plot.py

import tensorflow as tf
import matplotlib.pyplot as plt
import pandas as pd

def plot_run(name, log):
    
    # Gather data
    reward = []
    run_len = []
    for e in tf.compat.v1.train.summary_iterator(log):
        for v in e.summary.value:
            if v.tag == 'rollout/ep_rew_mean':
                reward.append({
                    'step': e.step,
                    'value': v.simple_value
                })
            if v.tag == 'rollout/ep_len_mean':
                run_len.append({
                    'step': e.step,
                    'value': v.simple_value
                })
    reward_df = pd.DataFrame(reward)
    run_len_df = pd.DataFrame(run_len)

    plt.figure(figsize=(4,4))    
    plt.title('Mean episode length')
    plt.xlabel('step')
    plt.ylabel('length')
    plt.grid()
    plt.tight_layout()
    plt.plot(run_len_df.step, run_len_df.value)
    plt.savefig(f'plots/ppo_length_{name}.png')
    plt.clf()
    
    plt.figure(figsize=(4,4))
    plt.title('Mean episode reward')
    plt.xlabel('step')
    plt.ylabel('reward')
    plt.grid()
    plt.tight_layout()
    plt.plot(reward_df.step, reward_df.value)
    plt.savefig(f'plots/ppo_reward_{name}.png')


if __name__ == '__main__':

    plot_run('1m', 'logs\PPO_13\events.out.tfevents.1746853937.JOHN-LAPTOP.33828.0')
    plot_run('2m', 'logs\PPO_14\events.out.tfevents.1746855975.JOHN-LAPTOP.40200.0')