K-Sort-Arena / model /matchmaker.py
YangZhoumill's picture
release code
a4b32da
raw
history blame contribute delete
No virus
4.19 kB
import numpy as np
import json
from trueskill import TrueSkill
import paramiko
import io, os
import sys
import random
sys.path.append('../')
from serve.constants import SSH_SERVER, SSH_PORT, SSH_USER, SSH_PASSWORD, SSH_SKILL
trueskill_env = TrueSkill()
ssh_matchmaker_client = None
sftp_matchmaker_client = None
def create_ssh_matchmaker_client(server, port, user, password):
global ssh_matchmaker_client, sftp_matchmaker_client
ssh_matchmaker_client = paramiko.SSHClient()
ssh_matchmaker_client.load_system_host_keys()
ssh_matchmaker_client.set_missing_host_key_policy(paramiko.AutoAddPolicy())
ssh_matchmaker_client.connect(server, port, user, password)
transport = ssh_matchmaker_client.get_transport()
transport.set_keepalive(60)
sftp_matchmaker_client = ssh_matchmaker_client.open_sftp()
def is_connected():
global ssh_matchmaker_client, sftp_matchmaker_client
if ssh_matchmaker_client is None or sftp_matchmaker_client is None:
return False
if not ssh_matchmaker_client.get_transport().is_active():
return False
try:
sftp_matchmaker_client.listdir('.')
except Exception as e:
print(f"Error checking SFTP connection: {e}")
return False
return True
def ucb_score(trueskill_diff, t, n):
exploration_term = np.sqrt((2 * np.log(t + 1e-5)) / (n + 1e-5))
ucb = -trueskill_diff + 1.0 * exploration_term
return ucb
def update_trueskill(ratings, ranks):
new_ratings = trueskill_env.rate(ratings, ranks)
return new_ratings
def serialize_rating(rating):
return {'mu': rating.mu, 'sigma': rating.sigma}
def deserialize_rating(rating_dict):
return trueskill_env.Rating(mu=rating_dict['mu'], sigma=rating_dict['sigma'])
def save_json_via_sftp(ratings, comparison_counts, total_comparisons):
global sftp_matchmaker_client
if not is_connected():
create_ssh_matchmaker_client(SSH_SERVER, SSH_PORT, SSH_USER, SSH_PASSWORD)
data = {
'ratings': [serialize_rating(r) for r in ratings],
'comparison_counts': comparison_counts.tolist(),
'total_comparisons': total_comparisons
}
json_data = json.dumps(data)
with sftp_matchmaker_client.open(SSH_SKILL, 'w') as f:
f.write(json_data)
def load_json_via_sftp():
global sftp_matchmaker_client
if not is_connected():
create_ssh_matchmaker_client(SSH_SERVER, SSH_PORT, SSH_USER, SSH_PASSWORD)
with sftp_matchmaker_client.open(SSH_SKILL, 'r') as f:
data = json.load(f)
ratings = [deserialize_rating(r) for r in data['ratings']]
comparison_counts = np.array(data['comparison_counts'])
total_comparisons = data['total_comparisons']
return ratings, comparison_counts, total_comparisons
class RunningPivot(object):
running_pivot = []
def matchmaker(num_players, k_group=4, not_run=[]):
trueskill_env = TrueSkill()
ratings, comparison_counts, total_comparisons = load_json_via_sftp()
ratings = ratings[:num_players]
comparison_counts = comparison_counts[:num_players, :num_players]
# Randomly select a player
# selected_player = np.random.randint(0, num_players)
comparison_counts[RunningPivot.running_pivot, :] = float('inf')
comparison_counts[not_run, :] = float('inf')
selected_player = np.argmin(comparison_counts.sum(axis=1))
RunningPivot.running_pivot.append(selected_player)
RunningPivot.running_pivot = RunningPivot.running_pivot[-5:]
print(RunningPivot.running_pivot)
selected_trueskill_score = trueskill_env.expose(ratings[selected_player])
trueskill_scores = np.array([trueskill_env.expose(p) for p in ratings])
trueskill_diff = np.abs(trueskill_scores - selected_trueskill_score)
n = comparison_counts[selected_player]
ucb_scores = ucb_score(trueskill_diff, total_comparisons, n)
# Exclude self, select opponent with highest UCB score
ucb_scores[selected_player] = -float('inf')
ucb_scores[not_run] = -float('inf')
opponents = np.argsort(ucb_scores)[-k_group + 1:].tolist()
# Group players
model_ids = [selected_player] + opponents
random.shuffle(model_ids)
return model_ids