https://github.com/peteryushunli/fantasy-football-weekly-email
leverage ESPN and Sleeper APIs, and GPT-4 to automate weekly emails to your fantasy league
https://github.com/peteryushunli/fantasy-football-weekly-email
Last synced: 4 months ago
JSON representation
leverage ESPN and Sleeper APIs, and GPT-4 to automate weekly emails to your fantasy league
- Host: GitHub
- URL: https://github.com/peteryushunli/fantasy-football-weekly-email
- Owner: peteryushunli
- Created: 2023-09-16T21:53:16.000Z (over 1 year ago)
- Default Branch: main
- Last Pushed: 2023-10-30T02:15:10.000Z (over 1 year ago)
- Last Synced: 2024-08-14T07:07:39.911Z (8 months ago)
- Language: Python
- Size: 3.23 MB
- Stars: 6
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
- jimsghstars - peteryushunli/fantasy-football-weekly-email - leverage ESPN and Sleeper APIs, and GPT-4 to automate weekly emails to your fantasy league (Python)
README
# fantasy-football-weekly-summary-email
Python functions and notebooks that sends automate weekly emails to your fantasy league for ESPN and Sleeper Leagues
## Overview
This project contains Python scripts and a Jupyter notebook that sends automated weekly emails to for ESPN and Sleeper Fantasy Football leagues. The main functionality is encapsulated in the espn_functions.py and sleeper.py files, which contains various functions for interacting with Sleeper and ESPN's Fantasy Football API, as well as OpenAI to for email generation. These functions are then used in a Jupyter notebook (espn_notebook.ipynb) to run specific analysis tasks, such as generating weekly summaries and matchups.
# Requirements & Setup
1. Custom Packages
- [OpenAI](https://github.com/openai/openai-python)
- [Sleeper API Wrapper](https://github.com/SwapnikKatkoori/sleeper-api-wrapper/blob/master/README.md)
- [Tabulate](https://pypi.org/project/tabulate/)
- [Timeout Decorator](https://pypi.org/project/timeout-decorator/)
2. The current code requires the following information to be filled into the **config.json**, which is broken out in the following sections
- Sleeper (do not fill out if you are playing in a ESPN league)
- [league_id](https://support.sleeper.com/en/articles/4121798-how-do-i-find-my-league-id)
- owner_dict: maps the [owner_id](https://docs.sleeper.com/#getting-users-in-a-league) to the name of the team owner (I created this b/c our user and display names on Sleeper)
- division_dict (optional and only applicable to leagues with divisions): maps owner names with division numbers. this dict is only used in customized playoff seeding
- ESPN (vice versa for players on Sleeper leagues)
- espn_s2: credentials for login. here is a [helpful YouTube video that shows how to obtain this and the swid](https://www.youtube.com/watch?v=tNcND9lVycA)
- swid": see video above
- league_id: can be found in your ESPN league URL
- owner_dict: similar to the owner_dict in Sleeper, this maps the teamID to the name of the owner
- OpenAi
- API key: [Requires a registered OpenAI account](https://openai.com/blog/openai-api)
- gmail
- GMAIL_USER: your gmail address (if you are not using gmail, the SMTP code might require edits)
- GMAIL_PW: this is your [Gmail App password](https://support.google.com/accounts/answer/185833?hl=en) (not your regular login password)
# Code Process
The code runs in 3 main parts: **Data Ingestion, Summary Generation and Email Sending**
For [ESPN leagues, run the espn_notebook](https://github.com/peteryushunli/fantasy-football-weekly-email/blob/main/espn_notebook.ipynb), and [for Sleeper leagues, run the sleeper_notebook](https://github.com/peteryushunli/fantasy-football-weekly-email/blob/main/sleeper_notebook.ipynb).
*Also both the espn_functions.py and sleeper_functions.py can be exectutable python scripts, but this has not been tested extensively.*
## 1A. ESPN Data Pull
The full ingestion and data manipulation can be run as:
~~~
from espn_functions import *
standings_df, matchup_df, HP_Owner, HP_Player, HP_Score, HT_Owner, HT_Score = run_espn_weekly(week = X, year = 2023)
~~~
The outputs for this function will all be used for the summary generation or included in the body of the email and they are:
- Standings Table
- Weekly Matchup and Performance Table
- Highest Scoring Player, Owner and Score
- Highest Scoring Team and Score
The process consist of the following main functions:
- **load_league**: loads primary league data JSON from ESPN's API
- **load_records**: extracts the team records from the league data. maps the owner to the owner_dict (from the config json)
- **load_schedule**: extracts team schedule
- **load_weekly_stats**: loads each team's player's performances. also uses the owner_dict
- **transform_weekly**: pivots the data from the load_weekly_stats function
- **iterate_weeks_espn**: loads weekly data since beginning of the season to record weekly scores
- **rank_playoff_seeds**: custom function to calculate playoff seeds based on my league settings (edit or comment out the function from the run_espn_weekly function for your own league's rules and settings)
## 1B. Sleeper Ingestion
Full ingestion and data manipulation can be run as:
~~~
from sleeper_functions import *
matchup_df, updated_standings, HT_Owner, HT_Score, HP_Owner, HP_Player, HP_Score = run_sleeper_weekly(week = 3)
~~~
Same outputs as that of the ESPN Section
The process consist of the following main functions:
- **owners**: loads owner_id and maps to owner name via owner_dict
- **player_info**: outputs a dict with player id and full name
- **weekly_matchup**: loads the weekly matchup data as a dataframe
- **rank_playoff_seeds**: custom function to calculate playoff seeds that also maps to the division dict from the config. (edit or comment out the function from the run_espn_weekly function for your own league's rules and settings)
- **iterate_weeks**: loads weekly data since beginning of the season to record weekly scores
## 2. Summary Generation
~~~
summary = generate_summary(week, matchup_df, standings_df, model = 'gpt-4')
~~~
Here I prefer to use the GPT-4 model for its superior comprehension and generation capabilities, as well as its longer maximum token length. The next best alternative is GPT-3.5-turbo (or ChatGPT), but depending on your prompt and response length, you might need to go up to GPT-3.5-turbo-16k for a longer token length.
Also the text prompt can edit in the **instruction_prompt** variable found in the respective .py function files.
## 3. Email Sending
~~~
send_email(user, week, summary, updated_standings.to_html(), HT_Owner, HT_Score, HP_Owner, HP_Player, HP_Score)
~~~