Training #5

train_64n_1000e

实验目录 train_64n_1000e · 创建者 root · 2026-06-12 13:25

已完成

第 1000 / 1000 轮 100.0%

平均最终收益0.03805697336792946

平均奖励-0.08103617280721664

平均交易次数85.1875

轨迹计算耗时243.77838503103703

Live Metrics

训练曲线

平均最终收益平均奖励

Config

训练配置

total_updates: 1000
trajectories_per_update: 64
num_workers: 1
torch_num_threads: 1
max_cached_stocks_per_worker: 0
max_segment_length_days: 180
max_stock_files: 0
checkpoint_interval: 10
max_history_checkpoints: 10
learning_rate: 0.00015
seed: 20260426
mlp_layers: 5
resume_training: True