File size: 4,317 Bytes
9897e20 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 | import torch
import pandas as pd
from box import Box
import numpy as np
import os
import logging
from gpudrive.env.dataset import SceneDataLoader
from eval_utils import (
load_config,
make_env,
load_policy,
evaluate_policy,
)
import random
import torch
import numpy as np
def set_seed(seed: int):
random.seed(seed)
np.random.seed(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed_all(seed) # If using CUDA
torch.backends.cudnn.deterministic = True
logging.basicConfig(level=logging.INFO)
SEED = 42 # Set to any fixed value
set_seed(SEED)
if __name__ == "__main__":
# Load configurations
eval_config = load_config("examples/experimental/config/eval_config")
model_config = load_config("examples/experimental/config/model_config")
train_loader = SceneDataLoader(
root=eval_config.train_dir,
batch_size=eval_config.num_worlds,
dataset_size=eval_config.num_worlds,
sample_with_replacement=False,
)
# Make environment
env = make_env(eval_config, train_loader)
for model in model_config.models:
logging.info(f"Evaluating model {model.name}")
# Load policy
policy = load_policy(
path_to_cpt=model_config.models_path,
model_name=model.name,
device=eval_config.device,
env=env,
)
# Create dataloaders for train and test sets
train_loader = SceneDataLoader(
root=eval_config.train_dir,
batch_size=eval_config.num_worlds,
dataset_size=model.train_dataset_size
if model.name != "random_baseline"
else 1000,
sample_with_replacement=False,
shuffle=False,
)
test_loader = SceneDataLoader(
root=eval_config.test_dir,
batch_size=eval_config.num_worlds,
dataset_size=eval_config.test_dataset_size
if model.name != "random_baseline"
else 1000,
sample_with_replacement=False,
shuffle=True,
)
# Rollouts
logging.info(
f"Rollouts on {len(set(train_loader.dataset))} train scenes / {len(set(test_loader.dataset))} test scenes"
)
df_res_train = evaluate_policy(
env=env,
policy=policy,
data_loader=train_loader,
dataset_name="train",
deterministic=False,
render_sim_state=False,
)
df_res_test = evaluate_policy(
env=env,
policy=policy,
data_loader=test_loader,
dataset_name="test",
deterministic=False,
render_sim_state=False,
)
# Concatenate train/test results
df_res = pd.concat([df_res_train, df_res_test])
# Add metadata
df_res["model_name"] = model.name
df_res["train_dataset_size"] = model.train_dataset_size
# Store
if not os.path.exists(eval_config.res_path):
os.makedirs(eval_config.res_path)
tab_agg_perf = df_res.groupby("dataset")[
[
"goal_achieved_frac",
"collided_frac",
"off_road_frac",
"other_frac",
]
].agg(["mean", "std"])
tab_agg_perf = tab_agg_perf * 100
tab_agg_perf = tab_agg_perf.round(1)
print("Scene-based metrics \n")
print(tab_agg_perf)
print("")
print("Agent-based metrics \n")
total_agents = df_res["controlled_agents_in_scene"].sum()
collision_rate = (df_res["collided_count"].sum() / total_agents) * 100
offroad_rate = (df_res["off_road_count"].sum() / total_agents) * 100
goal_rate = (df_res["goal_achieved_count"].sum() / total_agents) * 100
other_rate = (df_res["other_count"].sum() / total_agents) * 100
print(f"Total agents: {total_agents} in {df_res.shape[0]} scenes")
print(f"Collision rate: {collision_rate}")
print(f"Offroad rate: {offroad_rate}")
print(f"Goal rate: {goal_rate}")
print(f"Other rate: {other_rate}")
df_res.to_csv(f"{eval_config.res_path}/{model.name}.csv", index=False)
logging.info(f"Saved at {eval_config.res_path}/{model.name}.csv \n")
|