grocery-cadence

Use scripts and LLMs to analyze grocery purchases

Grocery Purchase Pattern Analyzer

Things have been busy enough lately that I found myself wishing I had grocery auto-ordering set up. I wasn't sure on what cadences I buy things, but I have been keeping my grocery list in Apple Reminders for years. I add items there, then mark them Complete when I purchase the item.

This repo contains scripts to analyze grocery purchasing patterns from a structured csv with items and purchase date. It also has a script to extract this info from a Reminders list in the Apple's Reminders app. The analysis groups grocery items by similarity and calculates the cadence with which items are bought to recommend reorder schedules.

Setup and Requirements

1. AppleScript: Extracts completed items from the specified Reminders list (Mac only).
2. Bash Script: Runs AppleScript (extract from specified Reminders list name, save to csv, run analysis)
3. Python Script: Analyzes data, provides item grouping and cadence analysis.
4. Ollama: Used for the complex mode in item grouping (requires Gemma 2b model). Instructions to get set up here.

Install

git clone https://github.com/elliemdaw/grocery-cadence.git
cd grocery-cadence

(or fork a copy into your own repos)

Usage

Setup

Highly recommend using a python virtual environment (for all python things :) ). After you've created one and started it...

pip install requirements.txt
ollama serve
ollama run gemma2:2b

In the project folder: chmod +x extract_and_analyze.sh

If you're using the Apple Script, make sure your permissions are good to go: System Preferences > Security & Privacy > Automation make sure that iterminal/IDE is allowed access to the Reminders app.

Extract data and analyze in Simple or Complex mode

# Run extraction and analysis in simple mode
./extract_and_analyze.sh "<name of Reminders list>" "/path/to/grocery_data.csv" "simple"

# Run in complex mode (requires Ollama)
./extract_and_analyze.sh "<name of Reminders list>" "/path/to/grocery_data.csv" "complex"

Just use the analysis script

Use this if you aren't on a Mac or already have your grocery data.

The analysis script expects a csv with one header row and the two columns: item, purchase datetime "%Y-%m-%dT%H:%M:%S"

# Simple Mode
python3 analyze_grocery_categories.py --mode simple --input "/path/to/grocery_data.csv"

# Complex Mode
python3 analyze_grocery_categories.py --mode complex --input "/path/to/grocery_data.csv"

Assumptions and Limitations

• On my machine, the complex method takes about 20 seconds to run (just using the python script)
• The Apple Script is extremely slow, to speed it up it grabs reminders for the prior 365 days
• Bash script calls the Apple Script, so therefore the bash script won't work on other platforms.
• Python script assumes a csv with one header row and two data columns (in order): item, purchase date formatted as "%Y-%m-%dT%H:%M:%S"
• The complex grouping requires Gemma2 to return JSON (I haven't had issues so far... knock on wood)
• This runs locally using Ollama, it isn't sending your grocery data to a cloud LLM

Additional improvements TODO

• Better parsing in case of invalid JSON from LLM
• Configurability of model
• Source grocery data from elsewhere
• Speed up the Apple Script...