ProgramDBFlowModule/utils.py

""" This code is an implementation of Funsearch (https://www.nature.com/articles/s41586-023-06924-6) and is heavily inspired by the original code (https://github.com/google-deepmind/funsearch)

    **Citation**:
    
    @Article{FunSearch2023,
        author  = {Romera-Paredes, Bernardino and Barekatain, Mohammadamin and Novikov, Alexander and Balog, Matej and Kumar, M. Pawan and Dupont, Emilien and Ruiz, Francisco J. R. and Ellenberg, Jordan and Wang, Pengming and Fawzi, Omar and Kohli, Pushmeet and Fawzi, Alhussein},
        journal = {Nature},
        title   = {Mathematical discoveries from program search with large language models},
        year    = {2023},
        doi     = {10.1038/s41586-023-06924-6}
    }
"""

import numpy as np
import scipy
def add_gaussian_noise(array,mean=0,std=1):
    return array + np.random.normal(mean,std,array.shape)

def get_versioned_function_name(function_name,i):
    pass

def rename_function(implementation,function_name):
    pass

def string_concatenation(strings,split_char='\n\n'):
    pass

def make_header_like(implementation,function_name):
    pass

def rename_function_calls(implementation,function_name):
    pass

def extract_template_from_program(program):
    pass


def _softmax(logits: np.ndarray, temperature: float, epsilon = 1e-6) -> np.ndarray:
    """Returns the tempered softmax of 1D finite `logits`."""
    if not np.all(np.isfinite(logits)):
        non_finites = set(logits[~np.isfinite(logits)])
        raise ValueError(f'`logits` contains non-finite value(s): {non_finites}')
    if not np.issubdtype(logits.dtype, np.floating):
        logits = np.array(logits, dtype=np.float32)

    result = scipy.special.softmax(logits / temperature, axis=-1)
    
    #Non zero mass to prevent zero probabilities
    result += epsilon  # Add epsilon to prevent zeros
    result /= np.sum(result, axis=-1, keepdims=True)  # Normalize

    # Ensure that probabilities sum to 1 to prevent error in `np.random.choice`.
    index = np.argmax(result)
    result[index] = 1 - np.sum(result[0:index]) - np.sum(result[index+1:])
    return result