Add sep_value

scripts/cleanba.py

@@ -8,7 +8,7 @@ from datetime import datetime, timedelta, timezone
 from collections import deque
 from dataclasses import dataclass, field, asdict
 from types import SimpleNamespace
-from typing import List, NamedTuple, Optional
+from typing import List, NamedTuple, Optional, Literal
 from functools import partial
 
 import ygoenv
@@ -28,7 +28,8 @@ from ygoai.rl.ckpt import ModelCheckpoint, sync_to_gcs, zip_files
 from ygoai.rl.jax.agent import RNNAgent, ModelArgs
 from ygoai.rl.jax.utils import RecordEpisodeStatistics, masked_normalize, categorical_sample
 from ygoai.rl.jax.eval import evaluate, battle
-from ygoai.rl.jax import clipped_surrogate_pg_loss, vtrace_2p0s, mse_loss, entropy_loss, simple_policy_loss, ach_loss, policy_gradient_loss
+from ygoai.rl.jax import clipped_surrogate_pg_loss, mse_loss, entropy_loss, simple_policy_loss, \
+    ach_loss, policy_gradient_loss, vtrace, vtrace_2p0s, truncated_gae
 from ygoai.rl.jax.switch import truncated_gae_2p0s as gae_2p0s_switch
 
 
@@ -116,6 +117,10 @@ class Args:
 
     upgo: bool = True
     """Toggle the use of UPGO for advantages"""
+    sep_value: bool = True
+    """Whether separate value function computation for each player"""
+    value: Literal["vtrace", "gae"] = "vtrace"
+    """the method to learn the value function"""
     gae_lambda: float = 0.95
     """the lambda for the general advantage estimation"""
     c_clip_min: float = 0.001
@@ -738,13 +743,23 @@ def main():
 
         # Advantages and target values
         if args.switch:
+            if args.value == "vtrace" or args.sep_value:
+                raise NotImplementedError
             target_values, advantages = gae_2p0s_switch(
                 next_value, new_values_, rewards, next_dones, switch_or_mains,
                 args.gamma, args.gae_lambda, args.upgo)
         else:
             # TODO: TD(lambda) for multi-step
             ratios_ = reshape_time_series(ratios)
-            target_values, advantages = vtrace_2p0s(
+            if args.value == "gae":
+                if not args.sep_value:
+                    raise NotImplementedError
+                target_values, advantages = truncated_gae(
+                    next_value, new_values_, rewards, next_dones, switch_or_mains,
+                    args.gamma, args.gae_lambda, args.upgo)
+            else:
+                vtrace_fn = vtrace if args.sep_value else vtrace_2p0s
+                target_values, advantages = vtrace_fn(
                     next_value, ratios_, new_values_, rewards, next_dones, switch_or_mains, args.gamma,
                     args.rho_clip_min, args.rho_clip_max, args.c_clip_min, args.c_clip_max)
 

ygoai/rl/jax/__init__.py

@@ -7,68 +7,6 @@ import chex
 import distrax
 
 
-# class VTraceOutput(NamedTuple):
-#     q_estimate: jnp.ndarray
-#     errors: jnp.ndarray
-
-
-# def vtrace(
-#     v_tm1,
-#     v_t,
-#     r_t,
-#     discount_t,
-#     rho_tm1,
-#     lambda_=1.0,
-#     c_clip_min: float = 0.001,
-#     c_clip_max: float = 1.007,
-#     rho_clip_min: float = 0.001,
-#     rho_clip_max: float = 1.007,
-#     stop_target_gradients: bool = True,
-# ):
-#     """
-
-#     Args:
-#       v_tm1: values at time t-1.
-#       v_t: values at time t.
-#       r_t: reward at time t.
-#       discount_t: discount at time t.
-#       rho_tm1: importance sampling ratios at time t-1.
-#       lambda_: mixing parameter; a scalar or a vector for timesteps t.
-#       clip_rho_threshold: clip threshold for importance weights.
-#       stop_target_gradients: whether or not to apply stop gradient to targets.
-#     """
-#     # Clip importance sampling ratios.
-#     lambda_ = jnp.ones_like(discount_t) * lambda_
-
-#     c_tm1 = jnp.clip(rho_tm1, c_clip_min, c_clip_max) * lambda_
-#     clipped_rhos_tm1 = jnp.clip(rho_tm1, rho_clip_min, rho_clip_max)
-
-#     # Compute the temporal difference errors.
-#     td_errors = clipped_rhos_tm1 * (r_t + discount_t * v_t - v_tm1)
-
-#     # Work backwards computing the td-errors.
-#     def _body(acc, xs):
-#         td_error, discount, c = xs
-#         acc = td_error + discount * c * acc
-#         return acc, acc
-
-#     _, errors = jax.lax.scan(
-#         _body, 0.0, (td_errors, discount_t, c_tm1), reverse=True)
-
-#     # Return errors, maybe disabling gradient flow through bootstrap targets.
-#     errors = jax.lax.select(
-#         stop_target_gradients,
-#         jax.lax.stop_gradient(errors + v_tm1) - v_tm1,
-#         errors)
-#     targets_tm1 = errors + v_tm1
-#     q_bootstrap = jnp.concatenate([
-#         lambda_[:-1] * targets_tm1[1:] + (1 - lambda_[:-1]) * v_tm1[1:],
-#         v_t[-1:],
-#     ], axis=0)
-#     q_estimate = r_t + discount_t * q_bootstrap
-#     return VTraceOutput(q_estimate=q_estimate, errors=errors)
-
-
 def entropy_loss(logits):
     return distrax.Softmax(logits=logits).entropy()
 
@@ -255,6 +193,57 @@ def vtrace_rnad(
     return targets, q_estimate
 
 
+def vtrace_loop(carry, inp, gamma, rho_min, rho_max, c_min, c_max):
+    v, next_value, last_return, next_q, next_main = carry
+    ratio, cur_value, next_done, reward, main = inp
+
+    v = jnp.where(next_done, 0, v)
+    next_value = jnp.where(next_done, 0, next_value)
+    
+    sign = jnp.where(main == next_main, 1, -1)
+    v = v * sign
+    next_value = next_value * sign
+
+    discount = gamma * (1.0 - next_done)
+
+    q_t = reward + discount * v
+
+    rho_t = jnp.clip(ratio, rho_min, rho_max)
+    c_t = jnp.clip(ratio, c_min, c_max)
+    sig_v = rho_t * (reward + discount * next_value - cur_value)
+    v = cur_value + sig_v + c_t * discount * (v - next_value)
+
+    # UPGO advantage (not corrected by importance sampling, unlike V-trace)
+    last_return = last_return * sign
+    next_q = next_q * sign
+    last_return = reward + discount * jnp.where(
+        next_q >= next_value, last_return, next_value)
+
+    next_q = reward + discount * next_value
+
+    carry = v, cur_value, last_return, next_q, main
+    return carry, (v, q_t, last_return)
+
+
+def vtrace(
+    next_value, ratios, values, rewards, next_dones, mains,
+    gamma, rho_min=0.001, rho_max=1.0, c_min=0.001, c_max=1.0, upgo=False,
+):
+    v = last_return = next_q = next_value
+    next_main = jnp.ones_like(next_value, dtype=jnp.bool_)
+    carry = v, next_value, last_return, next_q, next_main
+
+    _, (targets, q_estimate, return_t) = jax.lax.scan(
+        partial(vtrace_loop, gamma=gamma, rho_min=rho_min, rho_max=rho_max, c_min=c_min, c_max=c_max),
+        carry, (ratios, values, next_dones, rewards, mains), reverse=True
+    )
+    advantages = q_estimate - values
+    if upgo:
+        advantages += return_t - values
+    targets = jax.lax.stop_gradient(targets)
+    return targets, advantages
+
+
 def vtrace_2p0s_loop(carry, inp, gamma, rho_min, rho_max, c_min, c_max):
     v1, v2, next_values1, next_values2, reward1, reward2, xi1, xi2, \
         last_return1, last_return2, next_q1, next_q2 = carry
@@ -412,24 +401,46 @@ def truncated_gae_2p0s(
 
 
 def truncated_gae_loop(carry, inp, gamma, gae_lambda):
-    lastgaelam, next_value = carry
-    cur_value, next_done, reward = inp
-    nextnonterminal = 1.0 - next_done
+    lastgaelam, next_value, last_return, next_q, next_main = carry
+    cur_value, next_done, reward, main = inp
+    
+    lastgaelam = jnp.where(next_done, 0, lastgaelam)
+    next_value = jnp.where(next_done, 0, next_value)
 
-    delta = reward + gamma * next_value * nextnonterminal - cur_value
-    lastgaelam = delta + gamma * gae_lambda * nextnonterminal * lastgaelam
-    carry = lastgaelam, cur_value
-    return carry, lastgaelam
+    sign = jnp.where(main == next_main, 1, -1)
+    lastgaelam = lastgaelam * sign
+    next_value = next_value * sign
 
+    discount = gamma * (1.0 - next_done)
+
+    delta = reward + discount * next_value - cur_value
+    lastgaelam = delta + discount * gae_lambda * lastgaelam
+
+    # UPGO advantage
+    last_return = last_return * sign
+    next_q = next_q * sign
+    last_return = reward + discount * jnp.where(
+        next_q >= next_value, last_return, next_value)
+
+    next_q = reward + discount * next_value
 
-def truncated_gae(next_value, values, rewards, next_dones, gamma, gae_lambda):
+    carry = lastgaelam, cur_value, last_return, next_q, main
+    return carry, (lastgaelam, last_return)
+
+
+def truncated_gae(
+    next_value, values, rewards, next_dones, mains, gamma, gae_lambda, upgo=False):
     lastgaelam = jnp.zeros_like(next_value)
-    carry = lastgaelam, next_value
-    _, advantages = jax.lax.scan(
+    last_return = next_q = next_value
+    next_main = jnp.ones_like(next_value, dtype=jnp.bool_)
+    carry = lastgaelam, next_value, last_return, next_q, next_main
+    _, (advantages, return_t) = jax.lax.scan(
         partial(truncated_gae_loop, gamma=gamma, gae_lambda=gae_lambda),
-        carry, (values, next_dones, rewards), reverse=True
+        carry, (values, next_dones, rewards, mains), reverse=True
     )
     targets = values + advantages
+    if upgo:
+        advantages += return_t - values
     targets = jax.lax.stop_gradient(targets)
     return targets, advantages