NN_trainer_module.py

 
import os
import shutil
import pandas as pd
import awkward as ak
 
import basf2 as b2
from ROOT import Belle2
import modularAnalysis as ma
from skim.WGs.fei import feiHadronicB0
from b2pandas_utils import VariablesToHDF5
 
from smartBKG.utils.preprocess import load_particle_list, preprocessed
 
 
class SaveFlag(b2.Module):
    """
    Save event numbers to a Parquet file.
 
    Arguments:
        out_file (str): Output file path for saving the event numbers.
 
    Returns:
        None
 
    Note:
        This module should be added after the skimming process.
    """
 
    def __init__(self, out_file=None):
        """
        Initialize the SaveFlag module.
 
        :param out_file: Output file path for saving the event numbers.
        """
        super().__init__()
        
        self.out_file = out_file
 
    def initialize(self):
        """
        Initialize the data store and the list to save event numbers before processing events.
        """
        
        self.eventInfo = Belle2.PyStoreObj('EventMetaData')
        
        self.pass_list = []
 
    def event(self):
        """
        Process each event and append event numbers to the pass list.
        """
        self.pass_list.append(self.eventInfo.getEvent())
 
    def terminate(self):
        """
        Finalize the module and save the pass list to a Parquet file.
        """
        ak.to_parquet(self.pass_list, self.out_file)
 
 
class TrainDataSaver(b2.Module):
    """
    Save MCParticles to Pandas Dataframe.
 
    Arguments:
        output_file (str): Filename to save training data.
            Ending with ``parquet`` indicating fast mode, which will generate the final parquet file for training.
            Ending with ``h5`` indicating advanced mode, which will produce a temperary h5 file for further preprocessing.
        flag_file (str): Filename of the flag file indicating passing events.
 
    Returns:
        None
    """
 
    def __init__(
        self,
        output_file,
        flag_file,
    ):
        """
        Initialize the TrainDataSaver module.
 
        :param output_file: Filename to save training data to.
        :param flag_file: Filename of the flag file indicating passing events.
        """
        super().__init__()
        
        self.output_file = output_file
        
        self.flag_list = ak.from_parquet(flag_file)
        
        self.fast_mode = output_file.endswith(".parquet")
 
        # delete output file if it already exists, since we will apend later
        if os.path.exists(output_file):
            os.remove(output_file)
 
    def initialize(self):
        """
        Initialize the data store and the dictionary to save particle features before processing events.
        """
        
        self.eventInfo = Belle2.PyStoreObj('EventMetaData')
        
        self.eventExtraInfo = Belle2.PyStoreObj('EventExtraInfo')
        
        self.df_dict = pd.DataFrame()
 
    def event(self):
        """
        Process each event and append event information to the dictionary.
        """
        evtNum = self.eventInfo.getEvent()
        self.df_dict = pd.concat([
            self.df_dict,
            load_particle_list(mcplist=Belle2.PyStoreArray("MCParticles"), evtNum=evtNum, label=(evtNum in self.flag_list))
            ])
 
    def terminate(self):
        """
        Append events on disk in either of the two different ways and free memory.
 
        In fast mode, the dataframe containing particle-level information and skim labels is preprocessed
        and saved as a parquet file which is ready for NN training.
 
        In advanced mode, the dataframe is saved as a h5 file and waits for combination with event-level information
        before preprocessing.
        """
        if self.fast_mode:
            ak.to_parquet(preprocessed(self.df_dict), self.output_file)
        else:
            self.df_dict.to_hdf(self.output_file, key='mc_information', mode='a', format='table', append=True)
        self.df_dict = pd.DataFrame()
 
 
class data_production():
    """
    Process data for training and save to Parquet file. Two modes are provided:
    Fast mode: save_vars set to None, produce the dataset with only the necessary information for the training.
    Advanced mode: save_vars set to a dictionary of event-level variables,
    run through hard-coded b2 steering code in self.process_b2script to produce the required particle lists
    and save the required variables, can be used for event-level cuts or evaluations of the NN performance.
 
    Arguments:
        in_dir (str): Input directory.
        out_dir (str): Output directory.
        job_id (int): Job ID for batch processing.
        save_vars (dict): Event-level variables to save for different particles.
            By default None for fast mode.
            In the example script having Y4S and B keys for the corresponding particle list.
 
    Returns:
        None
    """
 
    def __init__(self, in_dir, out_dir, job_id, save_vars=None):
        """
        Initialize the data_production object.
 
        :param in_dir: Input directory.
        :param out_dir: Output directory.
        :param job_id: Job ID for batch processing.
        :param save_vars: Event-level variables to save for different particles.
        By default None for fast mode.
        In the example script having Y4S and B keys for the corresponding particle list.
        """
        dataName = '_submdst'
        flagName = '_flag'
        
        self.data = f'{in_dir}{dataName}{job_id}.root'
        
        self.flag = f'{in_dir}{flagName}{job_id}.parquet'
        if save_vars is not None:
            
            self.out_temp = f'{out_dir}_temp{job_id}/'
            os.makedirs(out_dir, exist_ok=True)
            os.makedirs(self.out_temp, exist_ok=True)
            
            self.temp_file = {
                'MC': f'{self.out_temp}mc.h5',
                'Y4S': f'{self.out_temp}y4s.h5',
                'B': f'{self.out_temp}b.h5'
                }
        
        self.out_file = f'{out_dir}preprocessed{job_id}.parquet'
        
        self.save_vars = save_vars
 
    def process(self):
        """
        Process the b2 steering file and the data generation.
        """
        self.process_b2script()
        if self.save_vars is not None:
            self.merge_files()
 
    def process_b2script(self, num_events=2500):
        """
        Skimming process with TrainDataSaver module.
 
        :param num_events: Maximum number of events to process.
        """
        path = ma.create_path()
 
        ma.inputMdst(environmentType='default', filename=self.data, path=path)
        ma.buildEventShape(path=path)
        ma.buildEventKinematics(path=path)
 
        # process with advance mode
        if self.save_vars is not None:
            TrainDataSaver_module = TrainDataSaver(
                output_file=self.temp_file['MC'],
                flag_file=self.flag,
            )
            path.add_module(TrainDataSaver_module)
            ma.fillParticleListFromMC('Upsilon(4S):mc', '', path=path)
            v2hdf5_y4s = VariablesToHDF5(
                'Upsilon(4S):mc',
                self.save_vars['Y4S'],
                filename=self.temp_file['Y4S'],
            )
            path.add_module(v2hdf5_y4s)
 
            fei_skim = feiHadronicB0(udstOutput=False, analysisGlobaltag=ma.getAnalysisGlobaltag())
            fei_skim(path=path)
            fei_skim.postskim_path.add_module(
                    "BestCandidateSelection",
                    particleList="B0:generic",
                    variable="extraInfo(SignalProbability)",
                    outputVariable="rank_signalprob",
                    numBest=1,
                )
            # Key of saved table is the name of particle list
            v2hdf5_b = VariablesToHDF5(
                'B0:generic',
                self.save_vars['B'],
                filename=self.temp_file['B'],
            )
            fei_skim.postskim_path.add_module(v2hdf5_b)
        # process with fast mode
        else:
            TrainDataSaver_module = TrainDataSaver(
                output_file=self.out_file,
                flag_file=self.flag,
            )
            path.add_module(TrainDataSaver_module)
        b2.process(path, max_event=num_events)
        print(b2.statistics)
 
    def merge_files(self):
        """
        Merge file of particle-level information (MC) with those of event-level information (Y4S, B).
        Preprocess and save to disk as Parquet file in form of Awkward Array.
        """
        df = pd.read_hdf(self.temp_file['MC'], key='mc_information')
        df_y4s = pd.read_hdf(self.temp_file['Y4S'], key='Upsilon(4S):mc')
        df_b = pd.read_hdf(self.temp_file['B'], key='B0:generic')
        df_merged = df_y4s.merge(df_b.drop(axis=1, labels=['icand', 'ncand']), how="left")
        decorr_df = df_merged.rename({'evt': 'evtNum'}, axis=1)
        ak.to_parquet(preprocessed(df, decorr_df), self.out_file)
 
    def clean_up(self):
        """
        Clean up temporary files.
        """
        # uncomment if needed for batch job
        # os.remove(self.data)
        os.remove(self.flag)
        if self.save_vars is not None:
            shutil.rmtree(self.out_temp)