perfspec-learning/install.md

PerfSPEC Learning Phase INSTALLATION

By follwing thesed steps you can easy run PerfSPEC notebooks

TOC

Python

[!CAUTION] Due to TensorFlow requiremens Python 3.12 is the latest compatible

For Linux

I end up by downloading Python soruce and compile

For MacOS

Use Homebrew

brew install python@3.12 python3-pip

After install Python with *pip and virtualenv
Install TensorFlow requiremens

pip install tensorflow

Package manager

Use uv

An extremely fast Python package and project manager, written in Rust.

MacOS and Linux

sudo curl -LsSf https://astral.sh/uv/install.sh | sh

To add $HOME/.local/bin to your PATH, either restart your shell or run:

source $HOME/.local/bin/env (sh, bash, zsh)
source $HOME/.local/bin/env.fish (fish)

Notebook tool

Rather than use Jupyter this project use Marimo for a lot of good reasons !

Marimo

python -m venv marimo-env
source marimo-env/bin/activate
uv pip install marimo

Check with marimo tutorial intro --no-token

A web server should be run at http://localhost:2718 with a token if not use --no-token option for a non secure session

Marimo configuration is in $HOME/.config/marimo.toml

PerfSPEC Learning Phase

1 - git clone this repository

git clone https://repo.jesusperez.pro/jesus/perfspec-learning.git
cd perfspec-learning

2 - Make structures

For data use data_sample.tar.xc

tar xvJf data_sample.tar.xz 

For trained models use models_sample.tar.xz

tar xvJf models_sample.tar.xz 

Raw log file raw-audit-logs.log.xz is included in data directory

3 - Run notebook tool like this

Inside perfspec-learning path

cd  learning/python 
uvx marimo edit   --no-token   --watch

Connect with a browser to http://localhost:2718

Start with run_perfspec.py

[!CAUTION] A red box will appear at top left corner prompting to install required packages for the loaded notebook, in console where localhost:2718 appers listening uv will do the magic by installing them using pip

Note each notebook can import some extra packages like train_perfspec.py

In any case some notebook import content from others.

[!Tip] When all notebooks have being loaded it will be posible also to use them as a Python script by run it in a terminal o console like this:

python model_perfspec.py -- --help 

There is also an static html mode to be used, more info: [script mode][https://docs.marimo.io/guides/scripts/]

Some Python packages could be not detected and will require manual installation via command-line, simply for example, use:

uv pip install scikit-learn
uv pip install prettytable