File size: 1,406 Bytes
ee657a1 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 | # 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
|