[BUG]: Unable to exit early

[Ltypestar2scratch]

What happened?

When i pressed q it didn't show the exit and even when i pressed q then enter it dosent work it dosent exit early.
I don't want to wait for a lot of itterations to get results

Version

1.5.9

Operating System

Windows

Package Manager

pip

Interface

Script (i.e., python my_script.py)

Relevant log output

Extra Info

I am running PySR on a PGN file, I am trying to map my winrates to my move counts:
Here are my settings:
model = PySRRegressor(
populations=4,
niterations=100000000000,
binary_operators=["+", "-", "*", "/", "^", "mod"],
unary_operators=[
"abs", "sqrt", "sin", "cos", "tan", "floor"
],
model_selection="best",
loss="loss(prediction, target, weight) = weight * (prediction - target)^2",
)
Here is my entire script:
winr.py

[MilesCranmer]

What terminal are you using?

[Ltypestar2scratch]

I am using CMD with non-administrator privileges

[MilesCranmer]

Can you try it in a different terminal like powershell?

[Ltypestar2scratch]

I got the same result in PowerShell. When I ran winr.py I cannot exit early even when i spammed q or when i pressed q once then quickly hit enter

[MilesCranmer]

And IPython?

[Ltypestar2scratch]

I used CMD to open the IPython terminal and ran winr.py again, and the same thing happened again, no matter how much I pressed q or how quickly I spammed q then enter I can't exit early.

[MilesCranmer]

Can you run it inside IPython, rather than the script? i.e., copy-paste the python code into IPython and try it there.

[MilesCranmer]

Also can you please paste your script into a comment in triple backticks? If it's very long you can use the <details> ... </details> tag to toggle it. It's easier for me to read than uploading a file though.

[Ltypestar2scratch]

Can you run it inside IPython, rather than the script? i.e., copy-paste the python code into IPython and try it there.

Same thing. I cannot exit early.

[Ltypestar2scratch]

Also can you please paste your script into a comment in triple backticks? If it's very long you can use the <details> ... </details> tag to toggle it. It's easier for me to read than uploading a file though.

import pandas as pd
import numpy as np
from pysr import PySRRegressor
import sys
import os

PGN_FILE = r"ltypestar1_chess_archives/ltypestar1_all_games.pgn"
TARGET_USERNAME = "Ltypestar1" 

def parse_pgn_data(pgn_path, username):
    """
    Reads the PGN and extracts Game Length vs. Win/Loss.
    Returns a DataFrame.
    """
    print(f"Reading games from {pgn_path}...")
    data = []
    
    with open(pgn_path, "r", encoding="utf-8") as pgn:
        while True:
            game = chess.pgn.read_game(pgn)
            if game is None:
                break
            
            headers = game.headers
            
            if headers.get("White") == username:
                user_color = chess.WHITE
            elif headers.get("Black") == username:
                user_color = chess.BLACK
            else:
                continue

            result = headers.get("Result")
            is_win = 0.0
            if result == "1-0" and user_color == chess.WHITE:
                is_win = 1.0
            elif result == "0-1" and user_color == chess.BLACK:
                is_win = 1.0
            
            board = game.end().board()
            num_moves = board.fullmove_number
            
            data.append({"moves": num_moves, "win": is_win})

    return pd.DataFrame(data)

def main():
    if not os.path.exists(PGN_FILE):
        print(f"I can't find the file at {PGN_FILE}")
        return

    # 1. Load the data
    df_raw = parse_pgn_data(PGN_FILE, TARGET_USERNAME)
    print(f"Parsed {len(df_raw)} games!")

    df_agg = df_raw.groupby("moves").agg(
        win_rate=("win", "mean"),
        count=("win", "count")
    ).reset_index()

    df_agg = df_agg[df_agg["count"] >= 3]

    X = df_agg[["moves"]].to_numpy()
    y = df_agg["win_rate"].to_numpy()
    weights = np.ones_like(y)

    print("Training the Symbolic Regressor...")

    # 3. Configure PySR
    model = PySRRegressor(
        populations=4,
        niterations=100000000000,
        binary_operators=["+", "-", "*", "/", "^", "mod"],
        unary_operators=[
            "abs", "sqrt", "sin", "cos", "tan", "floor"
        ],
        model_selection="best",
        loss="loss(prediction, target, weight) = weight * (prediction - target)^2",
    )
    model.fit(X, y, weights=weights)

    print("\nHere is the best one:")
    print(model.get_best().equation)

if __name__ == "__main__":
    main()

[MilesCranmer]

Thanks!

Ok it seems like a bug. I'm not sure why though.

If you are comfortable at all with Julia (or maybe ask an LLM to translate Python->Julia) then you could try changing the input_stream to something else. The default is stdin which is supposed to capture inputs, but I have no clue why it's not here. I would have expected input_stream=devnull to result in your behavior, but clearly you are using the default which is stdin.

I would recommend installing juliacall (key PySR dependency) and seeing if you can capture stdin from Python at all while using it.

[MilesCranmer]

I might actually have a guess for what it is:

In the code here:

"""Check if the user typed 'q' and <enter> or <ctl-c>."""
function check_for_user_quit(reader::StdinReader)::Bool
    if reader.can_read_user_input
        bytes = bytesavailable(reader.stream)
        if bytes > 0
            # Read:
            data = read(reader.stream, bytes)
            control_c = 0x03
            quit = 0x71
            if length(data) > 1 && (data[end] == control_c || data[end - 1] == quit)
                return true
            end
        end
    end
    return false
end

We are currently checking for a quit with

data[end - 1] == quit

This is valid on all operating systems that only send a single newline character after the "q". But I guess what might be happening is Windows is sending a \r\n rather than just a \n. And therefore on Windows, this should be data[end - 2] == quit. Just a guess though.

[Ltypestar2scratch]

I have a theory:
It may be because of multiprocessing
IDK why but I have a generally weird vibe that this is part of the solution