DSSC.py

import multiprocessing
from time import strftime
import tempfile
import shutil
from tqdm.auto import tqdm
import os
import warnings

import karabo_data as kd
from karabo_data.geometry2 import DSSC_1MGeometry
import ToolBox as tb
import numpy as np
import xarray as xr
import matplotlib.pyplot as plt
import h5py


class DSSC:
    
    def __init__(self, semester, proposal, topic='SCS'):
        """ Create a DSSC object to process DSSC data.
        
            inputs:
                semester: semester string
                proposal: proposal number string
                topic: topic, by default SCS
            
        """
        self.semester = semester
        self.proposal = proposal
        self.topic = topic
        self.tempdir = None
        self.save_folder = f'/gpfs/exfel/exp/{self.topic}/{self.semester}/{self.proposal}/usr/condensed_runs/'
        
        print('DSSC configuration')
        print(f'Topic: {self.topic}')
        print(f'Semester: {self.semester}')
        print(f'Proposal: {self.proposal}')
        print(f'Default save folder: {self.save_folder}')
        
        if not os.path.exists(self.save_folder):
            warnings.warn(f'Default save folder does not exist: {self.save_folder}')
        
    def __del__(self):
        # deleting temporay folder
        if self.tempdir:
            shutil.rmtree(self.tempdir)
    
    def open_run(self, run_nr, isDark=False):
        """ Open a run with karabo-data and prepare the virtual dataset for multiprocessing
        
            inputs:
                run_nr: the run number
                isDark: a boolean to specify if the run is a dark run or not
        
        """
        
        print('Opening run data with karabo-data')
        self.run_nr = run_nr
        self.xgm = None
        self.scan_vname = None
        
        self.run = kd.open_run(self.proposal, self.run_nr)
        self.isDark = isDark
        self.plot_title = f'{self.semester} run: {self.run_nr}'
        
        self.fpt = self.run.detector_info('SCS_DET_DSSC1M-1/DET/0CH0:xtdf')['frames_per_train']
        self.nbunches = self.run.get_array('SCS_RR_UTC/MDL/BUNCH_DECODER', 'sase3.nPulses.value')
        self.nbunches = np.unique(self.nbunches)
        if len(self.nbunches) == 1:
            self.nbunches = self.nbunches[0]
        else:
            warnings.warn('not all trains have same length DSSC data')
            print(f'nbunches: {self.nbunches}')
            self.nbunches = self.nbunches[-1]
            
        print(f'DSSC frames per train: {self.fpt}')
        print(f'SA3 bunches per train: {self.nbunches}')
        
        if self.tempdir is not None:
            shutil.rmtree(self.tempdir)
        
        self.tempdir = tempfile.mkdtemp()
        print(f'Temporary directory: {self.tempdir}')

        print('Creating virtual dataset')
        self.vdslist = self.create_virtual_dssc_datasets(self.run, path=self.tempdir)
        
        # create a dummy scan variable for dark run
        # for other type or run, use DSSC.define_run function to overwrite it
        self.scan = xr.DataArray(np.ones_like(self.run.train_ids), dims=['trainId'],
                                 coords={'trainId': self.run.train_ids})
        self.scan_vname = 'dummy'
        self.vds_scan = None
        
    def define_scan(self, vname, bins):
        """
            Prepare the binning of the DSSC data.
            
            inputs:
                vname: variable name for the scan, can be a mnemonic string from ToolBox
                    or a dictionnary with ['source', 'key'] fields
                bins: step size (or bins_edge but not yet implemented)
        """

        if type(vname) is dict:
            self.scan = self.run.get_array(vname['source'], vname['key'])
        elif type(vname) is str:
            if vname not in tb.mnemonics:
                raise ValueError(f'{vname} not found in the ToolBox mnemonics table')
            self.scan = self.run.get_array(tb.mnemonics[vname]['source'], tb.mnemonics[vname]['key'])
        else:
            raise ValueError(f'vname should be a string or a dict. We got {type(vname)}')
            
        if (type(bins) is int) or (type(bins) is float):
            self.scan = bins * np.round(self.scan / bins)
        else:
            # TODO: digitize the data
            raise ValueError(f'To be implemented')
        self.scan_vname = vname
        
        self.vds_scan = os.path.join(self.tempdir, 'scan_variable.h5')
        if os.path.isfile(self.vds_scan):
            os.remove(self.vds_scan)
            
        self.scan = self.scan.to_dataset(name='scan_variable')
        self.scan['xgm_pumped'] = self.xgm[:, :self.nbunches:2].mean('dim_0')
        self.scan['xgm_unpumped'] = self.xgm[:, 1:self.nbunches:2].mean('dim_0')
        self.scan.to_netcdf(self.vds_scan, group='data')

        self.scan_counts = xr.DataArray(np.ones(len(self.scan['scan_variable'])),
                                        dims=['scan_variable'],
                                        coords={'scan_variable': self.scan['scan_variable'].values},
                                        name='counts')
        
        fig, (ax1, ax2) = plt.subplots(nrows=2, figsize=[5, 5])
        ax1.plot(self.scan.groupby('scan_variable').mean('trainId').coords['scan_variable'].values,
                 self.scan_counts.groupby('scan_variable').sum(),
                 'o-', ms=2)
        ax1.set_xlabel('scan variable')
        ax1.set_ylabel('# trains')
        ax1.set_title(self.plot_title)
        
        ax2.plot(self.scan['scan_variable'])
        ax2.set_xlabel('train #')
        ax2.set_ylabel(f'{vname}')

        plt.tight_layout()
            
    def load_xgm(self):
        """ Loads pulse resolved dedicated SAS3 data from the SCS XGM.
        
        """
        if self.xgm is None:
            self.xgm = self.run.get_array(tb.mnemonics['SCS_SA3']['source'],
                                          tb.mnemonics['SCS_SA3']['key'], roi=kd.by_index[:self.nbunches])
            
    def plot_xgm_hist(self, nbins=100):
        """ Plots an histogram of the SCS XGM dedicated SAS3 data.
        
            inputs:
                nbins: number of the bins for the histogram.
        """
        if self.xgm is None:
            self.load_xgm()
            
        hist, bins_edges = np.histogram(self.xgm, nbins, density=True)
        width = 1.0 * (bins_edges[1] - bins_edges[0])
        bins_center = 0.5*(bins_edges[:-1] + bins_edges[1:])
        
        plt.figure(figsize=(5,3))
        plt.bar(bins_center, hist, align='center', width=width)
        plt.xlabel(f"{tb.mnemonics['SCS_SA3']['source']}{tb.mnemonics['SCS_SA3']['key']}")
        plt.ylabel('density')
        plt.title(self.plot_title)
        plt.tight_layout()
        
    def xgm_filter(self, xgm_low=-np.inf, xgm_high=np.inf):
        """ Filters the data by train. If one pulse within a train has an SASE3 SCS XGM value below
            xgm_low or above xgm_high, that train will be dropped from the dataset.
            
            inputs:
                xgm_low: low threshold value
                xgm_high: high threshold value
        """
                   
        if self.xgm is None:
            self.load_xgm()
        
        if self.isDark:
            warnings.warn(f'This run was loaded as dark. Filtering on xgm makes no sense. Aborting')
            return
        
        self.xgm_low = xgm_low
        self.xgm_high = xgm_high
        
        valid = ((self.xgm > self.xgm_low) * (self.xgm < self.xgm_high)).prod('dim_0').astype(bool)
        xgm_valid = self.xgm.where(valid)
        xgm_valid = xgm_valid.dropna('trainId')
        self.scan = self.scan.sel({'trainId': xgm_valid.trainId})
        nrejected = len(self.run.train_ids) - len(self.scan)
        print((f'Rejecting {nrejected} out of {len(self.run.train_ids)} trains due to xgm '
               f'thresholds: [{self.xgm_low}, {self.xgm_high}]'))

    def load_geom(self):
        """ Loads and return the DSSC geometry.
        """
        quad_pos = [
            (-124.100,    3.112),  # TR
            (-133.068, -110.604),  # BR
            (   0.988, -125.236),  # BL
            (   4.528,   -4.912)   # TL
            ]
        path = '/gpfs/exfel/sw/software/exfel_environments/misc/git/karabo_data/docs/dssc_geo_june19.h5'
        geom = DSSC_1MGeometry.from_h5_file_and_quad_positions(path, quad_pos)
        return geom

    def create_virtual_dssc_datasets(self, run, path=''):
        """ Create virtual datasets for each 16 DSSC modules used for the multiprocessing.
            
            input:
                run: karabo-data run
                path: string where the virtual files are created
        """
        
        vds_list = []
        for m in tqdm(range(16)):
            vds_filename = os.path.join(path, f'dssc{m}_vds.h5')
            if os.path.isfile(vds_filename):
                os.remove(vds_filename)
            module_vds = run.get_virtual_dataset(f'SCS_DET_DSSC1M-1/DET/{m}CH0:xtdf',
                                                 'image.data', filename=vds_filename)
            vds_list.append([vds_filename, module_vds])
        return vds_list
    
    def crunch(self):
        """ Crunch through the DSSC data using multiprocessing.
        
        """
        if self.vds_scan is None:
            # probably a dark run with a dummy scan variable
            self.vds_scan = os.path.join(self.tempdir, 'scan_variable.h5')
            if os.path.isfile(self.vds_scan):
                os.remove(self.vds_scan)
            
            self.scan = self.scan.to_dataset(name='scan_variable')
            self.scan.to_netcdf(self.vds_scan, group='data')

        max_GB = 400
        # max_GB / (8byte * 16modules * 128px * 512px * N_pulses)
        self.chunksize = int(max_GB * 1024**3 // (8 * 16 * 128 * 512 * self.fpt))
        
        print('processing', self.chunksize, 'trains per chunk')
                   
        jobs = []
        for m in range(16):
            jobs.append(dict(
            module=m,
            fpt=self.fpt,
            vdf_module=os.path.join(self.tempdir, f'dssc{m}_vds.h5'),
            chunksize=self.chunksize,
            vdf_scan=self.vds_scan,
            nbunches=self.nbunches,
            run_nr=self.run_nr,
            ))
            
        timestamp = strftime('%X')
        print(f'start time: {timestamp}')

        with multiprocessing.Pool(16) as pool:
            module_data = pool.map(process_one_module, jobs)
    
        print('finished:', strftime('%X'))
    
        # rearange the multiprocessed data
        self.module_data = xr.concat(module_data, dim='module')
        self.module_data['run'] = self.run_nr
        self.module_data = self.module_data.transpose('scan_variable', 'module', 'x', 'y')
                   
        self.module_data = xr.merge([self.module_data, self.scan.groupby('scan_variable').mean('trainId')])
        self.module_data = self.module_data.squeeze()

    def save(self, save_folder=None, overwrite=False):
        """ Save the crunched data.
        
            inputs:
                save_folder: string of the fodler where to save the data.
                overwrite: boolean whether or not to overwrite existing files.
        """
        if save_folder is None:
            save_folder = this.save_folder

        if self.isDark:
            fname = f'run{self.run_nr}_el.h5'  # no scan
        else:
            fname = f'run{self.run_nr}_by-delay_el.h5'  # run with delay scan (change for other scan types!)


        save_path = os.path.join(save_folder, fname)
        file_exists = os.path.isfile(save_path)

        if not file_exists or (file_exists and overwrite):
            if file_exists:
                warnings.warn(f'Overwriting file: {save_path}')
                os.remove(save_path)
            self.module_data.to_netcdf(save_path, group='data')
            os.chmod(save_path, 0o664)
            print('saving: ', save_path)
        else:
            print('file', save_path, 'exists and overwrite is False')

# since 'self' is not pickable, this function has to be outside the DSSC class so that it can be used
# by the multiprocessing pool.map function
def process_one_module(job):
    module = job['module']
    fpt = job['fpt']
    data_vdf = job['vdf_module']
    scan_vdf = job['vdf_scan']
    chunksize = job['chunksize']
    nbunches = job['nbunches']

    image_path = f'INSTRUMENT/SCS_DET_DSSC1M-1/DET/{module}CH0:xtdf/image/data'
    npulse_path = f'INDEX/SCS_DET_DSSC1M-1/DET/{module}CH0:xtdf/image/count'
    with h5py.File(data_vdf, 'r') as m:
        all_trainIds = m['INDEX/trainId'][()]
    n_trains = len(all_trainIds)
    chunk_start = np.arange(n_trains, step=chunksize, dtype=int)

    # load scan variable
    scan = xr.open_dataset(scan_vdf, group='data')['scan_variable']
    scan.name = 'scan'
    len_scan = len(scan.groupby(scan))

    # create empty dataset to add actual data to
    module_data = xr.DataArray(np.empty([len_scan, 128, 512], dtype=np.float64),
                               dims=['scan_variable', 'x', 'y'],
                               coords={'scan_variable': np.unique(scan)})
    module_data = module_data.to_dataset(name='pumped')
    module_data['unpumped'] = xr.full_like(module_data['pumped'], 0)
    module_data['sum_count'] = xr.DataArray(np.zeros_like(np.unique(scan)), dims=['scan_variable'])
    module_data['module'] = module

    # crunching
    with h5py.File(data_vdf, 'r') as m:
        #fpt_calc = int(len(m[image_path]) / n_trains)
        #assert fpt_calc == fpt, f'data length does not match expected value (module {module})'
        all_trainIds = m['INDEX/trainId'][()]
        frames_per_train = m[npulse_path][()]
        trains_with_data = all_trainIds[frames_per_train == fpt]
        #print(np.unique(pulses_per_train), '/', fpt)
        #print(len(trains_with_data))
        chunk_start = np.arange(len(all_trainIds), step=chunksize, dtype=int)
        trains_start = 0
                   
        # This line is the strange hack from https://github.com/tqdm/tqdm/issues/485
        print(' ', end='', flush=True)
        for c0 in tqdm(chunk_start, desc=f'pool.map#{module:02d}', position=module):
            chunk_dssc = np.s_[int(c0 * fpt):int((c0 + chunksize) * fpt)]  # for dssc data
            data = m[image_path][chunk_dssc].squeeze()
            data = data.astype(np.float64)
            n_trains = int(data.shape[0] // fpt)
            trainIds_chunk = np.unique(trains_with_data[trains_start:trains_start + n_trains])
            trains_start += n_trains
            n_trains_actual = len(trainIds_chunk)
            coords = {'trainId': trainIds_chunk}
            data = np.reshape(data, [n_trains_actual, fpt, 128, 512])[:, :int(2 * nbunches)]
            data = xr.DataArray(data, dims=['trainId', 'pulse', 'x', 'y'], coords=coords)
            data_pumped = (data[:, ::4]).mean('pulse')
            data_unpumped = (data[:, 2::4]).mean('pulse')
            data = data_pumped.to_dataset(name='pumped')
            data['unpumped'] = data_unpumped
            data['sum_count'] = xr.DataArray(np.ones(n_trains_actual), dims=['trainId'], coords=coords)
            # grouping and summing
            data['scan_variable'] = scan  # this only adds scan data for matching trainIds
            data = data.dropna('trainId')
            data = data.groupby('scan_variable').sum('trainId')
            where = {'scan_variable': data.scan_variable}
            for var in ['pumped', 'unpumped', 'sum_count']:
                module_data[var].loc[where] = module_data[var].loc[where] + data[var]
    for var in ['pumped', 'unpumped']:
        module_data[var] = module_data[var] / module_data.sum_count
    #module_data = module_data.drop('sum_count')
    return module_data