def run_mva(client, parameters, df):
mva_path = parameters.pop("mva_path", "./")
mkdir(mva_path)
mva_models = parameters.pop("mva_models", {})
saved_models = parameters.pop("saved_models", {})
training_datasets = parameters.pop("training_datasets", {})
features = parameters.pop("training_features", [])
do_training = parameters.pop("mva_do_training", False)
do_evaluation = parameters.pop("mva_do_evaluation", False)
do_plotting = parameters.pop("mva_do_plotting", False)
channels_to_use = parameters.get("mva_channels", ["ggh_0jets"])
for channel in channels_to_use:
out_dir = f"{mva_path}/{channel}"
mkdir(out_dir)
parameters["plots_path"] = out_dir
trainer = Trainer(
df=df[df.channel == channel],
channel=channel,
ds_dict=training_datasets,
features=features,
out_path=out_dir,
training_cut="(dimuon_mass > 110) & (dimuon_mass < 150)",
)
# trainer.shape_in_eta_bins(shape_of="dimuon_mass", nbins=10)
if do_training:
trainer.add_models(mva_models.copy())
trainer.run_training(client)
if channel in saved_models.keys():
if len(saved_models[channel].keys()) > 0:
trainer.add_saved_models(saved_models[channel])
if do_evaluation:
trainer.run_evaluation(client)
trainer.shape_in_bins(shape_of="dimuon_mass", nbins=6)
# trainer.shape_in_bins(shape_of="max_abs_eta", nbins=10)
for model_name in trainer.models.keys():
score_name = f"{model_name}_score"
df.loc[df.channel == channel, score_name] = trainer.df.loc[
:, score_name
]
if do_plotting:
trainer.plot_roc_curves()
parameters_tmp = parameters.copy()
parameters_tmp["hist_vars"] = []
parameters_tmp["plot_vars"] = []
parameters_tmp["regions"] = ["h-peak", "h-sidebands"]
parameters_tmp["channels"] = [channel]
all_models = []
if channel in mva_models.keys():
all_models += list(mva_models[channel].keys())
if channel in saved_models.keys():
all_models += list(saved_models[channel].keys())
all_models = list(set(all_models))
for model_name in all_models:
score_name = f"{model_name}_score"
parameters_tmp["hist_vars"].append(score_name)
parameters_tmp["plot_vars"].append(score_name)
parameters_tmp["variables_lookup"][score_name] = Variable(
score_name, score_name, 50, 0, 1
)
hist_df = to_histograms(client, parameters_tmp, trainer.df)
plotter(client, parameters_tmp, hist_df)
parameters["hist_vars"] += ["score_" + m for m in parameters["bdt_models"]]
# parameters['plot_vars'] = ['dimuon_mass']
parameters["plot_vars"] = parameters["hist_vars"]
parameters["datasets"] = datasets
all_paths = {}
for year in parameters["years"]:
all_paths[year] = {}
for dataset in datasets:
paths = glob.glob(
f"{parameters['path']}/"
f"{year}_{parameters['label']}/"
f"{dataset}/*.parquet"
)
all_paths[year][dataset] = paths
if args.remake_hists:
for year in parameters["years"]:
print(f"Processing {year}")
for dataset, path in tqdm.tqdm(all_paths[year].items()):
if len(path) == 0:
continue
df = load_dataframe(client, parameters, inputs=[path])
if not isinstance(df, dd.DataFrame):
continue
to_histograms(client, parameters, df=df)
if args.plot:
yields = plotter(client, parameters)
"signals": ["ggh_powheg", "vbf_powheg"],
}
if __name__ == "__main__":
tick = time.time()
client = Client(processes=True,
n_workers=1,
threads_per_worker=1,
memory_limit="4GB")
file_name = "dy_delphes_stage1_output.parquet"
path = f"{os.getcwd()}/tests/samples/{file_name}"
out_df = load_dataframe(client, parameters, inputs=[path])
out_hist = to_histograms(client, parameters, df=out_df)
out_plot = plotter(client, parameters, hist_df=out_hist)
out_tmp = to_templates(client, parameters, hist_df=out_hist)
elapsed = round(time.time() - tick, 3)
print(f"Finished everything in {elapsed} s.")
slicer = {
"region": "h-peak",
"channel": "ggh_0jets",
"val_sumw2": "value",
"dimuon_mass": slice(None),
}
assert almost_equal(out_hist["hist"][0][slicer].sum(), 3349.189725131393)
assert almost_equal(sum(out_plot), 3349.189725131393)
assert almost_equal(sum(out_tmp), 3349.189725131393)