pyruns 0.0.4

Lightweight web UI for managing, running, and monitoring Python experiments.

Pyruns — Python Experiment Runner & Monitor

🧪 A lightweight web UI for managing, running, and monitoring Python experiments.

---

✨ Features

Feature	Description
Generator	Load YAML configs, edit parameters in a structured form or raw YAML editor, batch-generate tasks with product (|) and zip ((|)) syntax
Manager	Card-grid overview of all tasks with status filters, search, batch run/delete, adjustable columns
Monitor	Real-time ANSI-colored log viewer, task list with live status, export reports to CSV/JSON
System Metrics	Live CPU, RAM, and GPU summary (count × avg utilization) in the header
Auto Config Detection	pyr script.py detects argparse parameters or pyruns.read() calls automatically
Workspace Settings	Customise UI defaults (refresh intervals, grid columns, workers) via _pyruns_.yaml

📦 Installation

pip install pyruns

Dependencies

• Python ≥ 3.8
• NiceGUI — web UI framework
• PyYAML — YAML parsing
• psutil — system metrics
• nvidia-smi (optional) — GPU metrics

🚀 Quick Start

CLI Mode (recommended)

pyr your_script.py       # Launch UI for your script
pyr dev your_script.py   # Launch with hot-reload (for development)
pyr help                 # Show usage instructions

pyr will:

1. Detect parameters from your script (argparse or pyruns.read())
2. Generate _pyruns_/config_default.yaml (for argparse scripts)
3. Create _pyruns_/_pyruns_.yaml with editable UI defaults
4. Open the web UI at http://localhost:8080

In Your Script

import pyruns

# Under pyr — load() auto-reads the task config, no read() needed
config = pyruns.load()
print(config.lr, config.epochs)

# Record metrics for the Monitor page
for epoch in range(100):
    loss = train(config)
    pyruns.add_monitor(epoch=epoch, loss=loss)

When running standalone (python train.py), specify a config explicitly:

pyruns.read("path/to/config.yaml")   # explicit path
config = pyruns.load()               # then load as usual

⚙️ Workspace Settings

On first launch, pyr creates _pyruns_/_pyruns_.yaml:

ui_port: 8080                      # web UI port
header_refresh_interval: 3         # metrics refresh (seconds)
generator_form_columns: 2          # parameter editor columns
generator_auto_timestamp: true     # auto-name tasks with timestamp
manager_columns: 5                 # task card grid columns
manager_max_workers: 1             # parallel worker count
manager_execution_mode: thread     # thread | process
manager_poll_interval: 1           # Manager polling (seconds)
monitor_poll_interval: 1           # Monitor polling (seconds)

Edit this file to customise the UI for your workflow.

📋 Batch Syntax

# Product (cartesian): 3 × 2 = 6 combinations
lr: 0.001 | 0.01 | 0.1
batch_size: 32 | 64

# Zip (paired): lengths must match
seed: (1 | 2 | 3)
name: (exp_a | exp_b | exp_c)

📄 License

MIT

---

Pyruns — Python 实验管理与监控工具

🧪 一个轻量级 Web UI，用于管理、运行和监控 Python 实验。

---

✨ 功能特性

功能	说明
Generator	加载 YAML 配置，结构化表单 / 原始 YAML 编辑，支持 | 笛卡尔积和 (|) 配对批量生成
Manager	卡片网格管理任务，状态过滤、搜索、批量运行/删除
Monitor	实时 ANSI 彩色日志查看，任务状态监控，导出 CSV/JSON
系统指标	顶栏实时显示 CPU、RAM、GPU 概览（数量 × 平均利用率）
自动检测	pyr script.py 自动提取 argparse 参数或检测 pyruns.read()
工作区配置	通过 _pyruns_.yaml 自定义刷新间隔、网格列数、并行数等

📦 安装

pip install pyruns

🚀 快速开始

pyr your_script.py       # 启动 UI
pyr dev your_script.py   # 热加载模式（开发调试用）
pyr help                 # 查看使用说明

在脚本中使用

import pyruns

# pyr 模式下，load() 自动读取任务配置，无需手动 read()
config = pyruns.load()

# 记录训练指标（Monitor 页面可查看）
pyruns.add_monitor(epoch=1, loss=0.5, acc=92.3)

手动运行时（python train.py）：

pyruns.read("path/to/config.yaml")   # 指定配置路径
config = pyruns.load()

⚙️ 工作区配置

首次启动时自动生成 _pyruns_/_pyruns_.yaml，可编辑以自定义 UI 默认值：

ui_port: 8080                      # Web UI 端口
header_refresh_interval: 3         # 顶栏刷新间隔（秒）
generator_form_columns: 2          # 参数编辑器列数
generator_auto_timestamp: true     # 自动时间戳命名
manager_columns: 5                 # 任务卡片网格列数
manager_max_workers: 1             # 默认并行数
manager_execution_mode: thread     # thread | process
manager_poll_interval: 1           # Manager 轮询间隔（秒）
monitor_poll_interval: 1           # Monitor 轮询间隔（秒）

📋 批量生成语法

# 笛卡尔积：3 × 2 = 6 种组合
lr: 0.001 | 0.01 | 0.1
batch_size: 32 | 64

# 配对组合：长度必须一致
seed: (1 | 2 | 3)
name: (exp_a | exp_b | exp_c)

📄 开源协议

MIT