Initial commit

CrabNebula.yaml

+data_files:
+    mc:
+        train_sample: 
+            magic1:
+                input_mask: "../../../MCs/MAGIC/ST.03.07/za05to35/Train_sample/1.Calibrated/GA_M1*root"
+                hillas_output: "../../../MCs/MAGIC/ST.03.07/za05to35/Train_sample/2.Hillas/iv_hillas_m1.h5"
+            magic2:
+                input_mask: "../../../MCs/MAGIC/ST.03.07/za05to35/Train_sample/1.Calibrated/GA_M2*root"
+                hillas_output: "../../../MCs/MAGIC/ST.03.07/za05to35/Train_sample/2.Hillas/iv_hillas_m2.h5"
+                
+        test_sample: 
+            magic1:
+                input_mask: "../../../MCs/MAGIC/ST.03.07/za05to35/Test_sample/1.Calibrated/GA_M1*root"
+                hillas_output: "../../../MCs/MAGIC/ST.03.07/za05to35/Test_sample/2.Hillas/iv_hillas_m1.h5"
+                reco_output: "../../../MCs/MAGIC/ST.03.07/za05to35/Test_sample/3.Reco/iv_reco_m1.h5"
+            magic2:
+                input_mask: "../../../MCs/MAGIC/ST.03.07/za05to35/Test_sample/1.Calibrated/GA_M2*root"
+                hillas_output: "../../../MCs/MAGIC/ST.03.07/za05to35/Test_sample/2.Hillas/iv_hillas_m2.h5"
+                reco_output: "../../../MCs/MAGIC/ST.03.07/za05to35/Test_sample/3.Reco/iv_reco_m2.h5"
+                
+    data:
+       train_sample: 
+           magic1:
+               input_mask: "../../../Data/MAGIC/Off/Train_sample/1.Calibrated/20*_M1_*root"
+               hillas_output: "../../../Data/MAGIC/Off/Train_sample/2.Hillas/iv_hillas_m1.h5"
+           magic2:
+               input_mask: "../../../Data/MAGIC/Off/Train_sample/1.Calibrated/20*_M2_*root"
+               hillas_output: "../../../Data/MAGIC/Off/Train_sample/2.Hillas/iv_hillas_m2.h5"
+               
+       test_sample: 
+           magic1:
+               input_mask: "../../../Data/MAGIC/CrabNebula/1.Calibrated/20*_M1_*root"
+               hillas_output: "../../../Data/MAGIC/CrabNebula/ctapipe/2.Hillas/iv_hillas_m1.h5"
+               reco_output: "../../../Data/MAGIC/CrabNebula/ctapipe/3.Reco/reco_m1.h5"
+           magic2:
+               input_mask: "../../../Data/MAGIC/CrabNebula/1.Calibrated/20*_M2_*root"
+               hillas_output: "../../../Data/MAGIC/CrabNebula/ctapipe/2.Hillas/iv_hillas_m2.h5"
+               reco_output: "../../../Data/MAGIC/CrabNebula/ctapipe/3.Reco/reco_m2.h5"
+                
+
+image_cleanining:
+    magic:
+        charge_thresholds:
+            picture_thresh: 6
+            boundary_thresh: 3.5
+            min_number_picture_neighbors: 1
+
+        time_thresholds:
+            # 1.5 slices x 1.64 GHz
+            time_limit: 2.46
+            # 4.5 slices x 1.64 GHz
+            max_time_off: 7.38
+            min_number_neighbors: 1
+
+energy_rf:
+    save_name: "RFs/energy_rf.joblib"
+    cuts: "(multiplicity > 1) & (intensity > 30) & (length > 0.0) & (leakage1_intensity < 0.15)"
+    settings:
+        n_estimators: 100
+        min_samples_leaf: 10
+        n_jobs: 3
+    
+    features:
+        ['slope', 'length', 'width', 'intensity', 'skewness', 'kurtosis', 'x', 'y', 'psi', 'leakage1_intensity', 'leakage2_intensity']
+            
+direction_rf:
+    save_name: "RFs/direction_rf.joblib"
+    cuts: "(multiplicity > 1) & (intensity > 30) & (length > 0.0) & (leakage1_intensity < 0.15)"
+    settings:
+        n_estimators: 100
+        min_samples_leaf: 10
+        n_jobs: 3
+    
+    features:
+        'disp': ['slope', 'length', 'width', 'intensity', 'skewness', 'kurtosis', 'x', 'y', 'psi']
+        'pos_angle_shift': ['slope', 'length', 'width', 'intensity', 'skewness', 'kurtosis', 'x', 'y', 'psi']
+
+classifier_rf:
+    save_name: "RFs/classifier_rf.joblib"
+    cuts: "(multiplicity > 1) & (intensity > 30) & (length > 0.0) & (leakage1_intensity < 0.15)"
+    settings:
+        n_estimators: 100
+        min_samples_leaf: 10
+        n_jobs: 3
+    
+    features:
+        ['slope', 'length', 'width', 'intensity', 'skewness', 'kurtosis', 'x', 'y', 'psi', 'n_islands']
+
+event_list:
+    max_time_diff: 6.9e-4
+    cuts:
+        l3rate: "80 < value < 1000"
+        dc: "2.5e3 < value < 3.5e3"

gti_generator.py

+# coding: utf-8
+
+import glob
+import yaml
+import argparse
+import datetime
+
+import scipy
+import uproot
+import pandas
+
+
+def info_message(text, prefix='info'):
+    """
+    This function prints the specified text with the prefix of the current date
+
+    Parameters
+    ----------
+    text: str
+
+    Returns
+    -------
+    None
+
+    """
+
+    date_str = datetime.datetime.now().strftime("%Y-%m-%dT%H:%M:%S")
+    print(f"({prefix:s}) {date_str:s}: {text:s}")
+
+
+def identify_time_edges(times, criterion, max_time_diff=6.9e-4):
+    """
+    Identifies the time interval edges, corresponding to the True
+    state of the specified condition. Neighbouring time intervals, 
+    separated by not more than max_time_diff are joined together.
+
+    Parameters
+    ----------
+    times: array_like
+        Array of the time data points.
+    criterion: array_like
+        Array of True/False values, indicating the goodness of the 
+        corresponding data points.
+    max_time_diff: float, optional
+        Maximal time difference between the time intervals, below which 
+        they are joined into one.
+
+    Returns
+    -------
+    parts_edges: list
+        List of start/stop pairs, describing the identified time intervals.
+
+    """
+
+    times = scipy.array(times)
+    wh = scipy.where(criterion == True)
+    
+    if len(wh[0]) == 0:
+        print("No time intervals excluded!")
+        
+        return [[0,0]]
+    
+    # The above function will result in indicies of non-zero bins.
+    # But we want indicies of their _edges_.
+    non_zero_bin_edges = []
+    
+    for i in wh[0]:
+        if i-1 >= 0:
+            if abs(times[i]-times[i-1]) < max_time_diff:
+                non_zero_bin_edges.append(i-1)
+        
+        non_zero_bin_edges.append(i)
+        
+        if i+1 < len(times):
+            if abs(times[i]-times[i+1]) < max_time_diff:
+                non_zero_bin_edges.append(i+1)
+        
+    non_zero_bin_edges = scipy.unique(non_zero_bin_edges)
+    
+    if len(non_zero_bin_edges) > 2:
+        # Finding the elements, that separate the observational time intervals
+        # During one time interval diff should return 1 (current bin has index i, the index of the next time bin is i+1).
+        # Here we're looking for diffs that are not equal to 1.
+        #division_indicies = (scipy.diff(non_zero_bin_edges[1:-1])-1).nonzero()
+        #division_indicies = division_indicies[0]
+        cond_id = scipy.diff(non_zero_bin_edges[1:-1]) > 1
+        cond_time = scipy.diff(times[non_zero_bin_edges[1:-1]]) > max_time_diff
+        division_indicies = scipy.where(cond_id | cond_time)
+        division_indicies = division_indicies[0]
+        
+        # Concatenating to the found elements the beginning and the end of the observational time.
+        # Also adding i+1 elements, to correctly switch to the next observational time interval.
+        parts_edges_idx = scipy.concatenate(
+                (
+                    [non_zero_bin_edges[0]], 
+                    non_zero_bin_edges[1:-1][division_indicies], 
+                    non_zero_bin_edges[1:-1][division_indicies+1], 
+                    [non_zero_bin_edges[-1]]
+                )
+            )
+    else:
+        parts_edges_idx = scipy.array(non_zero_bin_edges)
+    parts_edges_idx.sort()
+    
+    # Transorming edges indicies to the real values and transforming them to the [start, stop] list.
+    parts_edges = times[parts_edges_idx]
+    parts_edges = parts_edges.reshape(-1,2)
+    
+    return parts_edges
+
+
+def intersect_time_intervals(intervals1, intervals2):
+    """
+    Intersects two lists of (TStart, TStop) pairs. Returned list 
+    contains the start/stop invervals, common in both input lists.
+
+    Parameters
+    ----------
+    interval1: list
+        First list of (TStart, TStop) lists (or tuples).
+    interval2: list
+        Second list of (TStart, TStop) lists (or tuples).
+
+    Returns
+    -------
+    joint_intervals: list
+        A list of (TStart, TStop) lists, representing the start/stop invervals, 
+        common in both input lists.
+
+    """
+
+
+    joined_intervals  = []
+    
+    # Comparing 1st to 2nd
+    for interv1 in intervals1:
+        tstart = None
+        tstop = None
+        
+        for interv2 in intervals2:
+            if (interv2[0] >= interv1[0]) and (interv2[0] <= interv1[1]):
+                tstart = interv2[0]
+                tstop = min(interv2[1], interv1[1])
+            elif (interv2[1] >= interv1[0]) and (interv2[1] <= interv1[1]):
+                tstart = max(interv2[0], interv1[0])
+                tstop = interv2[1]
+            elif (interv2[0] < interv1[0]) and (interv2[1] > interv1[1]):
+                tstart = interv1[0]
+                tstop = interv1[1]
+
+        if tstart and tstop:
+            joined_intervals.append([tstart, tstop])
+ 
+    return joined_intervals
+
+
+def gti_generator(config):
+    """
+    GTI list generator.
+
+    Parameters
+    ----------
+    config: dict
+        Configuration dictionary (e.g. read from a YAML file).
+
+    Returns
+    -------
+    joint_intervals: list
+        A list of (TStart, TStop) lists, representing the identified GTIs.
+
+    """
+
+    if 'event_list' not in config:
+        raise ValueError('GTI generator error: the configuration file is missing the "event_list" section. Exiting.')
+
+    # Identifying the files to read
+    info_message("looking for the files", "GTI generator")
+    file_mask = config["data_files"]["data"]["test_sample"]["magic2"]["input_mask"]
+
+    file_list = glob.glob(file_mask)
+
+    if not file_list:
+        raise ValueError("No files to process")
+
+    # Containers for the data points
+    dfs = {
+        'dc': pandas.DataFrame(),
+        'l3rate': pandas.DataFrame()
+    }
+
+    # Removing the containers, not specified in the configuration card
+    if "l3rate" not in config['event_list']['cuts']:
+        del dfs['l3rate']
+
+    if "dc" not in config['event_list']['cuts']:
+        del dfs['dc']
+
+    # Looping over the data files
+    for fnum, file_name in enumerate(file_list):
+        info_message(f"processing file {fnum+1} / {len(file_list)}", "GTI generator")
+
+        with uproot.open(file_name) as input_stream:
+            
+            # --- DC ---
+            if "dc" in dfs:
+                mjd = input_stream["Camera"]["MTimeCamera.fMjd"].array()
+                millisec = input_stream["Camera"]["MTimeCamera.fTime.fMilliSec"].array()
+                nanosec = input_stream["Camera"]["MTimeCamera.fNanoSec"].array()
+
+                df_ = pandas.DataFrame()
+                
+                df_['mjd'] = mjd + (millisec / 1e3 + nanosec / 1e9) / 86400
+                df_['value'] = input_stream["Camera"]["MReportCamera.fMedianDC"].array()
+                
+                dfs['dc'] = dfs['dc'].append(df_)
+            
+            # --- L3 rate ---
+            if "l3rate" in dfs:
+                mjd = input_stream["Trigger"]["MTimeTrigger.fMjd"].array()
+                millisec = input_stream["Trigger"]["MTimeTrigger.fTime.fMilliSec"].array()
+                nanosec = input_stream["Trigger"]["MTimeTrigger.fNanoSec"].array()
+
+                df_ = pandas.DataFrame()
+                
+                df_['mjd'] = mjd + (millisec / 1e3 + nanosec / 1e9) / 86400
+                df_['value'] = input_stream["Trigger"]["MReportTrigger.fL3Rate"].array()
+                
+                
+                dfs['l3rate'] = dfs['l3rate'].append(df_)
+
+    # Sorting data points by date is needed for the time intervals identification
+    for key in dfs:
+        dfs[key] = dfs[key].sort_values(by=['mjd'])
+
+    info_message("identifying GTIs", "GTI generator")
+
+    time_intervals_list = []
+
+    # Identifying DC-related GTIs
+    if "dc" in dfs:
+        cut = config['event_list']['cuts']['dc']
+        
+        selection = dfs['dc'].query(cut)
+
+        _, idx, _ = scipy.intersect1d(dfs['dc']["mjd"], selection["mjd"], return_indices=True)
+
+        criterion = scipy.repeat(False, len(dfs['dc']["mjd"]))
+        criterion[idx] = True
+
+        time_intervals = identify_time_edges(dfs['dc']["mjd"], criterion, max_time_diff=config['event_list']['max_time_diff'])
+
+        time_intervals_list.append(time_intervals)
+
+    # Identifying L3-related GTIs
+    if "l3rate" in dfs:
+        cut = config['event_list']['cuts']['l3rate']
+        
+        selection = dfs['l3rate'].query(cut)
+
+        _, idx, _ = scipy.intersect1d(dfs['l3rate']["mjd"], selection["mjd"], return_indices=True)
+
+        criterion = scipy.repeat(False, len(dfs['l3rate']["mjd"]))
+        criterion[idx] = True
+
+        time_intervals = identify_time_edges(dfs['l3rate']["mjd"], criterion, max_time_diff=config['event_list']['max_time_diff'])
+
+        time_intervals_list.append(time_intervals)
+
+    if not len(time_intervals_list):
+        raise ValueError("No valid selection cuts provided in the configuration file.")
+
+    joint_intervals = time_intervals_list[0]
+
+    # Joining all found GTIs
+    for i in range(1, len(time_intervals_list)):
+        joint_intervals = intersect_time_intervals(joint_intervals, time_intervals_list[i])
+
+    return joint_intervals
+    
+
+# =================
+# === Main code ===
+# =================
+
+if __name__ == "__main__":
+    # --------------------------
+    # Adding the argument parser
+    arg_parser = argparse.ArgumentParser(description="""
+    This tools fits the direction random forest regressor on the specified events files.
+    """)
+
+    arg_parser.add_argument("--config", default="config.yaml",
+                            help='Configuration file to steer the code execution.')
+
+    parsed_args = arg_parser.parse_args()
+    # --------------------------
+
+    # ------------------------------
+    # Reading the configuration file
+
+    file_not_found_message = """
+    Error: can not load the configuration file {:s}.
+    Please check that the file exists and is of YAML or JSON format.
+    Exiting.
+    """
+
+    try:
+        config = yaml.load(open(parsed_args.config, "r"))
+    except IOError:
+        print(file_not_found_message.format(parsed_args.config))
+        exit()
+
+    if 'event_list' not in config:
+        print('Error: the configuration file is missing the "event_list" section. Exiting.')
+        exit()
+    # ------------------------------
+
+    joint_intervals = gti_generator(config)
+
+    print("--- Identified GTI intervals [MJD] ---")
+    for interval in joint_intervals:
+        print(interval)
\ No newline at end of file

train_classifier_rf.py

+# coding: utf-8
+
+import argparse
+import yaml
+import datetime
+
+import scipy
+import pandas as pd
+import numpy as np
+
+import sklearn.metrics
+
+from ctapipe.reco.event_processing import EventClassifierPandas
+
+from astropy import units as u
+
+from matplotlib import pyplot, colors
+
+from MyFunctions import GetHist
+
+
+def info_message(text, prefix='info'):
+    """
+    This function prints the specified text with the prefix of the current date
+
+    Parameters
+    ----------
+    text: str
+
+    Returns
+    -------
+    None
+
+    """
+
+    date_str = datetime.datetime.now().strftime("%Y-%m-%dT%H:%M:%S")
+    print(f"({prefix:s}) {date_str:s}: {text:s}")
+    
+
+def evaluate_performance(data, class0_name='event_class_0'):
+    data = data.dropna()
+    
+    report = {
+        "gammaness": dict(),
+        "metrics": dict()
+    }
+    
+    for event_class in data['true_event_class'].unique():
+        events = data.query(f'true_event_class == {event_class}')
+        hist = GetHist(events[class0_name], bins=100, range=(0, 1))
+        hist['Hist'] = hist['Hist'] / hist['Hist'].sum()
+        hist['Cumsum'] = 1 - scipy.cumsum(hist['Hist'])
+        
+        report['gammaness'][event_class] = hist
+
+    if 'mean' in class0_name:
+        class_names = list(filter(lambda name: 'event_class_' in name and '_mean' in name, data.columns))
+    else:
+        class_names = list(filter(lambda name: 'event_class_' in name and '_mean' not in name, data.columns))
+        
+    proba = data[class_names].values
+    predicted_class = proba.argmax(axis=1)
+
+    report['metrics']['acc'] = sklearn.metrics.accuracy_score(data['true_event_class'], predicted_class)
+    
+    true_class = scipy.clip(data['true_event_class'], 0, 1)
+    true_class = 1 - true_class
+    report['metrics']['auc_roc'] = sklearn.metrics.roc_auc_score(true_class, proba[:, 0])
+
+    return report
+
+
+def load_data_sample(sample):
+    shower_data = pd.DataFrame()
+
+    for telescope in sample:
+        info_message(f'Loading {telescope} data...', prefix='ClassifierRF')
+        
+        hillas_data = pd.read_hdf(sample[telescope]['hillas_output'], key='dl1/hillas_params')
+        hillas_data.set_index(['obs_id', 'event_id', 'tel_id'], inplace=True)
+        
+        shower_data = shower_data.append(hillas_data)
+        
+    shower_data.sort_index(inplace=True)
+    
+    return shower_data
+
+
+def get_weights(mc_data, bkg_data, alt_edges, intensity_edges):
+    mc_hist, _, _ = scipy.histogram2d(mc_data['tel_alt'], 
+                                      mc_data['intensity'], 
+                                      bins=[alt_edges, intensity_edges])
+    bkg_hist, _, _ = scipy.histogram2d(bkg_data['tel_alt'], 
+                                       bkg_data['intensity'], 
+                                       bins=[alt_edges, intensity_edges])
+    
+    availability_hist = scipy.clip(mc_hist, 0, 1) * scipy.clip(bkg_hist, 0, 1)
+    
+    # --- MC weights ---
+    mc_alt_bins = scipy.digitize(mc_data['tel_alt'], alt_edges) - 1
+    mc_intensity_bins = scipy.digitize(mc_data['intensity'], intensity_edges) - 1
+    
+    # Treating the out-of-range events
+    mc_alt_bins[mc_alt_bins == len(alt_edges) - 1] = len(alt_edges) - 2
+    mc_intensity_bins[mc_intensity_bins == len(intensity_edges) - 1] = len(intensity_edges) - 2
+
+    mc_weights = 1 / mc_hist[mc_alt_bins, mc_intensity_bins]
+    mc_weights *= availability_hist[mc_alt_bins, mc_intensity_bins]
+    
+    # --- Bkg weights ---
+    bkg_alt_bins = scipy.digitize(bkg_data['tel_alt'], alt_edges) - 1
+    bkg_intensity_bins = scipy.digitize(bkg_data['intensity'], intensity_edges) - 1
+
+    # Treating the out-of-range events
+    bkg_alt_bins[bkg_alt_bins == len(alt_edges) - 1] = len(alt_edges) - 2
+    bkg_intensity_bins[bkg_intensity_bins == len(intensity_edges) - 1] = len(intensity_edges) - 2
+
+    bkg_weights = 1 / bkg_hist[bkg_alt_bins, bkg_intensity_bins] 
+    bkg_weights *= availability_hist[bkg_alt_bins, bkg_intensity_bins]
+    
+    # --- Storing to a data frame ---
+    mc_weight_df = pd.DataFrame(data={'event_weight': mc_weights}, 
+                                index=mc_data.index)
+    bkg_weight_df = pd.DataFrame(data={'event_weight': bkg_weights}, 
+                                 index=bkg_data.index)
+    
+    return mc_weight_df, bkg_weight_df
+
+# =================
+# === Main code ===
+# =================
+
+# --------------------------
+# Adding the argument parser
+arg_parser = argparse.ArgumentParser(description="""
+This tools fits the event classification random forest on the specified events files.
+""")
+
+arg_parser.add_argument("--config", default="config.yaml",
+                        help='Configuration file to steer the code execution.')
+
+parsed_args = arg_parser.parse_args()
+# --------------------------
+
+# ------------------------------
+# Reading the configuration file
+
+file_not_found_message = """
+Error: can not load the configuration file {:s}.
+Please check that the file exists and is of YAML or JSON format.
+Exiting.
+"""
+
+try:
+    config = yaml.load(open(parsed_args.config, "r"))
+except IOError:
+    print(file_not_found_message.format(parsed_args.config))
+    exit()
+
+if 'classifier_rf' not in config:
+    print('Error: the configuration file is missing the "classifier_rf" section. Exiting.')
+    exit()
+# ------------------------------
+
+
+# --------------------
+# --- Train sample ---
+info_message('Loading MC train data...', prefix='ClassifierRF')
+mc_data = load_data_sample(config['data_files']['mc']['train_sample'])
+
+info_message('Loading "off" train data...', prefix='ClassifierRF')
+bkg_data = load_data_sample(config['data_files']['data']['train_sample'])
+
+# True event classes
+mc_data['true_event_class'] = 0
+bkg_data['true_event_class'] = 1
+
+# Dropping data with the wrong altitude
+bkg_data = bkg_data.query('tel_alt < 1.5707963267948966')
+
+# Dropping extra keys
+bkg_data.drop('mjd', axis=1, inplace=True)
+mc_data.drop(['true_energy', 'true_alt', 'true_az'], axis=1, inplace=True)
+
+# Computing event weights
+sin_edges = scipy.linspace(0, 1, num=51)
+alt_edges = scipy.arcsin(sin_edges)
+intensity_edges = scipy.logspace(1, 5, num=51)
+
+mc_weights, bkg_weights = get_weights(mc_data, bkg_data, 
+                                      alt_edges, intensity_edges)
+
+mc_data = mc_data.join(mc_weights)
+bkg_data = bkg_data.join(bkg_weights)
+
+# Merging the train sample
+shower_data_train = mc_data.append(bkg_data)