File size: 10,832 Bytes
580868a |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 |
import json
import os
from collections import Counter
from typing import Any, Dict, List, Optional, Tuple
import matplotlib.pyplot as plt
import numpy as np
from torch.utils.tensorboard import SummaryWriter
def append_leafstats(tree1: Dict, tree2: Dict):
"""
Append each corresponding leaf of tree2 to tree1.
"""
for key, value in tree2.items():
if key not in tree1:
tree1[key] = value
elif isinstance(value, dict):
append_leafstats(tree1[key], value)
elif isinstance(value, list):
if isinstance(tree1[key], list):
tree1[key].extend(value)
else:
tree1[key] = [tree1[key]] + value
else:
if isinstance(tree1[key], list):
tree1[key].append(value)
else:
tree1[key] = [tree1[key], value]
def get_mean_leafstats(tree: Dict) -> Dict:
"""
scores a leafstats where each leaf is replaced by the mean of the leaf values.
"""
result = {}
for key, value in tree.items():
if isinstance(value, dict):
result[key] = get_mean_leafstats(value)
elif isinstance(value, list):
cleaned_values = [v for v in value if v is not None]
result[key] = np.mean(cleaned_values) if cleaned_values else None
else:
result[key] = value
return result
def get_mean_leafstats(trees: List[Dict]) -> Dict:
"""
Computes a smart element-wise mean across multiple leafstats.
For each path that exists in any of the input trees:
- If the path contains arrays in multiple trees, computes the element-wise mean
- Handles arrays of different lengths by taking means of available values at each index
- If a path exists in only some trees, still computes the mean using available data
Args:
trees: List of leafstats dictionaries to compute means from
Returns:
A new leafstats where each leaf is the element-wise mean of the corresponding array leaves in the input trees
"""
if not trees:
return {}
if len(trees) == 1:
return trees[0]
# Collect all possible keys at this level
all_keys = set()
for tree in trees:
all_keys.update(tree.keys())
result = {}
for key in all_keys:
# Collect all values at this key position
values_at_key = []
for tree in trees:
if key in tree:
values_at_key.append(tree[key])
# If all values are dictionaries, recursively compute means
if all(isinstance(v, dict) for v in values_at_key):
result[key] = get_mean_leafstats(values_at_key)
# If any value is a list, compute element-wise means
elif any(isinstance(v, list) for v in values_at_key):
# First, convert non-list values to lists (singleton)
list_values = []
for v in values_at_key:
if isinstance(v, list):
list_values.append(v)
else:
list_values.append([v])
# Find the maximum length among all lists
max_length = max(len(lst) for lst in list_values)
# Initialize result array
mean_array = []
# Compute element-wise means
for i in range(max_length):
# Collect values at this index that are not None
values_at_index = [
lst[i] for lst in list_values if i < len(lst) and lst[i] is not None
]
# If we have valid values, compute mean; otherwise, use None
if values_at_index:
mean_array.append(np.mean(values_at_index))
else:
mean_array.append(None)
result[key] = mean_array
# If all values are scalars (not dict or list), compute their mean
else:
# Filter out None values
non_none_values = [v for v in values_at_key if v is not None]
if non_none_values:
result[key] = np.mean(non_none_values)
else:
result[key] = None
return result
def get_var_leafstats(tree: Dict) -> Dict:
"""
scores a leafstats where each leaf is replaced by the variance of the leaf values.
"""
result = {}
for key, value in tree.items():
if isinstance(value, dict):
result[key] = get_var_leafstats(value)
elif isinstance(value, list):
cleaned_values = [v for v in value if v is not None]
result[key] = np.var(cleaned_values) if cleaned_values else None
else:
result[key] = 0 # Single value has variance 0
return result
def plot_leafstats(tree: Dict, folder: str, path: str = ""):
"""
Plots the leaves of the leafstats and saves them to the specified folder.
"""
os.makedirs(folder, exist_ok=True)
for key, value in tree.items():
new_path = f"{path}_{key}"
if isinstance(value, dict):
plot_leafstats(value, folder, new_path)
elif isinstance(value, list):
plt.figure()
plt.plot(value)
plt.title(new_path)
plt.savefig(os.path.join(folder, f"{new_path.replace('/', '_')}.png"))
plt.close()
def plot_EMA_leafstats(tree: Dict, folder: str, path: str = "", alpha: float = 0.1):
"""
Plots the exponential moving average of the leaves of the leafstats and saves them to the specified folder.
"""
os.makedirs(folder, exist_ok=True)
for key, value in tree.items():
new_path = f"{path}/{key}" if path else key
if isinstance(value, dict):
plot_EMA_leafstats(value, folder, new_path, alpha)
elif isinstance(value, list):
value = np.array(value)
nb_elements = len(value)
coefficients = (1 - alpha) ** np.arange(nb_elements, 0, -1)
value = np.cumsum(value * coefficients)
value /= (1 - alpha) ** np.arange(nb_elements - 1, -1, -1) # renormalize
out_path = f"EMA_alpha_{alpha}_{path}_{key}"
out_path = os.path.join(folder, f"{out_path.replace('/', '_')}.png")
plt.figure()
plt.plot(value)
plt.title(new_path)
plt.savefig(out_path)
plt.close()
def plot_SMA_leafstats(tree: Dict, folder: str, path: str = "", window: int = 33):
"""
Plots the simple moving average of the leaves of the leafstats and saves them to the specified folder.
"""
os.makedirs(folder, exist_ok=True)
for key, value in tree.items():
new_path = f"{path}/{key}" if path else key
if isinstance(value, dict):
plot_SMA_leafstats(value, folder, new_path, window)
elif isinstance(value, list):
value = np.array(value)
nb_elements = len(value)
assert window % 2 == 1 # Even numbers are annoying for centered windows
value = np.convolve(v=value, a=np.ones(window) / window, mode="same")
# Adjust out of window for start and finish
value[: window // 2] *= window / np.arange(window // 2 + 1, window)
value[-window // 2 + 1 :] *= window / np.arange(window - 1, window // 2, -1)
out_path = f"SMA_window_{window}_{path}_{key}"
out_path = os.path.join(folder, f"{out_path.replace('/', '_')}.png")
plt.figure()
plt.plot(value)
plt.title(new_path)
plt.savefig(out_path)
plt.close()
def save_leafstats(tree: Dict, folder: str, path: str = ""):
"""
Saves the leaves of the leafstats to the specified folder.
"""
os.makedirs(folder, exist_ok=True)
for key, value in tree.items():
new_path = f"{path}/{key}" if path else key
if isinstance(value, dict):
save_leafstats(value, folder, new_path)
elif (
isinstance(value, list)
or isinstance(value, np.ndarray)
or isinstance(value, float)
or isinstance(value, int)
):
with open(
os.path.join(folder, f"{new_path.replace('/', '_')}.json"), "w"
) as f:
json.dump(value, f, indent=4)
def tb_leafstats(tree: Dict, writer, path: str = ""):
"""
Logs the leaves of the leafstats to TensorBoard.
"""
for key, value in tree.items():
new_path = f"{path}/{key}" if path else key
if isinstance(value, dict):
tb_leafstats(value, writer, new_path)
elif isinstance(value, list):
for i, v in enumerate(value):
if v is not None:
writer.add_scalar(new_path, v, i)
def update_agent_statistics(input_path, output_file):
"""
Computes statistics for the current iteration and updates the global statistics file.
Args:
input_path (str): Path to the folder containing agent JSON files for the current iteration.
output_file (str): Path to the JSON file where statistics are stored.
"""
# Build leafstats by appending each dict from JSON files in "input_path" folder
leafstats = {}
for filename in os.listdir(input_path):
if filename.endswith(".json"):
with open(os.path.join(input_path, filename), "r") as f:
data = json.load(f)
append_leafstats(leafstats, data)
# Get epoch mean leafstats
mean_leafstats = get_mean_leafstats(leafstats)
# Add mean leafstats to global stats file
if os.path.exists(output_file):
with open(output_file, "r") as f:
global_stats = json.load(f)
else:
global_stats = {}
append_leafstats(global_stats, mean_leafstats)
with open(output_file, "w") as f:
json.dump(global_stats, f, indent=4)
def generate_agent_stats_plots(
global_stats_path, matplotlib_log_dir, tensorboard_log_dir, wandb_log_dir
):
"""
Visualizes the global statistics by logging them to TensorBoard and Weights & Biases.
Args:
global_stats_path (str): Path to the global statistics JSON file.
tensorboard_log_dir (str): Directory to save TensorBoard logs.
wandb_log_dir (str): Directory for Weights & Biases run metadata.
"""
os.makedirs(matplotlib_log_dir, exist_ok=True)
os.makedirs(tensorboard_log_dir, exist_ok=True)
os.makedirs(wandb_log_dir, exist_ok=True)
with open(global_stats_path, "r") as f:
global_stats = json.load(f)
plot_leafstats(global_stats, folder=matplotlib_log_dir)
# Log statistics to TensorBoard
writer = SummaryWriter(tensorboard_log_dir)
tb_leafstats(global_stats, writer)
writer.close()
# wb_leafstats(global_stats)
|