easydoe

中文文档 | A minimal CLI for generating design-of-experiments configs via Latin Hypercube Sampling.

Install

pip install -e .

Quick Start

doe takes two things: a config file and an output file.

# Basics
doe <config.yaml> -o <output.csv>

# Use the built-in template
doe src/easydoe/templates/default.yaml -o experiments.csv

# Or write your own config
doe my_config.yaml -o runs/batch_001.csv --seed 42

Config Format

n_samples: 30
seed: 42
output: experiments.csv

params:
  lr:
    type: log
    low: 1.0e-5
    high: 1.0e-2

  dropout:
    type: linear
    low: 0.1
    high: 0.5

  batch_size:
    type: choice
    values: [16, 32, 64]

Parameter Types

Type	Fields	Example
linear	low, high	Continuous uniform
log	low, high	Log-uniform (for lr, wd)
int	low, high	Integer uniform
choice	values	Discrete options

Output File

Format auto-detected by extension:

doe cfg.yaml -o experiments.csv    # CSV
doe cfg.yaml -o experiments.yaml   # YAML
doe cfg.yaml -o experiments.json   # JSON

Full Workflow Example

• Step 1: Write transformer_tune.yaml

n_samples: 30
seed: 2024

params:
  lr:
    type: log
    low: 1e-5
    high: 5e-4
  n_layers:
    type: int
    low: 2
    high: 8
  hidden_dim:
    type: choice
    values: [256, 512, 768]
  dropout:
    type: linear
    low: 0.0
    high: 0.3

• Step 2: Generate

doe transformer_tune.yaml -o runs/batch_1.csv

• Step 3: Train

import csv

with open('runs/batch_1.csv') as f:
    for row in csv.DictReader(f):
        lr = float(row['lr'])
        # ... train

CLI Options

Flag	Purpose	Example
--seed N	Fix random seed	doe cfg.yaml --seed 42
--dry-run	Preview without saving	doe cfg.yaml --dry-run
-o path	Override output path	doe cfg.yaml -o /tmp/test.csv