vacmar01

import numpy as np
from collections import defaultdict

def get_cell_bounds(idx): 
    x, y = idx; return ((x//3)*3, (x//3)*3+3), ((y//3)*3, (y//3)*3+3)

def get_used_numbers(grid, idx):
    x, y = idx; row, col = grid[x, :], grid[:, y]
    (x1,x2), (y1,y2) = get_cell_bounds(idx); box = grid[x1:x2, y1:y2].flatten()
    return np.unique(np.concatenate([row, col, box])[np.concatenate([row, col, box]) != 0])

vacmar01

# Fast Sudoku solver: bitmask constraints + MRV + forward-checking
# Drop-in replacement: solve(puzzle_str)->81-char string

# --- Precompute structures (indices 0..80) ---
ROWS = [list(range(r * 9, r * 9 + 9)) for r in range(9)]
COLS = [list(range(c, 81, 9)) for c in range(9)]
BOXS = []
for br in range(0, 9, 3):
    for bc in range(0, 9, 3):
        box = []

vacmar01

# /// script
# dependencies = [
#   "dspy",
#   "rich"
# ]
# ///

import dspy
import os
import inspect

vacmar01

➜  random_experiments uv run agent.py
Reading inline script metadata from `agent.py`
Enter your message or type 'exit' to quit.
👨‍💻User: what is the biggest file in the current folder?
Reasoning: To find the biggest file in the current folder, I first need to list the files
in the current directory. Once I have the list, I can inspect each file's size and
determine which one is the largest. The `list_files` function will give me the names of the
files, and then I can use `run_cmd` with a command like `ls -l` or `du -h` to get their
sizes.
Selected Function: list_files

vacmar01

from fasthtml.common import (
    fast_app,
    serve,
    Main,
    H1,
    Article,
    Div,
    Form,
    Textarea,
    NotStr,

vacmar01

from fasthtml.common import *
from dataclasses import make_dataclass, fields
import datetime
from pathlib import Path

import yaml

def create_dataclass_from_schema(schema_dict, cls_name="DynamicClass"):
    str2type = {
        "str": str,

vacmar01

Allgemeine Tips

• GPU in Google Colab einschalten: "Laufzeit -> Laufzeittyp ändern -> Hardwarebeschleuniger -> T4 GPU"

Links

• https://github.com/fastai/fastbook
• https://course.fast.ai/Lessons/lesson1.html
• https://github.com/fastai/fastbook/blob/master/01_intro.ipynb (https://colab.research.google.com/github/fastai/fastbook/blob/master/01_intro.ipynb)
• https://github.com/fastai/fastbook/blob/master/04_mnist_basics.ipynb (https://colab.research.google.com/github/fastai/fastbook/blob/master/04_mnist_basics.ipynb)
• https://docs.fast.ai/tutorial.datablock.html
• -https://www.kaggle.com/code/abdalimran/intuition-of-gradient-descent-for-machine-learning

vacmar01

How to Install Nginx, MySQL and PHP on an Ubuntu Droplet from DigitalOcean

This is meant to be as a reminder to myself and as a quick and dirty reference to others (heavily inspired by: https://www.techalyst.com/posts/laravel-hosting-with-digital-ocean-droplet-step-by-step-tutorial). It assumes you want to install a Laravel app afterwards, but it should be easily adaptable to other PHP apps.

I know that there are probably much more things you could config in your nginx (and maybe you should), but this works for now for me. I might update this with new things I learn.

1. Create DigitalOcean droplet

Choose the region that is nearest to you.

vacmar01

class Interpreter:
    def __init__(self, model, dl):
        self.model = model
        self.dl = dl
        if hasattr(model, "loss"):
            self.loss_func = self.model.loss
        self.losses = torch.empty(0)
        self.model.eval()
        for batch in tqdm.tqdm(self.dl):
            x, y = batch

vacmar01

Super Kurs und super (online kostenlos zu lesendes) Buch. Sehr anschaulich erklärt.

• https://www.fast.ai/
• https://github.com/fastai/fastbook

Sehr aufwendig produzierter Kurs vom MIT zum Thema Deep Learning mit mehreren fortgeschrittenen Themen in späteren Videos

• http://introtodeeplearning.com/ (MIT Boston)

"Introduction to Deep Learning" Kurs der TU München

• https://niessner.github.io/I2DL/ (TU München)