def write_notes(notes_df, outfile):
"""
write_notes(notes_df, infile)
Write notes from notes_df to infile
Parameters
----------
infile: string
filename of a wfdb file from the MIMIC3 matched dataset
notes_df: notes
DataFrame containing notes, times, etc
"""
arr, saType = df_to_sarray(notes_df)
with h5py.File(outfile, 'a') as f:
clin = f.require_group('/clinical')
note_ds = f.create_dataset('clinical/notes',
maxshape=(None, ),
data=arr,
dtype=saType,
compression="gzip",
compression_opts=9,
shuffle=True)
def write_labs(labs_df, outfile):
# TODO: convert flag to category and encode in .flag_info
arr, saType = df_to_sarray(labs_df)
# dt = h5py.special_dtype(vlen=str)
# comp_type = np.dtype([('time', dt), ('testid', 'i8'), ('value', dt), ('valuenum', 'f8'), ('flag', dt)])
# define array for writing to dataset
# arr_data = np.empty((0,), dtype=comp_type)
# for idx, row in labs_df.iterrows():
# arr = np.array([(str(row['charttime']), row['itemid'], row['value'], row['valuenum'], row['flag'])],
# dtype = comp_type)
# arr_data = np.append(arr_data, arr)
#create metadata
labs_grouped = labs_df.groupby('itemid')['itemid', 'label', 'category',
'fluid', 'valueuom',
'loinc_code'].first()
labs_grouped = labs_grouped.set_index('itemid')
test_info = labs_grouped.T.to_dict('dict')
with h5py.File(outfile, 'a') as f:
clin = f.require_group('/clinical')
lab_ds = f.create_dataset('clinical/labs',
maxshape=(None, ),
data=arr,
dtype=saType,
compression="gzip",
compression_opts=9,
shuffle=True)
lab_ds.attrs['.test_info'] = json.dumps(test_info)
def make_mapping (filename, mapper=None, overwrite=True):
if mapper is None:
mapper = LoincMapper(external_mapping_table="MIMICIII")
# open the file
f = audata.File.open(filename, readonly=False)
#get the numerics
#for each dataset, get the columns
#convert to dict
#convert to df
#append to dataset
#save as dataset
"""Numerics"""
signals = []
for dset_name in f['numerics'].list()['datasets']:
dset = f['numerics/'+dset_name]
columns = build_col_dict(dset, 'numerics', std_signals, mapper)
df=pd.DataFrame.from_dict(columns, orient='index')
signals.append(df)
"""Waveforms - if present"""
for dset_name in f['waveforms'].list()['datasets']:
dset = f['waveforms/'+dset_name]
columns = build_col_dict(dset, 'waveforms', std_signals, mapper)
df=pd.DataFrame.from_dict(columns, orient='index')
signals.append(df)
"""write to file"""
""" test for existance of mapping table first - option to overwrite """
signals = pd.concat(signals)
sarray, saType = df_to_sarray(signals)
# test for existance
if 'mapping' in f['/'].list()['datasets']:
print('Mapping table exists')
if overwrite:
del f.hdf['mapping']
f.hdf.create_dataset('mapping', data=sarray, dtype=saType)
else:
print('Not overwritten')
else:
f.hdf.create_dataset('mapping', data=sarray, dtype=saType)
f.close()
def write_labs(df, filename, test_metadata=False):
"""
Parameters
----------
df: DataFrame
Datetime index
filename: String
test_metadata
"""
with h5py.File(filename, 'a') as f:
origin = pd.to_datetime(json.loads(f['/'].attrs['.meta'])['time_origin'])
df.index = df.index.tz_localize('UTC').tz_convert('UTC')
df['time']=(df.index - origin).total_seconds()
del df['HADM_ID']
#TODO: If test_metadata is True, strip repeated elemebts (loinc, etc) and store in .test_info
df.columns = df.columns.str.strip().str.lower()
arr, saType = df_to_sarray(df)
clin = f.require_group('/clinical')
lab_ds = clin.create_dataset('labs', maxshape = (None, ), data = arr, dtype=saType,
compression="gzip", compression_opts = 9, shuffle = True)
lab_ds.attrs['.test_info'] = 'none'