wyj/katrain-qt
1
0 Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

exception handler and bug fixes

Browse Source
This commit is contained in:
Sander Land committed 2020-06-04 23:26:44 +02:00
1 parent 372a98ae1f
commit 20f4598410
6 files changed
+100 -23
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

katrain/__main__.py
+22 -2
View File
@@ -1,5 +1,8 @@
# first, logging level lower
import os
from kivy.base import ExceptionHandler, ExceptionManager
os.environ["KCFG_KIVY_LOG_LEVEL"] = os.environ.get("KCFG_KIVY_LOG_LEVEL", "warning")
os.environ['KIVY_AUDIO'] = "sdl2" # force working audio
@@ -481,6 +484,8 @@ class KaTrainApp(MDApp):
sys.exit(0)
def run_app():
app = KaTrainApp()
signal.signal(signal.SIGINT, app.signal_handler)
@@ -488,9 +493,24 @@ def run_app():
try:
app.run()
except Exception as e:
print(e)
app.on_request_close()
raise
print(f"FATAL ERROR: {e}")
traceback.print_exc()
if __name__ == "__main__":
class CrashHandler(ExceptionHandler):
def handle_exception(self, inst):
args = list(inst.args)
trace = traceback.format_exc()
message = args[0]
app = MDApp.get_running_app()
if app and app.gui:
app.gui.log("Exception: " + message + "\n" + trace,OUTPUT_ERROR)
else:
print("Exception: " + message + "\n" + trace)
return ExceptionManager.PASS
ExceptionManager.add_handler(CrashHandler())
run_app()
katrain/core/ai.py
+60 -16
View File
@@ -43,7 +43,9 @@ def fmt_moves(moves: List[Tuple[float, Move]]):
def policy_weighted_move(policy_moves, lower_bound, weaken_fac):
lower_bound, weaken_fac = max(0, lower_bound), max(0.01, weaken_fac)
weighted_coords = [(pv, pv ** (1 / weaken_fac), move) for pv, move in policy_moves if pv > lower_bound and not move.is_pass]
weighted_coords = [
(pv, pv ** (1 / weaken_fac), move) for pv, move in policy_moves if pv > lower_bound and not move.is_pass
]
if weighted_coords:
top = weighted_selection_without_replacement(weighted_coords, 1)[0]
ai_thoughts = f"Playing policy-weighted random move {top[2].gtp()} ({top[0]:.1%}) from {len(weighted_coords)} moves above lower_bound of {lower_bound:.1%}."
@@ -56,10 +58,19 @@ def policy_weighted_move(policy_moves, lower_bound, weaken_fac):
def generate_influence_territory_weights(ai_mode, ai_settings, policy_grid, size):
thr_line = ai_settings["threshold"] - 1 # zero-based
if ai_mode == AI_INFLUENCE:
weight = lambda x, y: (1 / ai_settings["line_weight"]) ** (max(0, thr_line - min(size[0] - 1 - x, x)) + max(0, thr_line - min(size[1] - 1 - y, y)))
weight = lambda x, y: (1 / ai_settings["line_weight"]) ** (
max(0, thr_line - min(size[0] - 1 - x, x)) + max(0, thr_line - min(size[1] - 1 - y, y))
)
else:
weight = lambda x, y: (1 / ai_settings["line_weight"]) ** (max(0, min(size[0] - 1 - x, x, size[1] - 1 - y, y) - thr_line))
weighted_coords = [(policy_grid[y][x] * weight(x, y), weight(x, y), x, y) for x in range(size[0]) for y in range(size[1]) if policy_grid[y][x] > 0]
weight = lambda x, y: (1 / ai_settings["line_weight"]) ** (
max(0, min(size[0] - 1 - x, x, size[1] - 1 - y, y) - thr_line)
)
weighted_coords = [
(policy_grid[y][x] * weight(x, y), weight(x, y), x, y)
for x in range(size[0])
for y in range(size[1])
if policy_grid[y][x] > 0
]
ai_thoughts = f"Generated weights for {ai_mode} according to weight factor {ai_settings['line_weight']} and distance from {thr_line + 1}th line. "
return weighted_coords, ai_thoughts
@@ -67,11 +78,18 @@ def generate_influence_territory_weights(ai_mode, ai_settings, policy_grid, size
def generate_local_tenuki_weights(ai_mode, ai_settings, policy_grid, cn, size):
var = ai_settings["stddev"] ** 2
mx, my = cn.move.coords
weighted_coords = [(policy_grid[y][x], math.exp(-0.5 * ((x - mx) ** 2 + (y - my) ** 2) / var), x, y) for x in range(size[0]) for y in range(size[1]) if policy_grid[y][x] > 0]
weighted_coords = [
(policy_grid[y][x], math.exp(-0.5 * ((x - mx) ** 2 + (y - my) ** 2) / var), x, y)
for x in range(size[0])
for y in range(size[1])
if policy_grid[y][x] > 0
]
ai_thoughts = f"Generated weights based on one minus gaussian with variance {var} around coordinates {mx},{my}. "
if ai_mode == AI_TENUKI:
weighted_coords = [(p, 1 - w, x, y) for p, w, x, y in weighted_coords]
ai_thoughts = f"Generated weights based on one minus gaussian with variance {var} around coordinates {mx},{my}. "
ai_thoughts = (
f"Generated weights based on one minus gaussian with variance {var} around coordinates {mx},{my}. "
)
return weighted_coords, ai_thoughts
@@ -92,7 +110,9 @@ def generate_ai_move(game: Game, ai_mode: str, ai_settings: Dict) -> Tuple[Move,
policy_grid = var_to_grid(cn.policy, size) # type: List[List[float]]
top_policy_move = policy_moves[0][1]
ai_thoughts += f"Using policy based strategy, base top 5 moves are {fmt_moves(policy_moves[:5])}. "
if (ai_mode == AI_POLICY and cn.depth <= ai_settings["opening_moves"]) or (ai_mode in [AI_LOCAL, AI_TENUKI] and not cn.move or cn.move.coords is None):
if (ai_mode == AI_POLICY and cn.depth <= ai_settings["opening_moves"]) or (
ai_mode in [AI_LOCAL, AI_TENUKI] and not (cn.move and cn.move.coords)
):
ai_mode = AI_WEIGHTED
ai_thoughts += f"Strategy override, using policy-weighted strategy instead. "
ai_settings = {"pick_override": 0.9, "weaken_fac": 1, "lower_bound": 0.02}
@@ -103,7 +123,7 @@ def generate_ai_move(game: Game, ai_mode: str, ai_settings: Dict) -> Tuple[Move,
elif ai_mode == AI_POLICY:
aimove = top_policy_move
ai_thoughts += f"Playing top policy move {aimove.gtp()}."
else: # weighted or pick-based
else: # weighted or pick-based
legal_policy_moves = [(pol, mv) for pol, mv in policy_moves if not mv.is_pass and pol > 0]
board_squares = size[0] * size[1]
if ai_mode == AI_RANK: # calibrated, override from 0.8 at start to ~0.4 at full board
@@ -115,7 +135,9 @@ def generate_ai_move(game: Game, ai_mode: str, ai_settings: Dict) -> Tuple[Move,
aimove = top_policy_move
ai_thoughts += f"Top policy move has weight > {override:.1%}, so overriding other strategies."
elif ai_mode == AI_WEIGHTED:
aimove, ai_thoughts = policy_weighted_move(policy_moves, ai_settings["lower_bound"], ai_settings["weaken_fac"])
aimove, ai_thoughts = policy_weighted_move(
policy_moves, ai_settings["lower_bound"], ai_settings["weaken_fac"]
)
elif ai_mode in AI_STRATEGIES_PICK:
if ai_mode != AI_RANK:
@@ -126,20 +148,33 @@ def generate_ai_move(game: Game, ai_mode: str, ai_settings: Dict) -> Tuple[Move,
if ai_mode in [AI_INFLUENCE, AI_TERRITORY, AI_LOCAL, AI_TENUKI]:
if cn.depth > ai_settings["endgame"] * board_squares:
weighted_coords = [(pol, 1, *mv.coords) for pol, mv in legal_policy_moves]
x_ai_thoughts = f"Generated equal weights as move number >= {ai_settings['endgame'] * size[0] * size[1]}. "
x_ai_thoughts = (
f"Generated equal weights as move number >= {ai_settings['endgame'] * size[0] * size[1]}. "
)
elif ai_mode in [AI_INFLUENCE, AI_TERRITORY]:
weighted_coords, x_ai_thoughts = generate_influence_territory_weights(ai_mode, ai_settings, policy_grid, size)
weighted_coords, x_ai_thoughts = generate_influence_territory_weights(
ai_mode, ai_settings, policy_grid, size
)
else: # ai_mode in [AI_LOCAL, AI_TENUKI]
weighted_coords, x_ai_thoughts = generate_local_tenuki_weights(ai_mode, ai_settings, policy_grid, cn, size)
weighted_coords, x_ai_thoughts = generate_local_tenuki_weights(
ai_mode, ai_settings, policy_grid, cn, size
)
ai_thoughts += x_ai_thoughts
else: # ai_mode in [AI_PICK, AI_RANK]:
weighted_coords = [(policy_grid[y][x], 1, x, y) for x in range(size[0]) for y in range(size[1]) if policy_grid[y][x] > 0]
weighted_coords = [
(policy_grid[y][x], 1, x, y)
for x in range(size[0])
for y in range(size[1])
if policy_grid[y][x] > 0
]
pick_moves = weighted_selection_without_replacement(weighted_coords, n_moves)
ai_thoughts += f"Picked {min(n_moves,len(weighted_coords))} random moves according to weights. "
if pick_moves:
new_top = [(p, Move((x, y), player=cn.next_player)) for p, wt, x, y in heapq.nlargest(5, pick_moves)]
new_top = [
(p, Move((x, y), player=cn.next_player)) for p, wt, x, y in heapq.nlargest(5, pick_moves)
]
aimove = new_top[0][1]
ai_thoughts += f"Top 5 among these were {fmt_moves(new_top)} and picked top {aimove.gtp()}. "
if new_top[0][0] < pass_policy:
@@ -159,12 +194,21 @@ def generate_ai_move(game: Game, ai_mode: str, ai_settings: Dict) -> Tuple[Move,
else:
if ai_mode == AI_JIGO:
sign = cn.player_sign(cn.next_player)
jigo_move = min(candidate_ai_moves, key=lambda move: abs(sign * move["scoreLead"] - ai_settings["target_score"]))
jigo_move = min(
candidate_ai_moves, key=lambda move: abs(sign * move["scoreLead"] - ai_settings["target_score"])
)
aimove = Move.from_gtp(jigo_move["move"], player=cn.next_player)
ai_thoughts += f"Jigo strategy found {len(candidate_ai_moves)} candidate moves (best {top_cand.gtp()}) and chose {aimove.gtp()} as closest to 0.5 point win"
elif ai_mode == AI_SCORELOSS:
c = ai_settings["strength"]
moves = [(d["pointsLost"], math.exp(min(200, -c * max(0, d["pointsLost"]))), Move.from_gtp(d["move"], player=cn.next_player),) for d in candidate_ai_moves]
moves = [
(
d["pointsLost"],
math.exp(min(200, -c * max(0, d["pointsLost"]))),
Move.from_gtp(d["move"], player=cn.next_player),
)
for d in candidate_ai_moves
]
topmove = weighted_selection_without_replacement(moves, 1)[0]
aimove = topmove[2]
ai_thoughts += f"ScoreLoss strategy found {len(candidate_ai_moves)} candidate moves (best {top_cand.gtp()}) and chose {aimove.gtp()} (weight {topmove[1]:.3f}, point loss {topmove[0]:.1f}) based on score weights."
katrain/core/base_katrain.py
+1 -1
View File
@@ -64,7 +64,7 @@ class KaTrainBase:
def log(self, message, level=OUTPUT_INFO):
if level == OUTPUT_ERROR:
print(f"ERROR: {message}", file=sys.stderr)
print(f"ERROR: {message}")
elif self.debug_level >= level:
print(message)
katrain/core/engine.py
+3 -2
View File
@@ -1,7 +1,7 @@
import copy
import json
import subprocess
import sys
import traceback
import threading
import time
from typing import Callable, Optional
@@ -157,7 +157,8 @@ class KataGoEngine:
if getattr(self.katrain, "update_state", None): # easier mocking etc
self.katrain.update_state()
except Exception as e:
self.katrain.log(f"Unexpected exception while processing KataGo output {line}", OUTPUT_ERROR)
traceback.print_exc(e)
self.katrain.log(f"Unexpected exception {e} while processing KataGo output {line}", OUTPUT_ERROR)
def send_query(self, query, callback, error_callback, next_move=None):
with self._lock:
katrain/gui/popups.py
+3
View File
@@ -213,6 +213,9 @@ class ConfigTimerPopup(QuickConfigGui):
class NewGamePopup(QuickConfigGui):
def __init__(self, katrain):
super().__init__(katrain)
for bw, info in katrain.players_info.items():
self.player_setup.update_players(bw, info)
self.rules_spinner.value_refs = [name for abbr, name in katrain.engine.RULESETS_ABBR]
def update_config(self, save_to_file=True):
tests/test_ai.py
+11 -2
View File
@@ -12,12 +12,13 @@ class TestAI:
def test_order(self):
assert set(AI_STRATEGIES_RECOMMENDED_ORDER) == set(AI_STRATEGIES)
@pytest.mark.skipif(os.environ.get('CI').lower() == 'true', reason='GH actions has no OpenCL')
@pytest.mark.skipif(os.environ.get('CI','').lower() == 'true', reason='GH actions has no OpenCL')
def test_ai_strategies(self):
katrain = KaTrainBase(force_package_config=True, debug_level=0)
engine = KataGoEngine(katrain, katrain.config("engine"))
game = Game(katrain, engine)
game = Game(katrain, engine)
n_rounds = 3
for _ in range(n_rounds):
for strategy in AI_STRATEGIES:
@@ -28,3 +29,11 @@ class TestAI:
assert played_node == game.current_node
assert game.current_node.depth == len(AI_STRATEGIES) * n_rounds
for strategy in AI_STRATEGIES:
game = Game(katrain, engine)
settings = katrain.config(f"ai/{strategy}")
move, played_node = generate_ai_move(game, strategy, settings)
katrain.log(f"Testing strategy on first move {strategy} -> {move}", OUTPUT_INFO)
assert game.current_node.depth == 1