276 lines
7.8 KiB
Python
276 lines
7.8 KiB
Python
import os
|
|
import re
|
|
from collections import defaultdict
|
|
|
|
import numpy as np
|
|
|
|
META_VARS_STR = (
|
|
"instrument",
|
|
"title",
|
|
"sample",
|
|
"user",
|
|
"ProposalID",
|
|
"original_filename",
|
|
"date",
|
|
"zebra_mode",
|
|
"proposal",
|
|
"proposal_user",
|
|
"proposal_title",
|
|
"proposal_email",
|
|
"detectorDistance",
|
|
)
|
|
|
|
META_VARS_FLOAT = (
|
|
"omega",
|
|
"mf",
|
|
"2-theta",
|
|
"chi",
|
|
"phi",
|
|
"nu",
|
|
"temp",
|
|
"wavelenght",
|
|
"a",
|
|
"b",
|
|
"c",
|
|
"alpha",
|
|
"beta",
|
|
"gamma",
|
|
"cex1",
|
|
"cex2",
|
|
"mexz",
|
|
"moml",
|
|
"mcvl",
|
|
"momu",
|
|
"mcvu",
|
|
"snv",
|
|
"snh",
|
|
"snvm",
|
|
"snhm",
|
|
"s1vt",
|
|
"s1vb",
|
|
"s1hr",
|
|
"s1hl",
|
|
"s2vt",
|
|
"s2vb",
|
|
"s2hr",
|
|
"s2hl",
|
|
)
|
|
|
|
META_UB_MATRIX = ("ub1j", "ub2j", "ub3j")
|
|
|
|
CCL_FIRST_LINE = (
|
|
("scan_number", int),
|
|
("h_index", float),
|
|
("k_index", float),
|
|
("l_index", float),
|
|
)
|
|
|
|
CCL_ANGLES = {
|
|
"bi": (
|
|
("twotheta_angle", float),
|
|
("omega_angle", float),
|
|
("chi_angle", float),
|
|
("phi_angle", float),
|
|
),
|
|
"nb": (
|
|
("gamma_angle", float),
|
|
("omega_angle", float),
|
|
("nu_angle", float),
|
|
("unkwn_angle", float),
|
|
),
|
|
}
|
|
|
|
CCL_SECOND_LINE = (
|
|
("n_points", int),
|
|
("angle_step", float),
|
|
("monitor", float),
|
|
("temperature", float),
|
|
("mag_field", float),
|
|
("date", str),
|
|
("time", str),
|
|
("scan_type", str),
|
|
)
|
|
|
|
AREA_METHODS = ("fit_area", "int_area")
|
|
|
|
|
|
def load_1D(filepath):
|
|
"""
|
|
Loads *.ccl or *.dat file (Distinguishes them based on last 3 chars in string of filepath
|
|
to add more variables to read, extend the elif list
|
|
the file must include '#data' and number of points in right place to work properly
|
|
|
|
:arg filepath
|
|
:returns det_variables
|
|
- dictionary of all detector/scan variables and dictinionary for every scan.
|
|
Names of these dictionaries are M + scan number. They include HKL indeces, angles,
|
|
monitors, stepsize and array of counts
|
|
"""
|
|
with open(filepath, "r") as infile:
|
|
_, ext = os.path.splitext(filepath)
|
|
det_variables = parse_1D(infile, data_type=ext)
|
|
|
|
return det_variables
|
|
|
|
|
|
def parse_1D(fileobj, data_type):
|
|
# read metadata
|
|
metadata = {}
|
|
for line in fileobj:
|
|
if "=" in line:
|
|
variable, value = line.split("=")
|
|
variable = variable.strip()
|
|
value = value.strip()
|
|
if variable in META_VARS_FLOAT:
|
|
metadata[variable] = float(value)
|
|
elif variable in META_VARS_STR:
|
|
metadata[variable] = value
|
|
elif variable in META_UB_MATRIX:
|
|
metadata[variable] = re.findall(r"[-+]?\d*\.\d+|\d+", value)
|
|
|
|
if "#data" in line:
|
|
# this is the end of metadata and the start of data section
|
|
break
|
|
|
|
# read data
|
|
scan = {}
|
|
if data_type == ".ccl":
|
|
ccl_first_line = (*CCL_FIRST_LINE, *CCL_ANGLES[metadata["zebra_mode"]])
|
|
ccl_second_line = CCL_SECOND_LINE
|
|
|
|
for line in fileobj:
|
|
s = {}
|
|
|
|
# first line
|
|
for param, (param_name, param_type) in zip(line.split(), ccl_first_line):
|
|
s[param_name] = param_type(param)
|
|
|
|
# second line
|
|
next_line = next(fileobj)
|
|
for param, (param_name, param_type) in zip(next_line.split(), ccl_second_line):
|
|
s[param_name] = param_type(param)
|
|
|
|
s["om"] = np.linspace(
|
|
s["omega_angle"] - (s["n_points"] / 2) * s["angle_step"],
|
|
s["omega_angle"] + (s["n_points"] / 2) * s["angle_step"],
|
|
s["n_points"],
|
|
)
|
|
|
|
# subsequent lines with counts
|
|
counts = []
|
|
while len(counts) < s["n_points"]:
|
|
counts.extend(map(int, next(fileobj).split()))
|
|
s["Counts"] = counts
|
|
|
|
scan[s["scan_number"]] = s
|
|
|
|
elif data_type == ".dat":
|
|
# skip the first 2 rows, the third row contans the column names
|
|
next(fileobj)
|
|
next(fileobj)
|
|
col_names = next(fileobj).split()
|
|
s = defaultdict(list)
|
|
|
|
for line in fileobj:
|
|
if "END-OF-DATA" in line:
|
|
# this is the end of data
|
|
break
|
|
|
|
for name, val in zip(col_names, line.split()):
|
|
s[name].append(float(val))
|
|
|
|
try:
|
|
s["h_index"] = float(metadata["title"].split()[-3])
|
|
s["k_index"] = float(metadata["title"].split()[-2])
|
|
s["l_index"] = float(metadata["title"].split()[-1])
|
|
except (ValueError, IndexError):
|
|
print("seems hkl is not in title")
|
|
|
|
s["om"] = np.array(s["om"])
|
|
|
|
s["temperature"] = metadata["temp"]
|
|
try:
|
|
s["mag_field"] = metadata["mf"]
|
|
except KeyError:
|
|
print("Mag_field not present in dat file")
|
|
|
|
s["omega_angle"] = metadata["omega"]
|
|
s["n_points"] = len(s["om"])
|
|
s["monitor"] = s["Monitor1"][0]
|
|
s["twotheta_angle"] = metadata["2-theta"]
|
|
s["chi_angle"] = metadata["chi"]
|
|
s["phi_angle"] = metadata["phi"]
|
|
s["nu_angle"] = metadata["nu"]
|
|
|
|
s["scan_number"] = 1
|
|
scan[s["scan_number"]] = dict(s)
|
|
|
|
else:
|
|
print("Unknown file extention")
|
|
|
|
for s in scan.values():
|
|
if s["h_index"].is_integer() and s["k_index"].is_integer() and s["l_index"].is_integer():
|
|
s["h_index"] = int(s["h_index"])
|
|
s["k_index"] = int(s["k_index"])
|
|
s["l_index"] = int(s["l_index"])
|
|
s["indices"] = "hkl"
|
|
else:
|
|
s["indices"] = "real"
|
|
|
|
metadata["data_type"] = data_type
|
|
|
|
return {"meta": metadata, "scan": scan}
|
|
|
|
|
|
def export_1D(data, path, area_method=AREA_METHODS[0], lorentz=False, hkl_precision=2):
|
|
"""Exports data in the .comm/.incomm format
|
|
|
|
Scans with integer/real hkl values are saved in .comm/.incomm files correspondingly. If no scans
|
|
are present for a particular output format, that file won't be created.
|
|
"""
|
|
zebra_mode = data["meta"]["zebra_mode"]
|
|
file_content = {".comm": [], ".incomm": []}
|
|
|
|
for key, scan in data["scan"].items():
|
|
if "fit" not in scan:
|
|
print("Scan skipped - no fit value for:", key)
|
|
continue
|
|
|
|
scan_str = f"{key:6}"
|
|
|
|
h, k, l = scan["h_index"], scan["k_index"], scan["l_index"]
|
|
if scan["indices"] == "hkl":
|
|
hkl_str = f"{h:6}{k:6}{l:6}"
|
|
else: # scan["indices"] == "real"
|
|
hkl_str = f"{h:8.{hkl_precision}f}{k:8.{hkl_precision}f}{l:8.{hkl_precision}f}"
|
|
|
|
area_n = scan["fit"][area_method].n
|
|
area_s = scan["fit"][area_method].s
|
|
|
|
# apply lorentz correction to area
|
|
if lorentz:
|
|
if zebra_mode == "bi":
|
|
twotheta_angle = np.deg2rad(scan["twotheta_angle"])
|
|
corr_factor = np.sin(twotheta_angle)
|
|
else: # zebra_mode == "nb":
|
|
gamma_angle = np.deg2rad(scan["gamma_angle"])
|
|
nu_angle = np.deg2rad(scan["nu_angle"])
|
|
corr_factor = np.sin(gamma_angle) * np.cos(nu_angle)
|
|
|
|
area_n = np.abs(area_n * corr_factor)
|
|
area_s = np.abs(area_s * corr_factor)
|
|
|
|
area_str = f"{area_n:10.2f}{area_s:10.2f}"
|
|
|
|
ang_str = ""
|
|
for angle, _ in CCL_ANGLES[zebra_mode]:
|
|
ang_str = ang_str + f"{scan[angle]:8}"
|
|
|
|
ref = file_content[".comm"] if scan["indices"] == "hkl" else file_content[".incomm"]
|
|
ref.append(scan_str + hkl_str + area_str + ang_str + "\n")
|
|
|
|
for ext, content in file_content.items():
|
|
if content:
|
|
with open(path + ext, "w") as out_file:
|
|
out_file.writelines(content)
|