario (does the agent avoid the method's known traps?) → deterministic numeric benchmark (does it recover the true effect?). New this cycle: heterogeneous treatment effects (CATE) and quantile treatment effects (QTE). - **Paper-WorkFlow flagship at 29/29 executable rigor gates**, now merged to its `main` with a drift-gated README badge — the credibility artifact is checked by CI, so it can never silently go stale ([RIGOR.md](https://github.com/brycewang-stanford/Paper-WorkFlow/blob/main/RIGOR.md)). - **1,150 vendored skills / 69 collections**, all license-audited, security-scanned, tracked in a machine-readable catalog with CI-enforced freshness, and importable as a single whole-repo router skill. - **Hardened weekly upstream syncs**: synced trees must pass the full gate before merge; this cycle both sync PRs were verified, patched (StatsPAI's SkillOpt execution-gate card restored), and merged. - **First-party plugin marketplace** (Python / R / Stata empirical-analysis plugins + guide) shipped in the previous cycle, now under a tagged version. ## Known Follow-Ups - Ecosystem benchmark comparison (AERS vs econometrics agents on the shared benchmark suite) — [PLAN-2026-07 week 3](https://github.com/brycewang-stanford/Auto-Empirical-Research-Skills/blob/main/docs/PLAN-2026-07.md). - Decomposition (Oaxaca–Blinder) and shift-share (Bartik) method families (weeks 3–4). - GitHub Pages catalog site, README demo screenshots, five-language README consistency script (week 4). ## Catalog - Top-level collections: **69** - `SKILL.md` files: **1150** - Cataloged agent tools: **335** ([`tools/CATALOG.md`](../tools/CATALOG.md)) ## Methodological rigor - Method families in the coverage map: **13** ([`RIGOR_COVERAGE.md`](RIGOR_COVERAGE.md)) - Eval scenarios: **26** (138 rubric items, 122 auto-checkable) ([`../eval-harness/`](../eval-harness/README.md)) - Benchmark tasks: **13** (70 gold checks, 58 required) ([`../benchmark/`](../benchmark/README.md)) ### Method families (skills tagged) - `iv`: 27 - `rdd`: 25 - `panel-fe`: 19 - `staggered-did`: 17 - `bayesian`: 13 - `did`: 10 - `event-study`: 10 - `matching`: 10 - `synthetic-control`: 10 - `cate`: 8 - `dml`: 8 - `survival`: 3 - `quantile`: 2 ### Eval scenarios by category - `causal-identification`: 14 - `citation-hygiene`: 1 - `reproducibility`: 2 - `research-integrity`: 1 - `runtime-safety`: 1 - `statistical-validity`: 2 - `writing-compliance`: 3 - `writing-style`: 2 ## License buckets - AGPL-3.0: 1 - Apache-2.0: 1 - CC-BY-NC-4.0 (non-commercial): 1 - CC-BY-SA-4.0: 1 - CC-BY-SA-4.0 (repository default): 4 - GPL-3.0: 1 - MIT: 35 - MIT Non-Commercial: 1 - Mixed (per-skill; see folder): 2 - UNKNOWN - check upstream: 22 ## Quality gates - `make check` — catalog/provenance/audit/eval freshness, repo hygiene, Paper-WorkFlow gates, stdlib unit tests, eval-harness lint + graded smoke, and the strict numeric benchmark (reference-gold drift fails the build). - External links and OpenSSF Scorecard run on their scheduled workflows.
当前收录版本
0.0.1
质量评测
满分 100。前四项来自 DeepSkill 收录的质量报告。
评分依据
详细介绍微软 RD-Agent 开源框架,提供自动化研发流程的完整解决方案
评分依据
关键词覆盖全面,包含 r-and-d、automation、model-development 等核心术语
评分依据
文档结构清晰,包含概述、安装、核心概念、高级功能等完整章节
评分依据
提供完整的安装命令和环境配置说明,包含 OpenAI 和 Azure OpenAI 两种配置方式
计算方式
取前四项原始分值的算术平均,四舍五入取整。可通过收录报告核对各项评分与说明,运行记录需单独查证。
权限与使用前检查
根据当前主文档自动整理,命中内容可能是示例或防护说明;不是实际权限声明,也未覆盖附件、外部依赖或运行行为。
-
安装或获取外部依赖
使用官方来源并锁定版本,在隔离环境检查安装脚本。当前文档检查不覆盖后续下载的代码。
主文档提及 · 第 31, 34 行
文档引用的域名或地址
这些地址可能用于文档链接或示例,不是已批准的联网清单。使用前核对实际目的地及发送的数据。
- github.com
- microsoft.github.io
- www.microsoft.com
安全评级与访问范围
risk · 静态证据核验,不代表运行安全保证。
使用前请由宿主核对并授权实际文件读写、外部服务和联网目的地;本站的核验不会授予任何运行权限。
用户讨论
还没有已核实的点评。