Spaces:
Sleeping
Sleeping
File size: 11,916 Bytes
e75a247 |
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 |
# Eval a given a training run name at the given steps, taking into account the chaning of the training runs
import sys
import pickle
import wandb
import argparse
import os
from src.utils.paths import get_path
from src.utils.wandb_utils import get_run_initial_steps, get_run_step_direct, get_run_step_ckpt, get_steps_from_file, get_run_by_name
parser = argparse.ArgumentParser()
parser.add_argument("--tag", "-tag", type=str, required=False, default="")
parser.add_argument("--input", "-input", type=str, required=False, default="Feb26_2025_E1000_N500_noPartonFilter_C_F") # --input Feb26_2025_E1000_N500_full
parser.add_argument("--clustering-suffix", "-c", type=str, required=False, default="") # -c MinSamples0
parser.add_argument("--no-submit", "-ns", action="store_true") # do not submit the slurm job
parser.add_argument("--submit-AKX", "-AKX", action="store_true")
parser.add_argument("--submit-AK8", "-AK8", action="store_true")
parser.add_argument("--parton-level", "-pl", action="store_true") # To be used together with 'fastjet_jets' and --submit-AKX
parser.add_argument("--gen-level", "-gl", action="store_true")
parser.add_argument("--overwrite", "-ow", action="store_true") # overwrite the slurm job if it exists
parser.add_argument("--pt-cutoff-jet", "-pt", type=float, default=100.0, help="pt cutoff for what is considered a jet")
parser.add_argument("--high-eta-only", "-he", action="store_true", help="Only evaluate high eta jets (eta > 1.5)")
parser.add_argument("--low-eta-only", "-le", action="store_true", help="Only evaluate low eta jets (eta < 1.5)")
parser.add_argument("--ds-cap", "-ds", type=int, default=10000, help="dataset cap ")
args = parser.parse_args()
api = wandb.Api()
DSCAP = args.ds_cap
def get_eval_run_names(tag):
# from the api, get all the runs with the tag that are finished
runs = api.runs(
path="fcc_ml/svj_clustering",
filters={"tags": {"$in": [tag.strip()]}}
)
return [run.name for run in runs if run.state == "finished"], [run.config for run in runs if run.state == "finished"]
def get_log_number(tag):
numbers = set()
for file in os.listdir("jobs/slurm_files"):
if tag in file:
numbers.add(int(file.split("_")[-1].split(".")[0]))
if len(numbers) == 0:
return 0
return max(list(numbers)) + 1
def get_slurm_file_text_AKX(tag, log_number):
bindings = "-B /t3home/gkrzmanc/ -B /work/gkrzmanc/"
partition = "standard"
account = "t3"
d = "jobs/logs/{}".format(tag)
err = d + "_{}_CPUerr.txt".format(log_number)
log = d + "_{}_CPUlog.txt".format(log_number)
suffix_pl = "--parton-level" if args.parton_level else ""
suffix_gl = "--gen-level" if args.gen_level else ""
pl_folder = "_PL" if args.parton_level else ""
gl_folder = "_GL" if args.gen_level else ""
if args.pt_cutoff_jet != 100.0:
pt_cutoff_suffix = f"_pt_{args.pt_cutoff_jet}"
pt_cutoff_suffix_cmd = " --pt-jet-cutoff {}".format(args.pt_cutoff_jet)
else:
pt_cutoff_suffix = ""
pt_cutoff_suffix_cmd = ""
if args.high_eta_only:
pt_cutoff_suffix += "_high_eta"
pt_cutoff_suffix_cmd += " --high-eta-only"
elif args.low_eta_only:
pt_cutoff_suffix += "_low_eta"
pt_cutoff_suffix_cmd += " --low-eta-only"
file = f"""#!/bin/bash
#SBATCH --partition={partition} # Specify the partition
#SBATCH --account={account} # Specify the account
#SBATCH --mem=25000 # Request 10GB of memory
#SBATCH --time=06:00:00 # Set the time limit to 1 hour
#SBATCH --job-name=SVJan_AKX{pl_folder}{gl_folder}_{str(log_number)} # Name the job
#SBATCH --error={err} # Redirect stderr to a log file
#SBATCH --output={log} # Redirect stderr to a log file
#SBATCH --mail-type=FAIL
#SBATCH [email protected]
source env.sh
export APPTAINER_TMPDIR=/work/gkrzmanc/singularity_tmp
export APPTAINER_CACHEDIR=/work/gkrzmanc/singularity_cache
nvidia-smi
srun singularity exec {bindings} docker://gkrz/lgatr:v3 python -m scripts.analysis.count_matched_quarks --input {args.input} --output {args.input}/batch_eval_2k/{tag}{pt_cutoff_suffix}/AKX{pl_folder}{gl_folder} --jets-object fastjet_jets {suffix_pl} {suffix_gl} --dataset-cap {DSCAP} {pt_cutoff_suffix_cmd}
"""
return file
def get_slurm_file_text_AK(tag, log_number):
bindings = "-B /t3home/gkrzmanc/ -B /work/gkrzmanc/"
partition = "standard"
account = "t3"
d = "jobs/logs/{}".format(tag)
err = d + "_{}_CPUerr.txt".format(log_number)
log = d + "_{}_CPUlog.txt".format(log_number)
file = f"""#!/bin/bash
#SBATCH --partition={partition} # Specify the partition
#SBATCH --account={account} # Specify the account
#SBATCH --mem=25000 # Request 10GB of memory
#SBATCH --time=02:00:00 # Set the time limit to 1 hour
#SBATCH --job-name=SVJan # Name the job
#SBATCH --error={err} # Redirect stderr to a log file
#SBATCH --output={log} # Redirect stderr to a log file
#SBATCH --mail-type=END,FAIL
#SBATCH [email protected]
source env.sh
export APPTAINER_TMPDIR=/work/gkrzmanc/singularity_tmp
export APPTAINER_CACHEDIR=/work/gkrzmanc/singularity_cache
nvidia-smi
srun singularity exec {bindings} docker://gkrz/lgatr:v3 python -m scripts.analysis.count_matched_quarks --input {args.input} --output {args.input}/batch_eval_2k/{tag}/AK8 --dataset-cap 1500
srun singularity exec {bindings} docker://gkrz/lgatr:v3 python -m scripts.analysis.count_matched_quarks --input {args.input} --output {args.input}/batch_eval_2k/{tag}/AK8_GenJets --jets-object genjets --dataset-cap {DSCAP}
"""
return file
def get_slurm_file_text(tag, eval_job_name, log_number, aug_suffix = ""):
bindings = "-B /t3home/gkrzmanc/ -B /work/gkrzmanc/ -B /pnfs/psi.ch/cms/trivcat/store/user/gkrzmanc/ "
partition = "standard"
account = "t3"
d = "jobs/logs/{}".format(tag)
err = d + "_{}_CPUerr.txt".format(log_number)
log = d + "_{}_CPUlog.txt".format(log_number)
clust_suffix = ""
if args.clustering_suffix != "":
clust_suffix = f" --clustering-suffix {args.clustering_suffix}"
pt_cutoff_suffix_cmd = f" --pt-jet-cutoff {args.pt_cutoff_jet}"
pt_cutoff_suffix = ""
if args.pt_cutoff_jet != 100.0:
pt_cutoff_suffix = f"_pt_{args.pt_cutoff_jet}"
if args.high_eta_only:
pt_cutoff_suffix += "_high_eta"
#aug_suffix += " --high-eta-only"
elif args.low_eta_only:
pt_cutoff_suffix += "_low_eta"
#aug_suffix += " --low-eta-only"
file = f"""#!/bin/bash
#SBATCH --partition={partition} # Specify the partition
#SBATCH --account={account} # Specify the account
#SBATCH --mem=25000 # Request 10GB of memory
#SBATCH --time=02:00:00 # Set the time limit to 1 hour
#SBATCH --job-name=SVJ_CPU_{eval_job_name}_{str(log_number)} # Name the job
#SBATCH --error={err} # Redirect stderr to a log file
#SBATCH --output={log} # Redirect stderr to a log file
#SBATCH --mail-type=FAIL
#SBATCH [email protected]
source env.sh
export APPTAINER_TMPDIR=/work/gkrzmanc/singularity_tmp
export APPTAINER_CACHEDIR=/work/gkrzmanc/singularity_cache
nvidia-smi
srun singularity exec {bindings} docker://gkrz/lgatr:v3 python -m scripts.analysis.count_matched_quarks --input {args.input} --output {args.input}/batch_eval_2k/{tag}{pt_cutoff_suffix}/{eval_job_name}{args.clustering_suffix} --eval-dir train/{eval_job_name} --jets-object model_jets --dataset-cap {DSCAP} {aug_suffix} {clust_suffix} {pt_cutoff_suffix_cmd}
"""
return file
runs, run_config = get_eval_run_names(args.tag)
print("RUNS:", runs)
if args.submit_AK8:
# Submit also ak and ak8
if not os.path.exists("jobs/slurm_files"):
os.makedirs("jobs/slurm_files")
if not os.path.exists("jobs/logs"):
os.makedirs("jobs/logs")
log_number = get_log_number(args.tag)
slurm_file_text = get_slurm_file_text_AK(args.tag, log_number)
# write the file to jobs/slurm_files
with open("jobs/slurm_files/evalCPU_{}_{}.slurm".format(args.tag, log_number), "w") as f:
f.write(slurm_file_text)
print("Wrote file to jobs/slurm_files/evalCPU_{}_{}.slurm".format(args.tag, log_number))
if not args.no_submit:
os.system("sbatch jobs/slurm_files/evalCPU_{}_{}.slurm".format(args.tag, log_number))
print("---- Submitted AK8 run -----")
sys.exit(0)
def extract_n_events(filename):
if not os.path.exists(filename):
return -1
content = open(filename).read().strip()
try:
return int(content)
except:
return -1
if args.submit_AKX:
# Submit also AKX
if not os.path.exists("jobs/slurm_files"):
os.makedirs("jobs/slurm_files")
if not os.path.exists("jobs/logs"):
os.makedirs("jobs/logs")
log_number = get_log_number(args.tag)
slurm_file_text = get_slurm_file_text_AKX(args.tag, log_number)
# write the file to jobs/slurm_files
with open("jobs/slurm_files/evalCPU_{}_{}.slurm".format(args.tag, log_number), "w") as f:
f.write(slurm_file_text)
print("Wrote file to jobs/slurm_files/evalCPU_{}_{}.slurm".format(args.tag, log_number))
if not args.no_submit:
os.system("sbatch jobs/slurm_files/evalCPU_{}_{}.slurm".format(args.tag, log_number))
print("---- Submitted AKX run -----")
sys.exit(0)
for i, run in enumerate(runs):
#if get_run_by_name(run).state != "finished":
# print("Run not finished (failed or still in progress) - skipping", run)
# continue
conf = get_run_by_name(run).config
if( conf.get("parton_level") or conf.get("gen_level")) and args.pt_cutoff_jet != 100.0:
print("Skipping run", run, "because it is parton level or gen level and pt cutoff is not 100.0")
continue
aug_soft_p = conf.get("augment_soft_particles", False)
if aug_soft_p:
aug_suffix = "-aug-soft"
else:
aug_suffix = ""
if not os.path.exists("jobs/slurm_files"):
os.makedirs("jobs/slurm_files")
if not os.path.exists("jobs/logs"):
os.makedirs("jobs/logs")
log_number = get_log_number(args.tag)
pt_cutoff_suffix = ""
if args.pt_cutoff_jet != 100.0:
pt_cutoff_suffix = f"_pt_{args.pt_cutoff_jet}"
if args.high_eta_only:
pt_cutoff_suffix += "_high_eta"
aug_suffix += " --high-eta-only"
elif args.low_eta_only:
pt_cutoff_suffix += "_low_eta"
aug_suffix += " --low-eta-only"
slurm_file_text = get_slurm_file_text(args.tag, run, log_number, aug_suffix)
rel_path_save = f"{args.input}/batch_eval_2k/{args.tag}{pt_cutoff_suffix}/{run}{args.clustering_suffix}"
rel_path_save = get_path(rel_path_save, "results")
if not os.path.exists(rel_path_save):
os.makedirs(rel_path_save)
#if evaluated(rel_path_save):
n_events = extract_n_events(os.path.join(rel_path_save, "count_matched_quarks", "n_events.txt"))
if os.path.exists(os.path.join(rel_path_save, "count_matched_quarks", "n_events.txt")) and not args.overwrite and n_events > 0:
print("Skipping", run, "because this file exists:", os.path.join(rel_path_save, "count_matched_quarks", "n_events.txt"))
continue
else:
print("Evaluating", run)
# save run config here
with open(f"{rel_path_save}/run_config.pkl", "wb") as f:
pickle.dump(run_config[i], f)
# write the file to jobs/slurm_files
with open("jobs/slurm_files/evalCPU_{}_{}.slurm".format(args.tag, log_number), "w") as f:
f.write(slurm_file_text)
print("Wrote file to jobs/slurm_files/evalCPU_{}_{}.slurm".format(args.tag, log_number))
if not args.no_submit:
os.system("sbatch jobs/slurm_files/evalCPU_{}_{}.slurm".format(args.tag, log_number))
|