lambd: 0.95
learning_rate: 0.0003
learning_rate_schedule: linear
max_steps: 1e8
max_steps: 2e7
memory_size: 128
normalize: true
num_epoch: 3