def raw_to_dsp(ds, overwrite=False, nevt=None, test=False, verbose=2, block=8, group=''):
"""
Run raw_to_dsp on a set of runs.
[raw file] ---> [dsp_run{}.lh5] (digital signal processing results)
"""
for run in ds.runs:
raw_file = "/lfs/l1/legend/users/dandrea/pygama/pgt/tier1/pgt_longtrace_run0117-20200110-105115-calib_raw.lh5"
dsp_file = "/lfs/l1/legend/users/dandrea/pygama/pgt/tier2/pgt_longtrace_run0117-20200110-105115-calib_dsp.lh5"
#raw_file = ds.paths[run]["raw_path"]
#dsp_file = ds.paths[run]["dsp_path"]
print("raw_file: ",raw_file)
print("dsp_file: ",dsp_file)
if dsp_file is not None and overwrite is False:
continue
if dsp_file is None:
# declare new file name
dsp_file = raw_file.replace('raw_', 'dsp_')
if test:
print("test mode (dry run), processing raw file:", raw_file)
continue
print("Definition of new LH5 version")
#f_lh5 = lh5.Store()
#data = f_lh5.read_object("raw", raw_file)
#wf_in = data['waveform']['values'].nda
#dt = data['waveform']['dt'].nda[0] * unit_parser.parse_unit(data['waveform']['dt'].attrs['units'])
lh5_in = lh5.Store()
#groups = lh5_in.ls(raw_file, group)
f = h5py.File(raw_file,'r')
print("File info: ",f.keys())
for group in f.keys():
print("Processing: " + raw_file + '/' + group)
#data = lh5_in.read_object(group, raw_file)
data = f[group]['raw']
#wf_in = data['waveform']['values'].nda
#dt = data['waveform']['dt'].nda[0] * unit_parser.parse_unit(data['waveform']['dt'].attrs['units'])
wf_in = data['waveform']['values'][()]
dt = data['waveform']['dt'][0] * unit_parser.parse_unit(data['waveform']['dt'].attrs['units'])
# Parameters for DCR calculation
dcr_trap_int = 200
dcr_trap_flat = 1000
dcr_trap_startSample = 1200
# Set up processing chain
proc = ProcessingChain(block_width=block, clock_unit=dt, verbosity=verbose)
proc.add_input_buffer("wf", wf_in, dtype='float32')
# Basic Filters
proc.add_processor(mean_stdev, "wf[0:1000]", "bl", "bl_sig")
proc.add_processor(np.subtract, "wf", "bl", "wf_blsub")
proc.add_processor(pole_zero, "wf_blsub", 145*us, "wf_pz")
proc.add_processor(trap_norm, "wf_pz", 10*us, 5*us, "wf_trap")
proc.add_processor(asymTrapFilter, "wf_pz", 0.05*us, 2*us, 4*us, "wf_atrap")
# Timepoint calculation
proc.add_processor(np.argmax, "wf_blsub", 1, "t_max", signature='(n),()->()', types=['fi->i'])
proc.add_processor(time_point_frac, "wf_blsub", 0.95, "t_max", "tp_95")
proc.add_processor(time_point_frac, "wf_blsub", 0.8, "t_max", "tp_80")
proc.add_processor(time_point_frac, "wf_blsub", 0.5, "t_max", "tp_50")
proc.add_processor(time_point_frac, "wf_blsub", 0.2, "t_max", "tp_20")
proc.add_processor(time_point_frac, "wf_blsub", 0.05, "t_max", "tp_05")
proc.add_processor(time_point_thresh, "wf_atrap[0:2000]", 0, "tp_0")
# Energy calculation
proc.add_processor(np.amax, "wf_trap", 1, "trapEmax", signature='(n),()->()', types=['fi->f'])
proc.add_processor(fixed_time_pickoff, "wf_trap", "tp_0+(5*us+9*us)", "trapEftp")
proc.add_processor(trap_pickoff, "wf_pz", 1.5*us, 0, "tp_0", "ct_corr")
# Current calculation
proc.add_processor(avg_current, "wf_pz", 10, "curr(len(wf_pz)-10, f)")
proc.add_processor(np.amax, "curr", 1, "curr_amp", signature='(n),()->()', types=['fi->f'])
proc.add_processor(np.divide, "curr_amp", "trapEftp", "aoe")
# DCR calculation: use slope using 1000 samples apart and averaging 200
# samples, with the start 1.5 us offset from t0
proc.add_processor(trap_pickoff, "wf_pz", 200, 1000, "tp_0+1.5*us", "dcr_unnorm")
proc.add_processor(np.divide, "dcr_unnorm", "trapEftp", "dcr")
# Tail slope. Basically the same as DCR, except with no PZ correction
proc.add_processor(linear_fit, "wf_blsub[3000:]", "wf_b", "wf_m")
proc.add_processor(np.divide, "-wf_b", "wf_m", "tail_rc")
#add zac filter energy calculation
sigma = 10*us
flat = 1*us
decay = 160*us
proc.add_processor(zac_filter, "wf", sigma, flat, decay, "wf_zac(101, f)")
proc.add_processor(np.amax, "wf_zac", 1, "zacE", signature='(n),()->()', types=['fi->f'])
# Set up the LH5 output
lh5_out = lh5.Table(size=proc._buffer_len)
lh5_out.add_field("zacE", lh5.Array(proc.get_output_buffer("zacE"), attrs={"units":"ADC"}))
lh5_out.add_field("trapEmax", lh5.Array(proc.get_output_buffer("trapEmax"), attrs={"units":"ADC"}))
lh5_out.add_field("trapEftp", lh5.Array(proc.get_output_buffer("trapEftp"), attrs={"units":"ADC"}))
lh5_out.add_field("ct_corr", lh5.Array(proc.get_output_buffer("ct_corr"), attrs={"units":"ADC*ns"}))
lh5_out.add_field("bl", lh5.Array(proc.get_output_buffer("bl"), attrs={"units":"ADC"}))
lh5_out.add_field("bl_sig", lh5.Array(proc.get_output_buffer("bl_sig"), attrs={"units":"ADC"}))
lh5_out.add_field("A", lh5.Array(proc.get_output_buffer("curr_amp"), attrs={"units":"ADC"}))
lh5_out.add_field("AoE", lh5.Array(proc.get_output_buffer("aoe"), attrs={"units":"ADC"}))
lh5_out.add_field("dcr", lh5.Array(proc.get_output_buffer("dcr"), attrs={"units":"ADC"}))
lh5_out.add_field("tp_max", lh5.Array(proc.get_output_buffer("tp_95", unit=us), attrs={"units":"us"}))
lh5_out.add_field("tp_95", lh5.Array(proc.get_output_buffer("tp_95", unit=us), attrs={"units":"us"}))
lh5_out.add_field("tp_80", lh5.Array(proc.get_output_buffer("tp_80", unit=us), attrs={"units":"us"}))
lh5_out.add_field("tp_50", lh5.Array(proc.get_output_buffer("tp_50", unit=us), attrs={"units":"us"}))
lh5_out.add_field("tp_20", lh5.Array(proc.get_output_buffer("tp_20", unit=us), attrs={"units":"us"}))
lh5_out.add_field("tp_05", lh5.Array(proc.get_output_buffer("tp_05", unit=us), attrs={"units":"us"}))
lh5_out.add_field("tp_0", lh5.Array(proc.get_output_buffer("tp_0", unit=us), attrs={"units":"us"}))
lh5_out.add_field("tail_rc", lh5.Array(proc.get_output_buffer("tail_rc", unit=us), attrs={"units":"us"}))
print("Processing:\n",proc)
proc.execute()
#groupname = group[:group.rfind('/')+1]+"data"
groupname = group+"/data"
print("Writing to: " + dsp_file + "/" + groupname)
lh5_in.write_object(lh5_out, groupname, dsp_file)
def optimize_trap(dg):
"""
Generate a file with grid points to search, and events from the target peak.
Then run DSP a bunch of times on the small table, and fit the peak w/ the
peakshape function.
NOTE: run table-to-table DSP (no file I/O)
"""
f_peak = './temp_peak.lh5' # lh5
f_results = './temp_results.h5' # pandas
grp_data, grp_grid = '/optimize_data', '/optimize_grid'
# epar, elo, ehi, epb = 'energy', 0, 1e7, 10000 # full range
epar, elo, ehi, epb = 'energy', 3.88e6, 3.92e6, 500 # K40 peak
show_movie = True
write_output = True
n_rows = None # default None
with open('opt_trap.json') as f:
dsp_config = json.load(f, object_pairs_hook=OrderedDict)
# files to consider. fixme: right now only works with one file
sto = lh5.Store()
lh5_dir = os.path.expandvars(dg.config['lh5_dir'])
raw_list = lh5_dir + dg.fileDB['raw_path'] + '/' + dg.fileDB['raw_file']
f_raw = raw_list.values[0]
tb_raw = 'ORSIS3302DecoderForEnergy/raw/'
# quick check of the energy range
# ene_raw = sto.read_object(tb_raw+'/'+epar, f_raw).nda
# hist, bins, var = pgh.get_hist(ene_raw, range=(elo, ehi), dx=epb)
# plt.plot(bins[1:], hist, ds='steps')
# plt.show()
# exit()
# set grid parameters
# TODO: jason's suggestions, knowing the expected shape of the noise curve
# e_rises = np.linspace(-1, 0, sqrt(sqrt(3))
# e_rises # make another list which is 10^pwr of this list
# np.linspace(log_tau_min, log_tau_max) # try this too
e_rises = np.arange(1, 12, 1)
e_flats = np.arange(1, 6, 1)
# rc_consts = np.arange(54, 154, 10) # changing this here messes up DCR
# -- create the grid search file the first time --
# NOTE: this makes a linear grid, and is editable by the arrays above.
# jason also proposed a more active gradient-descent style search
# like with Brent's method. (https://en.wikipedia.org/wiki/Brent%27s_method)
if True:
# if not os.path.exists(f_peak):
print('Recreating grid search file')
# create the grid file
# NOTE: save it as an lh5 Table just as an example of writing/reading one
lists = [e_rises, e_flats] #, rc_consts]
prod = list(itertools.product(*lists)) # clint <3 stackoverflow
df_grid = pd.DataFrame(prod, columns=['rise', 'flat']) #,'rc'])
lh5_grid = {}
for i, dfcol in df_grid.iteritems():
lh5_grid[dfcol.name] = lh5.Array(dfcol.values)
tb_grid = lh5.Table(col_dict=lh5_grid)
sto.write_object(tb_grid, grp_grid, f_peak)
# filter events by onboard energy
ene_raw = sto.read_object(tb_raw + '/' + epar, f_raw).nda
# hist, bins, var = pgh.get_hist(ene_raw, range=(elo, ehi), dx=epb)
# plt.plot(bins[1:], hist, ds='steps')
# plt.show()
if n_rows is not None:
ene_raw = ene_raw[:n_rows]
idx = np.where((ene_raw > elo) & (ene_raw < ehi))
# create a filtered table with correct waveform and attrs
# TODO: move this into a function in lh5.py which takes idx as an input
tb_data, wf_tb_data = lh5.Table(), lh5.Table()
# read non-wf cols (lh5 Arrays)
data_raw = sto.read_object(tb_raw, f_raw, n_rows=n_rows)
for col in data_raw.keys():
if col == 'waveform': continue
newcol = lh5.Array(data_raw[col].nda[idx],
attrs=data_raw[col].attrs)
tb_data.add_field(col, newcol)
# handle waveform column (lh5 Table)
data_wfs = sto.read_object(tb_raw + '/waveform', f_raw, n_rows=n_rows)
for col in data_wfs.keys():
attrs = data_wfs[col].attrs
if isinstance(data_wfs[col], lh5.ArrayOfEqualSizedArrays):
# idk why i can't put the filtered array into the constructor
aoesa = lh5.ArrayOfEqualSizedArrays(attrs=attrs, dims=[1, 1])
aoesa.nda = data_wfs[col].nda[idx]
newcol = aoesa
else:
newcol = lh5.Array(data_wfs[col].nda[idx], attrs=attrs)
wf_tb_data.add_field(col, newcol)
tb_data.add_field('waveform', wf_tb_data)
tb_data.attrs = data_raw.attrs
sto.write_object(tb_data, grp_data, f_peak)
else:
print('Loading peak file. groups:', sto.ls(f_peak))
tb_grid = sto.read_object(grp_grid, f_peak)
tb_data = sto.read_object(grp_data, f_peak) # filtered file
# tb_data = sto.read_object(tb_raw, f_raw) # orig file
df_grid = tb_grid.get_dataframe()
# check shape of input table
print('input table attributes:')
for key in tb_data.keys():
obj = tb_data[key]
if isinstance(obj, lh5.Table):
for key2 in obj.keys():
obj2 = obj[key2]
print(' ', key, key2, obj2.nda.shape, obj2.attrs)
else:
print(' ', key, obj.nda.shape, obj.attrs)
# clear new colums if they exist
new_cols = ['e_fit', 'fwhm_fit', 'rchisq', 'xF_err', 'fwhm_ovr_mean']
for col in new_cols:
if col in df_grid.columns:
df_grid.drop(col, axis=1, inplace=True)
t_start = time.time()
def run_dsp(dfrow):
"""
run dsp on the test file, editing the processor list
alternate idea: generate a long list of processors with different names
"""
# adjust dsp config dictionary
rise, flat = dfrow
# dsp_config['processors']['wf_pz']['defaults']['db.pz.tau'] = f'{tau}*us'
dsp_config['processors']['wf_trap']['args'][1] = f'{rise}*us'
dsp_config['processors']['wf_trap']['args'][2] = f'{flat}*us'
# pprint(dsp_config)
# run dsp
pc, tb_out = build_processing_chain(tb_data, dsp_config, verbosity=0)
pc.execute()
# analyze peak
e_peak = 1460.
etype = 'trapEmax'
elo, ehi, epb = 4000, 4500, 3 # the peak moves around a bunch
energy = tb_out[etype].nda
# get histogram
hE, bins, vE = pgh.get_hist(energy, range=(elo, ehi), dx=epb)
xE = bins[1:]
# should I center the max at 1460?
# simple numerical width
i_max = np.argmax(hE)
h_max = hE[i_max]
upr_half = xE[(xE > xE[i_max]) & (hE <= h_max / 2)][0]
bot_half = xE[(xE < xE[i_max]) & (hE >= h_max / 2)][0]
fwhm = upr_half - bot_half
sig = fwhm / 2.355
# fit to gaussian: amp, mu, sig, bkg
fit_func = pgf.gauss_bkg
amp = h_max * fwhm
bg0 = np.mean(hE[:20])
x0 = [amp, xE[i_max], sig, bg0]
xF, xF_cov = pgf.fit_hist(fit_func, hE, bins, var=vE, guess=x0)
# collect results
e_fit = xF[0]
xF_err = np.sqrt(np.diag(xF_cov))
e_err = xF
fwhm_fit = xF[1] * 2.355 * 1460. / e_fit
fwhm_err = xF_err[2] * 2.355 * 1460. / e_fit
chisq = []
for i, h in enumerate(hE):
model = fit_func(xE[i], *xF)
diff = (model - h)**2 / model
chisq.append(abs(diff))
rchisq = sum(np.array(chisq) / len(hE))
fwhm_ovr_mean = fwhm_fit / e_fit
if show_movie:
plt.plot(xE,
hE,
ds='steps',
c='b',
lw=2,
label=f'{etype} {rise}--{flat}')
# peak shape
plt.plot(xE,
fit_func(xE, *x0),
'-',
c='orange',
alpha=0.5,
label='init. guess')
plt.plot(xE,
fit_func(xE, *xF),
'-r',
alpha=0.8,
label='peakshape fit')
plt.plot(np.nan,
np.nan,
'-w',
label=f'mu={e_fit:.1f}, fwhm={fwhm_fit:.2f}')
plt.xlabel(etype, ha='right', x=1)
plt.ylabel('Counts', ha='right', y=1)
plt.legend(loc=2)
# show a little movie
plt.show(block=False)
plt.pause(0.01)
plt.cla()
# return results
return pd.Series({
'e_fit': e_fit,
'fwhm_fit': fwhm_fit,
'rchisq': rchisq,
'fwhm_err': xF_err[0],
'fwhm_ovr_mean': fwhm_ovr_mean
})
# df_grid=df_grid[:10]
df_tmp = df_grid.progress_apply(run_dsp, axis=1)
df_grid[new_cols] = df_tmp
# print(df_grid)
if show_movie:
plt.close()
print('elapsed:', time.time() - t_start)
if write_output:
df_grid.to_hdf(f_results, key=grp_grid)
print(f"Wrote output file: {f_results}")
signature='(n),()->()',
types=['fi->f'])
proc.add_processor(np.divide, "curr_amp", "trapEftp", "aoe")
# DCR calculation: use slope using 1000 samples apart and averaging 200
# samples, with the start 1.5 us offset from t0
proc.add_processor(trap_pickoff, "wf_pz", 200, 1000, "tp_0+1.5*us",
"dcr_unnorm")
proc.add_processor(np.divide, "dcr_unnorm", "trapEftp", "dcr")
# Tail slope. Basically the same as DCR, except with no PZ correction
proc.add_processor(linear_fit, "wf_blsub[3000:]", "wf_b", "wf_m")
proc.add_processor(np.divide, "-wf_b", "wf_m", "tail_rc")
# Set up the LH5 output
lh5_out = lh5.Table(size=proc._buffer_len)
lh5_out.add_field(
"trapEmax",
lh5.Array(proc.get_output_buffer("trapEmax"), attrs={"units": "ADC"}))
lh5_out.add_field(
"trapEftp",
lh5.Array(proc.get_output_buffer("trapEftp"), attrs={"units": "ADC"}))
lh5_out.add_field(
"ct_corr",
lh5.Array(proc.get_output_buffer("ct_corr"), attrs={"units":
"ADC*ns"}))
lh5_out.add_field(
"bl", lh5.Array(proc.get_output_buffer("bl"), attrs={"units": "ADC"}))
lh5_out.add_field(
"bl_sig",
lh5.Array(proc.get_output_buffer("bl_sig"), attrs={"units": "ADC"}))
def process_ds(f_grid, f_opt, f_tier1, d_out, efilter):
"""
process the windowed raw file 'f_tier1' and create the DSP file 'f_opt'
"""
print("Grid file:",f_grid)
df_grid = pd.read_hdf(f_grid)
if os.path.exists(f_opt):
os.remove(f_opt)
if 'corr' in efilter:
bfilter = efilter.split('corr')[0]
try:
df_res = pd.read_hdf(f'{d_out}/{bfilter}_results.h5',key='results')
print("Extraction of best parameters for", bfilter)
except:
print(bfilter,"not optimized")
return
# open raw file
lh5_in = lh5.Store()
#groups = lh5_in.ls(f_tier1, '*/raw')
f = h5py.File(f_tier1,'r')
#print("File info: ",f.keys())
t_start = time.time()
#for group in groups:
for idx, ged in enumerate(f.keys()):
if idx == 4:
diff = time.time() - t_start
tot = diff/5 * len(df_grid) / 60
tot -= diff / 60
print(f"Estimated remaining time: {tot:.2f} mins")
print("Detector:",ged)
#data = lh5_in.read_object(group, f_tier1)
data = f[ged]['raw']
#wf_in = data['waveform']['values'].nda
#dt = data['waveform']['dt'].nda[0] * unit_parser.parse_unit(data['waveform']['dt'].attrs['units'])
wf_in = data['waveform']['values'][()]
dt = data['waveform']['dt'][0] * unit_parser.parse_unit(data['waveform']['dt'].attrs['units'])
bl_in = data['baseline'][()] #flashcam baseline values
# Set up DSP processing chain -- very minimal
block = 8 #waveforms to process simultaneously
proc = ProcessingChain(block_width=block, clock_unit=dt, verbosity=False)
proc.add_input_buffer("wf", wf_in, dtype='float32')
proc.add_input_buffer("bl", bl_in, dtype='float32')
wsize = wf_in.shape[1]
dt0 = data['waveform']['dt'][0]*0.001
#proc.add_processor(mean_stdev, "wf[0:1000]", "bl", "bl_sig")
proc.add_processor(np.subtract, "wf", "bl", "wf_blsub")
for i, row in df_grid.iterrows():
if 'corr' in efilter: ct_const = row
if 'trapE' in efilter:
if 'corr' in efilter: rise, flat, rc = float(df_res['rise'][idx]), float(df_res['flat'][idx]), float(df_res['rc'][idx])
else: rise, flat, rc = row
proc.add_processor(pole_zero, "wf_blsub", rc*us, "wf_pz")
proc.add_processor(trap_norm, "wf_pz", rise*us, flat*us, f"wf_trap_{i}")
proc.add_processor(asymTrapFilter, "wf_pz", 0.05*us, 4*us, 4*us, "wf_atrap")
proc.add_processor(time_point_thresh, "wf_pz", 0, "tp_0")
proc.add_processor(np.amax, f"wf_trap_{i}", 1, f"trapE_{i}", signature='(n),()->()', types=['fi->f'])
proc.add_processor(fixed_time_pickoff, f"wf_trap_{i}", f"tp_0+({rise*us}+{flat*us})", f"trapEftp_{i}")
if 'zacE' in efilter:
if 'corr' in efilter: sigma, flat, decay = float(df_res['sigma'][idx]), float(df_res['flat'][idx]), float(df_res['decay'][idx])
else: sigma, flat, decay = row
proc.add_processor(zac_filter(wsize, sigma/dt0, flat/dt0, decay/dt0),"wf", f"wf_zac_{i}(101, f)")
proc.add_processor(np.amax, f"wf_zac_{i}", 1, f"zacE_{i}", signature='(n),()->()', types=['fi->f'])
if 'cuspE' in efilter:
if 'corr' in efilter: sigma, flat, decay = float(df_res['sigma'][idx]), float(df_res['flat'][idx]), float(df_res['decay'][idx])
else: sigma, flat, decay = row
proc.add_processor(cusp_filter(wsize, sigma/dt0, flat/dt0, decay/dt0),"wf_blsub", f"wf_cusp_{i}(101, f)")
proc.add_processor(np.amax, f"wf_cusp_{i}", 1, f"cuspE_{i}", signature='(n),()->()', types=['fi->f'])
if 'corr' in efilter:
proc.add_processor(trap_pickoff, "wf_pz", 1.5*us, 0, "tp_0", "ct_corr")
#proc.add_processor(trap_pickoff, "wf_pz", rise*us, flat*us, "tp_0", "ct_corr")
proc.add_processor(np.multiply, ct_const, "ct_corr", f"ct_corr_cal_{i}")
proc.add_processor(np.add, f"ct_corr_cal_{i}", f"{bfilter}_{i}", f"{efilter}_{i}")
# Set up the LH5 output
lh5_out = lh5.Table(size=proc._buffer_len)
for i, row in df_grid.iterrows():
lh5_out.add_field(f"{efilter}_{i}", lh5.Array(proc.get_output_buffer(f"{efilter}_{i}"), attrs={"units":"ADC"}))
print("Processing:\n",proc)
proc.execute()
#groupname = group[:group.rfind('/')+1]+"data"
#groupname = df_key+"/"+group+"/data"
groupname = ged+"/data"
print("Writing to: " + f_opt + "/" + groupname)
lh5_in.write_object(lh5_out, groupname, f_opt)
print("")
#list the datasets of the output file
data_opt = lh5_in.ls(f_opt)
#data_opt_0 = lh5_in.ls(f_opt,'opt_0/*')
data_opt_0 = lh5_in.ls(f_opt,'g024/data/*')
diff = time.time() - t_start
print(f"Time to process: {diff:.2f} s")
def dsp_to_hit_cage(f_dsp, f_hit, dg, n_max=None, verbose=False, t_start=None):
"""
non-general placeholder for creating a pygama 'hit' file. uses pandas.
for every file, apply:
- energy calibration (peakfit results)
- timestamp correction
for a more general dsp_to_hit, maybe each function could be given in terms
of an 'apply' on a dsp dataframe ...
TODO: create entry config['rawe'] with list of energy pars to calibrate, as
in energy_cal.py
"""
rawe = ['trapEmax']
# create initial 'hit' DataFrame from dsp data
hit_store = lh5.Store()
data = hit_store.read_object(dg.config['input_table'], f_dsp)
df_hit = data.get_dataframe()
# 1. get energy calibration for this run from peakfit
cal_db = db.TinyDB(storage=MemoryStorage)
with open(dg.config['ecaldb']) as f:
raw_db = json.load(f)
cal_db.storage.write(raw_db)
runs = dg.file_keys.run.unique()
if len(runs) > 1:
print("sorry, I can't do combined runs yet")
exit()
run = runs[0]
for etype in rawe:
tb = cal_db.table(f'peakfit_{etype}').all()
df_cal = pd.DataFrame(tb)
df_cal['run'] = df_cal['run'].astype(int)
df_run = df_cal.loc[df_cal.run == run]
cal_pars = df_run.iloc[0][['cal0', 'cal1', 'cal2']]
pol = np.poly1d(cal_pars) # handy numpy polynomial object
df_hit[f'{etype}_cal'] = pol(df_hit[f'{etype}'])
# 2. compute timestamp rollover correction (specific to struck 3302)
clock = 100e6 # 100 MHz
UINT_MAX = 4294967295 # (0xffffffff)
t_max = UINT_MAX / clock
ts = df_hit['timestamp'].values / clock
tdiff = np.diff(ts)
tdiff = np.insert(tdiff, 0, 0)
iwrap = np.where(tdiff < 0)
iloop = np.append(iwrap[0], len(ts))
ts_new, t_roll = [], 0
for i, idx in enumerate(iloop):
ilo = 0 if i == 0 else iwrap[0][i - 1]
ihi = idx
ts_block = ts[ilo:ihi]
t_last = ts[ilo - 1]
t_diff = t_max - t_last
ts_new.append(ts_block + t_roll)
t_roll += t_last + t_diff
df_hit['ts_sec'] = np.concatenate(ts_new)
# 3. compute global timestamp
if t_start is not None:
df_hit['ts_glo'] = df_hit['ts_sec'] + t_start
# write to LH5 file
if os.path.exists(f_hit):
os.remove(f_hit)
sto = lh5.Store()
tb_name = dg.config['input_table'].replace('dsp', 'hit')
tb_lh5 = lh5.Table(size=len(df_hit))
for col in df_hit.columns:
tb_lh5.add_field(col, lh5.Array(df_hit[col].values,
attrs={'units': ''}))
print(col)
print(f'Writing table: {tb_name} in file:\n {f_hit}')
sto.write_object(tb_lh5, tb_name, f_hit)
def raw_to_dsp(ds,
overwrite=False,
nevt=None,
test=False,
verbose=2,
block=8,
group='daqdata'):
"""
Run raw_to_dsp on a set of runs.
[raw file] ---> [dsp_run{}.lh5] (digital signal processing results)
"""
for run in ds.runs:
raw_file = ds.paths[run]["raw_path"]
dsp_file = ds.paths[run]["dsp_path"]
if dsp_file is not None and overwrite is False:
continue
if dsp_file is None:
# declare new file name
dsp_file = raw_file.replace('raw', 'dsp')
if test:
print("test mode (dry run), processing raw file:", raw_file)
continue
# new LH5 version
lh5_in = lh5.Store()
data = lh5_in.read_object("/ORSIS3302DecoderForEnergy", raw_file)
wf_in = data['waveform']['values'].nda
dt = data['waveform']['dt'].nda[0] * unit_parser.parse_unit(
data['waveform']['dt'].attrs['units'])
# Parameters for DCR calculation
dcr_trap_int = 200
dcr_trap_flat = 1000
dcr_trap_startSample = 1200
# Set up processing chain
proc = ProcessingChain(block_width=block,
clock_unit=dt,
verbosity=verbose)
proc.add_input_buffer("wf", wf_in, dtype='float32')
proc.add_processor(mean_stdev, "wf[0:1000]", "bl", "bl_sig")
proc.add_processor(np.subtract, "wf", "bl", "wf_blsub")
proc.add_processor(pole_zero, "wf_blsub", 70 * us, "wf_pz")
proc.add_processor(asymTrapFilter, "wf_pz", 10 * us, 5 * us, 10 * us,
"wf_atrap")
proc.add_processor(np.amax,
"wf_atrap",
1,
"atrapE",
signature='(n),()->()',
types=['fi->f'])
# proc.add_processor(np.divide, "atrapmax", 10*us, "atrapE")
proc.add_processor(trap_norm, "wf_pz", 10 * us, 5 * us, "wf_trap")
proc.add_processor(np.amax,
"wf_trap",
1,
"trapE",
signature='(n),()->()',
types=['fi->f'])
proc.add_processor(avg_current, "wf_pz", 10, "curr")
proc.add_processor(np.amax,
"curr",
1,
"A_10",
signature='(n),()->()',
types=['fi->f'])
proc.add_processor(np.divide, "A_10", "trapE", "AoE")
proc.add_processor(trap_pickoff, "wf_pz", dcr_trap_int, dcr_trap_flat,
dcr_trap_startSample, "dcr")
# Set up the LH5 output
lh5_out = lh5.Table(size=proc._buffer_len)
lh5_out.add_field(
"trapE",
lh5.Array(proc.get_output_buffer("trapE"), attrs={"units": "ADC"}))
lh5_out.add_field(
"bl",
lh5.Array(proc.get_output_buffer("bl"), attrs={"units": "ADC"}))
lh5_out.add_field(
"bl_sig",
lh5.Array(proc.get_output_buffer("bl_sig"), attrs={"units":
"ADC"}))
lh5_out.add_field(
"A",
lh5.Array(proc.get_output_buffer("A_10"), attrs={"units": "ADC"}))
lh5_out.add_field(
"AoE",
lh5.Array(proc.get_output_buffer("AoE"), attrs={"units": "ADC"}))
lh5_out.add_field(
"dcr",
lh5.Array(proc.get_output_buffer("dcr"), attrs={"units": "ADC"}))
print("Processing:\n", proc)
proc.execute()
print("Writing to: ", dsp_file)
f_lh5.write_object(lh5_out, "data", dsp_file)
def build_processing_chain(lh5_in,
dsp_config,
outputs=None,
verbosity=1,
block_width=8):
"""
Produces a ProcessingChain object and an lh5 table for output parameters
from an input lh5 table and a json recipe.
Returns (proc_chain, lh5_out):
- proc_chain: ProcessingChain object that is bound to lh5_in and lh5_out;
all you need to do is handle file i/o for lh5_in/out and run execute
- lh5_out: output LH5 table
Required arguments:
- lh5_in: input LH5 table
- config: dict or name of json file containing a recipe for
constructing the ProcessingChain object produced by this function.
config is formated as a json dict with different processors. Config
should have a dictionary called processors, containing dictionaries
of the following format:
Key: parameter name: name of parameter produced by the processor.
can optionally provide multiple, separated by spaces
Values:
processor (req): name of gufunc
module (req): name of module in which to find processor
prereqs (req): name of parameters from other processors and from
input that are required to exist to run this
args (req): list of arguments for processor, with variables passed
by name or value. Names should either be inputs from lh5_in, or
parameter names for other processors. Names of the format db.name
will look up the parameter in the metadata.
kwargs (opt): kwargs used when adding processors to proc_chain
init_args (opt): args used when initializing a processor that has
static data (for factory functions)
default (opt): default value for db parameters if not found
unit (opt): unit to be used for attr in lh5 file.
There may also be a list called 'outputs', containing a list of parameters
to put into lh5_out.
Optional keyword arguments:
- outputs: list of parameters to put in the output lh5 table. If None,
use the parameters in the 'outputs' list from config
- verbosity: verbosity level:
0: Print nothing (except errors...)
1: Print basic warnings (default)
2: Print basic debug info
3: Print friggin' everything!
- block_width: number of entries to process at once.
"""
if isinstance(dsp_config, str):
with open(dsp_config) as f:
dsp_config = json.load(f)
elif dsp_config is None:
dsp_config = {'outputs': [], 'processors': {}}
else:
# We don't want to modify the input!
dsp_config = deepcopy(dsp_config)
if outputs is None:
outputs = dsp_config['outputs']
processors = dsp_config['processors']
# for processors with multiple outputs, add separate entries to the processor list
for key in list(processors):
keys = [k for k in re.split(",| ", key) if k != '']
if len(keys) > 1:
for k in keys:
processors[k] = key
# Recursive function to crawl through the parameters/processors and get
# a sequence of unique parameters such that parameters always appear after
# their dependencies. For parameters that are not produced by the ProcChain
# (i.e. input/db parameters), add them to the list of leafs
# https://www.electricmonk.nl/docs/dependency_resolving_algorithm/dependency_resolving_algorithm.html
def resolve_dependencies(par, resolved, leafs, unresolved=[]):
if par in resolved:
return
elif par in unresolved:
raise Exception('Circular references detected: %s -> %s' %
(par, edge))
# if we don't find a node, this is a leaf
node = processors.get(par)
if node is None:
if par not in leafs:
leafs.append(par)
return
# if it's a string, that means it is part of a processor that returns multiple outputs (see above); in that case, node is a str pointing to the actual node we want
if isinstance(node, str):
resolve_dependencies(node, resolved, leafs, unresolved)
return
edges = node['prereqs']
unresolved.append(par)
for edge in edges:
resolve_dependencies(edge, resolved, leafs, unresolved)
resolved.append(par)
unresolved.remove(par)
proc_par_list = [] # calculated from processors
input_par_list = [] # input from file and used for processors
copy_par_list = [] # copied from input to output
out_par_list = []
for out_par in outputs:
if out_par not in processors:
copy_par_list.append(out_par)
else:
resolve_dependencies(out_par, proc_par_list, input_par_list)
out_par_list.append(out_par)
if verbosity > 0:
print('Processing parameters:', str(proc_par_list))
print('Required input parameters:', str(input_par_list))
print('Copied output parameters:', str(copy_par_list))
print('Processed output parameters:', str(out_par_list))
proc_chain = ProcessingChain(block_width, lh5_in.size, verbosity=verbosity)
# Now add all of the input buffers from lh5_in (and also the clk time)
for input_par in input_par_list:
buf_in = lh5_in.get(input_par)
if buf_in is None:
print("I don't know what to do with " + input_par +
". Building output without it!")
elif isinstance(buf_in, lh5.Array):
proc_chain.add_input_buffer(input_par, buf_in.nda)
elif isinstance(buf_in, lh5.Table):
# check if this is waveform
if 't0' and 'dt' and 'values' in buf_in:
proc_chain.add_input_buffer(input_par, buf_in['values'].nda,
'float32')
clk = buf_in['dt'].nda[0] * unit_parser.parse_unit(
lh5_in['waveform']['dt'].attrs['units'])
if proc_chain._clk is not None and proc_chain._clk != clk:
print(
"Somehow you managed to set multiple clock frequencies...Using "
+ str(proc_chain._clk))
else:
proc_chain._clk = clk
# now add the processors
for proc_par in proc_par_list:
recipe = processors[proc_par]
module = importlib.import_module(recipe['module'])
func = getattr(module, recipe['function'])
args = recipe['args']
for i, arg in enumerate(args):
if isinstance(arg, str) and arg[0:3] == 'db.':
#TODO: ADD METADATA LOOKUP!
args[i] = recipe['defaults'][arg]
kwargs = recipe.get('kwargs',
{}) # might also need metadata lookup here
# if init_args are defined, parse any strings and then call func
# as a factory/constructor function
try:
init_args = recipe['init_args']
for i, arg in enumerate(init_args):
if isinstance(arg, str):
if arg[0:3] == 'db.':
#TODO: ADD METADATA LOOKUP!
init_args[i] = recipe['defaults'][arg]
else:
# see if string can be parsed by proc_chain
try:
init_args[i] = proc_chain.get_variable(arg)
except:
pass
if (verbosity > 1):
print("Building function", func.__name__, "from init_args",
init_args)
func = func(*init_args)
except:
pass
proc_chain.add_processor(func, *args, **kwargs)
# build the output buffers
lh5_out = lh5.Table(size=proc_chain._buffer_len)
# add inputs that are directly copied
for copy_par in copy_par_list:
buf_in = lh5_in.get(copy_par)
if isinstance(buf_in, lh5.Array):
lh5_out.add_field(copy_par, buf_in)
elif isinstance(buf_in, lh5.Table):
# check if this is waveform
if 't0' and 'dt' and 'values' in buf_in:
lh5_out.add_field(copy_par, buf_in['values'])
clk = buf_in['dt'].nda[0] * unit_parser.parse_unit(
lh5_in['waveform']['dt'].attrs['units'])
if proc_chain._clk is not None and proc_chain._clk != clk:
print(
"Somehow you managed to set multiple clock frequencies...Using "
+ str(proc_chain._clk))
else:
proc_chain._clk = clk
else:
print("I don't know what to do with " + input_par +
". Building output without it!")
# finally, add the output buffers to lh5_out and the proc chain
for out_par in out_par_list:
recipe = processors[out_par]
# special case for proc with multiple outputs
if isinstance(recipe, str):
i = [k for k in re.split(",| ", recipe) if k != ''].index(out_par)
recipe = processors[recipe]
unit = recipe['unit'][i]
else:
unit = recipe['unit']
try:
scale = convert(1, unit_parser.parse_unit(unit), clk)
except InvalidConversion:
scale = None
buf_out = proc_chain.get_output_buffer(out_par, unit=scale)
lh5_out.add_field(out_par, lh5.Array(buf_out, attrs={"units": unit}))
return (proc_chain, lh5_out)
def load_raw_data_example(f_raw):
"""
make a plot of the timestamps in a particular channel.
instead of accessing just the timestamp column, this is an example
of accessing the entire raw file (including waveforms) with LH5.
"""
sto = lh5.Store()
tb_name = 'g024/raw'
n_rows = 100 # np.inf to read all
# method 1: call load_nda to pull out only timestamp column (fast)
# par_data = lh5.load_nda([f_raw], ['timestamp'], tb_name)
# pprint(par_data)
# print(par_data['timestamp'].shape)
# exit()
# method 2: read all data, just to give a longer example of what we can access
# TODO: include an example of slicing/selecting rows with np.where
# read non-wf cols (lh5 Arrays)
data_raw, n_tot = sto.read_object(tb_name, f_raw, n_rows=n_rows)
# declare output table (must specify n_rows for size)
tb_raw = lh5.Table(size=n_tot)
for col in data_raw.keys():
if col in ['waveform', 'tracelist']: continue
# copy all values
newcol = lh5.Array(data_raw[col].nda, attrs=data_raw[col].attrs)
# copy a selection (using np.where)
# newcol = lh5.Array(data_raw[col].nda[idx], attrs=data_raw[col].attrs)
tb_raw.add_field(col, newcol)
df_raw = tb_raw.get_dataframe()
print(df_raw)
# load waveform column (nested LH5 Table)
data_wfs, n_tot = sto.read_object(tb_name + '/waveform',
f_raw,
n_rows=n_rows)
tb_wfs = lh5.Table(size=n_tot)
for col in data_wfs.keys():
attrs = data_wfs[col].attrs
if isinstance(data_wfs[col], lh5.ArrayOfEqualSizedArrays):
# idk why i can't put the filtered array into the constructor
aoesa = lh5.ArrayOfEqualSizedArrays(attrs=attrs, dims=[1, 1])
aoesa.nda = data_wfs[col].nda
# aoesa.nda = data_wfs[col].nda[idx] # with np.where selection
newcol = aoesa
else:
newcol = lh5.Array(data_wfs[col].nda, attrs=attrs)
# newcol = lh5.Array(data_wfs[col].nda[idx], attrs=attrs) # selection
tb_wfs.add_field(col, newcol)
tb_wfs.add_field('waveform', newcol)
tb_wfs.attrs = data_raw.attrs
# can write to file, to read back in for DSP, etc.
# sto.write_object(tb_raw, grp_data, f_peak)
print(tb_wfs)
print(tb_wfs['waveform'].shape)
