# Begin here with the core Evolutionary search, then get down to structures

# Path: src/evolution.py

# Game Plan
# 1. Take textbook data for the sales agent
# 2. Sales agent evolves to be better at selling
# 3. Customer needs to evolve to be better at judging the sales agent and decide against buying
# then we see what happens...

# I guess we will just focus on various iteration methodologies here
# the dataset will be used in interesting ways only

# Parse the sales_textbook.txt data || use each subpoint as a Tip Unit
file_path = '../data/open-source/sales_textbook.txt'
with open(file_path, 'r') as f:
    sales_textbook = f.read()
# crude filtering
sales_tips = []
for subpoint in sales_textbook.split('Subpoint:'):
    sales_tips += [p for p in subpoint.split('\n') if len(p)>400]

# print('Tip 1: ', sales_tips[1])
# print('Tip 2: ', sales_tips[2])

# from conversation import SalesSimulator

# only evolve the sales agent, the customer will require extra computation & selection
# --- basically the most hard-to-sell customer responses survive and form a DPO AIFT
# --- the sales agent will select the best prompts to use against the customer

import numpy as np
tip_indices = np.random.choice(len(sales_tips), 5, replace=False)
tips = 'Reference the following tips: \n <tip> \n' + '\n'.join([sales_tips[i] for i in tip_indices]) + '\n' + '</tip>'
print(tips)

# equip sales agent with tips