nanoresearch-experiment

1# Experiment Skill
2 
3## Purpose
4Take the experiment blueprint and produce a runnable Python code skeleton that implements the proposed method, baselines, training loops, evaluation harness, and ablation configurations.
5 
6## Tools Required
7None. This skill operates entirely through LLM code generation based on the experiment blueprint.
8 
9## Input
10- `experiment_blueprint`: Path to `papers/experiment_blueprint.json` produced by the planning skill
11 
12## Process
131. Parse the experiment blueprint for datasets, baselines, metrics, and ablation groups
142. Generate the project directory structure (data loaders, models, training, evaluation, configs)
153. Produce data loading and preprocessing code for each specified dataset
164. Implement model architecture stubs for the proposed method and each baseline
175. Generate training loop with logging, checkpointing, and early stopping
186. Implement the evaluation harness computing all specified metrics
197. Create configuration files for each ablation group
208. Add a main entry point that accepts a config and runs the full train-evaluate pipeline
21 
22## Output
23Produces `experiments/` directory containing:
24- `data/`: Data loading and preprocessing modules
25- `models/`: Model architecture implementations (proposed method and baselines)
26- `training/`: Training loop and optimization utilities
27- `evaluation/`: Metric computation and result aggregation
28- `configs/`: YAML configuration files for each experiment and ablation variant
29…