sources/compton_combiner.py

"""Script to combine compton measurements with runs and process these data
"""

import argparse
from configparser import ConfigParser
from datetime import datetime, timedelta, timezone
import os
import sqlite3
from typing import Union, Tuple, Optional

from compton_filter import CalibrdbHandler
from iminuit import Minuit
import matplotlib.dates as mdates
import matplotlib.pyplot as plt
from mysql.connector import connect, Error
import numpy as np
import pandas as pd
from tqdm import tqdm

SEASONS = {
    'name': ['HIGH2017', 'RHO2018', 'HIGH2019', 'LOW2020', 'HIGH2020', 'HIGH2021', 'NNBAR2021'],
    'start_run': [36872, 48938, 70014, 85224, 89973, 98116, 107342, None],
}

class RunsDBHandler():
    def __init__(self, host: str = 'cmddb', database: str = 'online', user: str = None, password: str = None):
        self.conn = connect(host = host, database = database, user = user, password = password)
        self.cur = self.conn.cursor()
        self.cur.execute("SET time_zone = '+07:00';")
        
    @property
    def fields(self) -> list:
        """Returns a list of available columns in the RunsDB
        """
        
        self.cur.execute("""DESCRIBE Runlog""")
        return self.cur.fetchall()
        
    def load_tables(self, range: Union[Tuple[int, Optional[int]], Tuple[datetime, datetime]], energy_point: Optional[float] = None, select_bad_runs: bool = False):
        """
        Parameters
        ----------
        range : Union[Tuple[int, Optional[int]], Tuple[datetime, datetime]]
            selection range
            int range defines an interval in runs
            datetime range defines a time interval (NSK: +7:00 time)
        energy_point : Optional[float]
            energy point name, MeV (default is  None)
        select_bad_runs : bool
            select runs with labels except (Y) (default is False)
        """
        
        cond = ""
        if isinstance(range[0], int):
            cond = f" AND run >= {range[0]} "
            if range[1] is not None:
                cond += f" AND run <= {range[1]} "
        elif isinstance(range[0], datetime):
            cond = f" AND starttime >= %s "
            if range[1] is not None:
                cond += " AND stoptime <= %s"
            else:
                range = (range[0], )
                
        energy_cond = ""
        if energy_point is not None:
            energy_cond = f" AND energy = {energy_point}"
            
        quality_cond = ' quality = "Y" '
        if select_bad_runs:
            quality_cond = ' quality <> "Y" '
            
        sql_query = f"""
        SELECT 
            run, 
            starttime, 
            stoptime, 
            energy, 
            luminosity
        FROM Runlog 
        WHERE 
            {quality_cond}
            {cond}
            {energy_cond}
            AND luminosity > 0
            AND stoptime > starttime
            AND nevent > 0
        ORDER BY run DESC"""
        
        if isinstance(range[0], datetime):
            self.cur.execute(sql_query, range)
        else:
            self.cur.execute(sql_query)
        
        field_names = [i[0] for i in self.cur.description]
        res = self.cur.fetchall()
        return res, field_names
        
    def __del__(self):
        self.conn.close()
        
class Combiner():
    """Combines a dataframe with runs and a dataframe with compton measurements together
    """
    
    def __init__(self, runsdb: Tuple[list, list], clbrdb: Tuple[list, list]):
        """
        Parameters
        ----------
        runsdb : Tuple[list, list]
            table of runs (rows and field names)
        clbrdb : Tuple[list, list]
            table of compton measurements (rows and field names)            
        """
        
        rdb_rows, r_fld = runsdb
        cdb_rows, c_fld = clbrdb
        
        self.conn = sqlite3.connect(":memory:", detect_types=sqlite3.PARSE_DECLTYPES|sqlite3.PARSE_COLNAMES)
        self.cur = self.conn.cursor()
        self.cur.execute(f"CREATE table runs (run, elabel, starttime timestamp, stoptime timestamp, luminosity)")
        self.cur.execute(f"CREATE table compton (begintime timestamp, endtime timestamp, e_mean, e_std, spread_mean, spread_std)")
        
        run_row_generator = map(lambda x: (x[r_fld.index("run")], x[r_fld.index("energy")],
                                           x[r_fld.index("starttime")], x[r_fld.index("stoptime")],
                                           x[r_fld.index("luminosity")]), rdb_rows)
        c_data_idx = c_fld.index("data")
        compton_row_generator = map(lambda x: (x[c_fld.index("begintime")], x[c_fld.index("endtime")], 
                                               float(x[c_data_idx][0]), float(x[c_data_idx][1]), 
                                               float(x[c_data_idx][2]), float(x[c_data_idx][3])), cdb_rows)
        
        self.cur.executemany(f"""INSERT into runs VALUES ({','.join(['?']*5)})""", run_row_generator)
        self.cur.executemany(f"""INSERT into compton VALUES ({','.join(['?']*6)})""", compton_row_generator)
        
        self.__create_combined_table()        
        
    def __create_combined_table(self):
        create_combined_query = """
        CREATE TABLE combined_table AS
        SELECT 
            runs.run AS run,
            runs.elabel AS elabel,
            runs.starttime as "run_start [timestamp]", 
            runs.stoptime AS "run_stop [timestamp]", 
            compton.begintime AS "compton_start [timestamp]", 
            compton.endtime AS "compton_stop [timestamp]",
            runs.luminosity, compton.e_mean, compton.e_std, compton.spread_mean, compton.spread_std
        FROM runs, compton
        WHERE 
            (runs.starttime BETWEEN compton.begintime AND compton.endtime) 
            OR (runs.stoptime BETWEEN compton.begintime AND compton.endtime)
            OR (compton.begintime BETWEEN runs.starttime AND runs.stoptime)
            OR (compton.endtime BETWEEN runs.starttime AND runs.stoptime);
        """
        self.cur.execute(create_combined_query)
        return
        
    def combined_table(self) -> pd.DataFrame:
        """Returns combined dataframe
        """
        
        sql_query = """
        SELECT * FROM combined_table;
        """
        df = pd.read_sql(sql_query, self.conn)
        df['common_duration'] = df[['run_stop', 'compton_stop']].min(axis=1) - df[['run_start', 'compton_start']].max(axis=1)
        df['run_duration'] = df['run_stop'] - df['run_start']
        df['run_in_measurement'] = df['common_duration']/df['run_duration']
        df = df.sort_values(by='run_in_measurement', ascending=False).drop_duplicates(subset='run').sort_values(by='run')
        df = df.drop(['run_duration', 'common_duration'], axis=1) #, 'run_start', 'run_stop'
        return df
        
    def __del__(self):
        self.conn.close()
        
class Likelihood():
    """
    Likelihood function
    """
    
    def __init__(self, means: np.array, sigmas: np.array, weights: np.array):
        """
        Parameters
        ----------
        means : np.array
            array of means, [MeV]
        sigmas : np.array
            array of standard deviations, [MeV]
        weights : np.array
            array of luminosities
        """
        
        self.means = means
        self.sigmas = sigmas
        self.weights = weights/weights.mean()
        
    def __call__(self, mean: float, sigma: float):
        """
        Calls likelihood calculation
        
        Parameters
        ----------
        mean : float
            expected mean
        sigma : float
            expected standard deviation
        """
        
        sigma_total = np.sqrt(sigma**2 + self.sigmas**2)
        ln_L = -np.sum( self.weights*( ((mean - self.means)**2)/(2*(sigma_total**2)) + np.log(sigma_total) ) )
        return -ln_L
    
def __estimate_point_with_closest(comb_df: pd.DataFrame, runs_df: pd.DataFrame, compton_df: pd.DataFrame):
    # estimate energy by the nearest points
    min_run_time = runs_df[runs_df.run == comb_df.iloc[0].at['run_first']].iloc[0].at['starttime']
    max_run_time = runs_df[runs_df.run == comb_df.iloc[0].at['run_last']].iloc[0].at['stoptime']

    nearest_row_before = compton_df.iloc[pd.Index(compton_df.endtime).get_loc(min_run_time, 'nearest')]
    nearest_row_after = compton_df.iloc[pd.Index(compton_df.begintime).get_loc(max_run_time, 'nearest')]

    # regulatization
    nearest_row_before['data'][1] = max(nearest_row_before['data'][3], 1e-3)
    nearest_row_after['data'][3] = max(nearest_row_after['data'][3], 1e-3)
    nearest_row_before['data'][1] = max(nearest_row_before['data'][1], 1e-3)
    nearest_row_after['data'][3] = max(nearest_row_after['data'][3], 1e-3)

    mean_energy = (nearest_row_before['data'][0] + nearest_row_after['data'][0])/2
    mean_spread = (nearest_row_before['data'][2] + nearest_row_after['data'][2])/2
    std_energy = np.sqrt(1/(1/(nearest_row_before['data'][1])**2 + 1/(nearest_row_after['data'][1])**2))
    std_spread = np.sqrt(1/(1/(nearest_row_before['data'][3])**2 + 1/(nearest_row_after['data'][3])**2))
    sys_energy = np.std([nearest_row_before['data'][0], nearest_row_after['data'][0]])


    return {
        'energy_point': comb_df.elabel.min(),
        'first_run': comb_df.run_first.min(),
        'last_run': comb_df.run_last.max(), 
        'mean_energy': mean_energy, 
        'mean_energy_stat_err': std_energy, 
        'mean_energy_sys_err': sys_energy, 
        'mean_spread': mean_spread,
        'mean_spread_stat_err': std_spread, 
        'used_lum': 0, 
        'comment': 'indirect measurement #2',
    }, pd.DataFrame([])
    
def calculate_point(comb_df: pd.DataFrame, runs_df: pd.DataFrame, compton_df: pd.DataFrame, rdb) -> dict:
    """Calculates parameters of the energy (mean, std, spread) in this dataFrame
    
    Parameters
    ----------
    comb_df : pd.DataFrame
        table of the measurements linked with runs
    runs_df : pd.DataFrame
        table of the runs
    compton_df : pd.DataFrame
        table of the comptons
    
    Returns
    -------
    dict, pd.DataFrame
        average parameters on this DataFrame, clean dataFrame
    """
        
    if (len(comb_df) == 1) and pd.isnull(comb_df.iloc[0].at['compton_start']):
        # no direct measurements of the compton during data runs
                
        min_Yruntime = runs_df[runs_df.run == comb_df.iloc[0].at['run_first']].iloc[0].at['starttime']
        max_Yruntime = runs_df[runs_df.run == comb_df.iloc[0].at['run_last']].iloc[0].at['stoptime']
        dlt0 = timedelta(days=1)
        # assymetric time because energy can be stable only after
        runs_df_with_bads = rdb.load_tables((min_Yruntime, max_Yruntime + dlt0), energy_point = comb_df.iloc[0].at['elabel'], select_bad_runs = True)
        
        if len(runs_df_with_bads[0]) == 0:
            return __estimate_point_with_closest(comb_df, runs_df, compton_df)
        
        runs_df_with_bads_df = pd.DataFrame(runs_df_with_bads[0], columns = runs_df_with_bads[1])
        min_run_time, max_run_time = min(min_Yruntime, runs_df_with_bads_df.starttime.min()), max(max_Yruntime, runs_df_with_bads_df.stoptime.max())
        
        compton_meas = compton_df.query('((begintime>=@min_run_time)&(begintime<=@max_run_time))|((endtime>=@min_run_time)&(endtime<=@max_run_time))').copy()
        res_df = pd.DataFrame(list(map(lambda x: {
            'compton_start': x[1]['begintime'],
            'compton_stop': x[1]['endtime'],
            'e_mean': float(x[1]['data'][0]),
            'e_std': float(x[1]['data'][1]),
            'spread_mean': float(x[1]['data'][2]),
            'spread_std': float(x[1]['data'][3]),
        }, compton_meas.iterrows())))
        res_df = res_df.query(f'abs(e_mean -{comb_df.iloc[0].at["elabel"]})<5')
        
        if len(res_df) == 0:
            return __estimate_point_with_closest(comb_df, runs_df, compton_df)
                        
        return {
            'energy_point': comb_df.elabel.min(),
            'first_run': comb_df.run_first.min(),
            'last_run': comb_df.run_last.max(), 
            'mean_energy': res_df.e_mean.mean(), 
            'mean_energy_stat_err': np.sqrt(1/np.sum(1/(res_df.e_std)**2)),
            'mean_energy_sys_err': np.abs(comb_df.iloc[0].at['elabel'] - res_df.e_mean.mean()), 
            'mean_spread': res_df.spread_mean.mean(),
            'mean_spread_stat_err':np.sqrt(1/np.sum(1/(res_df.spread_std)**2)),
            'used_lum': 0, 
            'comment': 'indirect measurement #1',
        }, res_df
        
    
    df = comb_df.loc[~comb_df.compton_start.isna()].copy()
    df.spread_std = np.where(df.spread_std < 1e-4, 1e-4, df.spread_std)
    
    df = df[df.e_std > 0]
    mean_energy = np.sum(df.e_mean*df.luminosity/(df.e_std**2))/np.sum(df.luminosity/(df.e_std**2))
    # std_energy = np.sqrt(1/np.sum((df.luminosity/df.luminosity.mean())/df.e_std**2))
    
    good_criterion = np.abs((df.e_mean - mean_energy)/np.sqrt(df.e_mean.std()**2 + df.e_std**2)) < 5
    #print(df[~good_criterion].elabel.value_counts())
    df = df[good_criterion]
    
    m = Minuit(Likelihood(df.e_mean, df.e_std, df.luminosity), mean=df.e_mean.mean(), sigma=df.e_mean.std())
    m.errordef = 0.5 
    m.limits['sigma'] = (0, None)
    m.migrad();
    sys_err = m.values['sigma']
    mean_en = m.values['mean']
    
    mean_spread = np.sum(df.spread_mean*df.luminosity/(df.spread_std**2))/np.sum(df.luminosity/(df.spread_std**2))
    std_spread = np.sqrt(1/np.sum((df.luminosity/df.luminosity.mean())/df.spread_std**2))
    
    res_dict = {
        'energy_point': comb_df.elabel.min(), 
        'first_run': comb_df.run_first.min(),
        'last_run': comb_df.run_last.max(), 
        'mean_energy': mean_en, 
        'mean_energy_stat_err': m.errors['mean'], 
        'mean_energy_sys_err': sys_err, 
        'mean_spread': mean_spread,
        'mean_spread_stat_err': std_spread, 
        'used_lum': df.luminosity.sum()/comb_df.luminosity_total.sum(), 
        'comment': '',
    }
    return res_dict, df
    
def process_intersected_compton_meas(combined_df: pd.DataFrame) -> pd.DataFrame:
    """Replaces compton measurements writed on the border of two energy points on NaNs
    """
    
    energy_point_borders = combined_df[['point_idx', 'elabel', 'run_start', 'run_stop']].groupby(['point_idx'], dropna=True).agg(
        elabel_start_time=('run_start', 'min'), elabel_stop_time=('run_stop', 'max'),
    )
    df_comb = combined_df.set_index('point_idx').join(energy_point_borders, how='left')
    
    df_comb['comptonmeas_in_elabel'] = (df_comb[['elabel_stop_time', 'compton_stop']].min(axis=1) - df_comb[['elabel_start_time', 'compton_start']].max(axis=1))/(df_comb['compton_stop'] - df_comb['compton_start'])
    #print(df_comb['comptonmeas_in_elabel'].dropna().sort_values())
    df_comb = df_comb.query('comptonmeas_in_elabel < 0.7')
    border_comptons = df_comb.compton_start.values
    #print(combined_df.compton_start.isin(border_comptons).sum())
    #print(combined_df.loc[combined_df.compton_start.isin(border_comptons)].elabel.value_counts())
    
    combined_df.loc[combined_df.compton_start.isin(border_comptons), ['compton_start', 'compton_stop', 'e_mean', 'e_std', 'spread_mean', 'spread_std']] = np.nan
    return combined_df

def process_combined(combined_df: pd.DataFrame, runs_df: pd.DataFrame, compton_df: pd.DataFrame, pics_folder: Optional[str] = None, rdb: Optional[RunsDBHandler] = None) -> pd.DataFrame:
    
    if pics_folder is not None:
        plt.ioff()
        plt.style.use('ggplot')
        locator = mdates.AutoDateLocator(minticks=5)
        formatter = mdates.ConciseDateFormatter(locator)
        formatter.formats = ['%y', '%b', '%d', '%H:%M', '%H:%M', '%S.%f', ]
        formatter.zero_formats = [''] + formatter.formats[:-1]
        formatter.zero_formats[3] = '%d-%b'
        formatter.offset_formats = ['', '%Y', '%b %Y', '%d %b %Y', '%d %b %Y', '%d %b %Y %H:%M', ]
    
    runs_df = runs_df.rename({'luminosity': 'luminosity_full', 'energy': 'elabel'}, axis=1)
    combined_df = pd.merge(combined_df.drop(['elabel'], axis=1), runs_df[['run', 'elabel', 'luminosity_full']], how='outer')
    combined_df = combined_df.sort_values(by='run')
    combined_df['luminosity'] = combined_df['luminosity'].fillna(0)
    
    combined_df['point_idx'] = np.cumsum(~np.isclose(combined_df.elabel, combined_df.elabel.shift(1), atol=1e-4))
    combined_df = process_intersected_compton_meas(combined_df)
    
    combined_df = combined_df.groupby(['point_idx', 'compton_start'], dropna=False).agg(
        elabel=('elabel', 'min'), elabel_test=('elabel', 'max'),
        run_first=('run', 'min'), run_last=('run', 'max'),
        luminosity=('luminosity', 'sum'), luminosity_total=('luminosity_full', 'sum'),
        compton_stop=('compton_stop', 'min'), compton_stop_test=('compton_stop', 'max'),
        e_mean=('e_mean', 'min'), e_mean_test=('e_mean', 'max'),
        e_std=('e_std', 'min'), e_std_test=('e_std', 'max'),
        spread_mean=('spread_mean', 'min'), spread_mean_test=('spread_mean', 'max'),
        spread_std=('spread_std', 'min'), spread_std_test=('spread_std', 'max'),
    ).reset_index().set_index('point_idx')
    #return combined_df
    
    result_df = pd.DataFrame(columns=['energy_point', 'first_run', 'last_run', 'mean_energy', 'mean_energy_stat_err', 'mean_energy_sys_err', 'mean_spread', 'mean_spread_stat_err', 'used_lum', 'comment'])
    
    for i, table in tqdm(combined_df.groupby('point_idx', dropna=False)):
        res_dict, good_df = calculate_point(table, runs_df, compton_df, rdb)
        result_df = result_df.append(res_dict, ignore_index=True)
        
        if pics_folder is not None:
            plt_table = good_df.dropna()
            if len(plt_table) == 0:
                continue
            
            total_error = np.sqrt(res_dict["mean_energy_stat_err"]**2 + res_dict["mean_energy_sys_err"]**2)
            half_timedelta = (plt_table.compton_stop - plt_table.compton_start)/2 
            time = plt_table.compton_start + half_timedelta
            dlt0, total_time = timedelta(days=1), plt_table.compton_stop.max() - plt_table.compton_stop.min()
            timelim = [plt_table.compton_start.min() - 0.05*total_time, plt_table.compton_stop.max() + 0.05*total_time]
            
            fig, ax = plt.subplots(1, 1, dpi=120, tight_layout=True)
            ax.errorbar(time, plt_table.e_mean, xerr=half_timedelta, yerr=plt_table.e_std, fmt='.')
            ax.axhline(res_dict['mean_energy'], color='black', zorder=3, label='Mean')
            ax.fill_between(timelim, 
                            [res_dict['mean_energy'] - total_error]*2, 
                            [res_dict['mean_energy'] + total_error]*2, color='green', zorder=1, alpha=0.4)
            ax.tick_params(axis='x', labelrotation=45)
            ax.xaxis.set_major_locator(locator)
            ax.xaxis.set_major_formatter(formatter)
            ax.set(title=f'{res_dict["energy_point"]}, E = {res_dict["mean_energy"]:.3f} ± {res_dict["mean_energy_stat_err"]:.3f} ± {res_dict["mean_energy_sys_err"]:.3f} MeV', 
                   xlabel='Time, NSK', ylabel='Energy, [MeV]', xlim=timelim)
            plt.savefig(f'{pics_folder}/{res_dict["first_run"]}_{res_dict["energy_point"]}.png', transparent=True)
            plt.close()
    
    return result_df

def final_table_to_clbrdb(df: pd.DataFrame, clbrdb: CalibrdbHandler, runs_df: pd.DataFrame, season: str):
    """Write good values from the averaged table into clbrdb
    """
    
    good_values = (df.comment=='')|((df.comment!='')&((df.mean_energy.astype(float) - df.energy_point).abs()<5))
    df_clbrdb = df.loc[good_values].drop(['comment', 'used_lum'], axis=1)
    
    df_clbrdb = pd.merge(df_clbrdb, runs_df[['run', 'starttime']], how='left', left_on='first_run', right_on='run').drop(['run'], axis=1)
    df_clbrdb = pd.merge(df_clbrdb, runs_df[['run', 'stoptime']], how='left', left_on='last_run', right_on='run').drop(['run'], axis=1)
    
    df_clbrdb = df_clbrdb.assign(writetime=lambda df: df['stoptime'])
    df_clbrdb = df_clbrdb[['writetime', 'starttime', 'stoptime',
                           'energy_point', 'first_run', 'last_run', 'mean_energy', 
                           'mean_energy_stat_err', 'mean_energy_sys_err', 'mean_spread', 'mean_spread_stat_err']].values.tolist()
    clbrdb.insert(df_clbrdb, 'Misc', 'RunHeader', 'Compton_run_avg', 'Default', comment = season)
    clbrdb.commit()
    
def save_csv(df: pd.DataFrame, filepath: str):
    """Saves csv file. Updates current file in filepath if exists"""
    
    if os.path.isfile(filepath):
        df_current = pd.read_csv(filepath)
        df_current = df_current.append(df, ignore_index=True)
        df_current = df_current.drop_duplicates(subset=['energy_point', 'first_run'], keep='last')
        df = df_current
    
    df.to_csv(filepath, index=False, float_format='%g')
    return

# python scripts/compton_combiner.py -s NNBAR2021 -c database.ini --csv_dir . --clbrdb
def main():
    parser = argparse.ArgumentParser(description = 'Mean compton energy measurements from clbrdb')
    parser.add_argument('-s', '--season', help = 'Name of the season')
    parser.add_argument('-c', '--config', help = 'Config file containing information for access to databases')
    parser.add_argument('--csv_dir', help = 'Save csv file with data in the folder or not if skip it')
    parser.add_argument('--clbrdb', action = 'store_true', help = 'Update Compton_run_avg clbrdb or not')
    parser.add_argument('--pics_folder', help = 'Path to the directory for saving the pictures')
    parser.add_argument('--only_last', action = 'store_true', help = 'Compute values of the last (in Compton_run_avg clbrdb) and new points only')
    
    args = parser.parse_args()
    # logging.info(f"Arguments: season: {args.season}, config {args.config}")

    parser = ConfigParser()
    parser.read(args.config);
    
    rdb = RunsDBHandler(**parser['cmdruns'])
    clbrdb = CalibrdbHandler(**parser['clbrDB'])
    
    idx = SEASONS['name'].index(args.season)
    runs_range = (SEASONS['start_run'][idx], SEASONS['start_run'][idx+1])
    
    if args.only_last:
        res_avg = clbrdb.load_table('Misc', 'RunHeader', 'Compton_run_avg', num_last_rows = 1)
        if len(res_avg[0]) != 0:
            begintime = res_avg[0][0][res_avg[1].index("begintime")]
            runs_range = (begintime, None)
    
    res_rdb = rdb.load_tables(runs_range)
    runs_df = pd.DataFrame(res_rdb[0], columns=res_rdb[1])
    
    tdlt0 = timedelta(days=2)
    time_range = (runs_df.starttime.min() - tdlt0, runs_df.stoptime.max() + tdlt0)
    
    res_clbrdb = clbrdb.load_table('Misc', 'RunHeader', 'Compton_run', num_last_rows = None, timerange = time_range)
    
    cb = Combiner(res_rdb, res_clbrdb)
    comb_df = cb.combined_table()
    
    compton_df = pd.DataFrame(res_clbrdb[0], columns=res_clbrdb[1])
    
    cdf = process_combined(comb_df, runs_df, compton_df, args.pics_folder, rdb)
    
    if args.csv_dir is not None:
        csv_path = os.path.join(args.csv_dir, f'{args.season}.csv')
        save_csv(cdf, csv_path)
        # cdf.to_csv(f'{args.season}.csv', index=False, float_format='%g')
    
    if args.clbrdb:
        final_table_to_clbrdb(cdf, clbrdb, runs_df, args.season)
    return 

if __name__ == "__main__":
    main()