import numpy as np
import json
from trueskill import TrueSkill
import paramiko
import io, os
import sys
sys.path.append('../')
from serve.constants import SSH_SERVER, SSH_PORT, SSH_USER, SSH_PASSWORD, SSH_SKILL


trueskill_env = TrueSkill()

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 create_ssh_client(server, port, user, password):
    ssh = paramiko.SSHClient()
    ssh.load_system_host_keys()
    ssh.set_missing_host_key_policy(paramiko.AutoAddPolicy())
    ssh.connect(server, port, user, password)
    return ssh

def save_json_via_sftp(ratings, comparison_counts, total_comparisons):
    ssh = create_ssh_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)
    sftp = ssh.open_sftp()
    with sftp.open(SSH_SKILL, 'w') as f:
        f.write(json_data)

def load_json_via_sftp():
    ssh = create_ssh_client(SSH_SERVER, SSH_PORT, SSH_USER, SSH_PASSWORD)
    sftp = ssh.open_sftp()
    with sftp.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


def matchmaker(num_players, k_group=4):
    trueskill_env = TrueSkill()
    
    ratings, comparison_counts, total_comparisons = load_json_via_sftp()

    # Randomly select a player
    selected_player = np.random.randint(0, num_players)

    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')  # minimize the score for the selected player to exclude it
    opponents = np.argsort(ucb_scores)[-k_group + 1:].tolist()

    # Group players
    model_ids = [selected_player] + opponents

    from serve.update_skill import Model_ID
    Model_ID.group = model_ids
        
    return model_ids