def _tensor_from_file(tm, hd5, **kwargs):
visits = get_visits(tm, hd5, **kwargs)
for visit in visits:
path = tm.path_prefix.replace("*", visit)
if path in hd5:
return np.array([visit])
raise KeyError(f"No visit with {tm.name}.")
def _tensor_from_file(tm, hd5, **kwargs):
visits = get_visits(tm, hd5, **kwargs)
# Get flag if it exists in kwargs, otherwise it's None
flag_filtered = kwargs.get("filtered")
if flag_filtered:
max_size = max(
len(
set(hd5[tm.path_prefix.replace("*", v)].keys()).
intersection(set(DEFINED_TMAPS[sig_type]), ), )
for v in visits)
else:
max_size = max(
len(hd5[tm.path_prefix.replace("*", v)].keys())
for v in visits)
shape = (len(visits), max_size)
tensor = np.full(shape, "", object)
for i, visit in enumerate(visits):
path = tm.path_prefix.replace("*", visit)
list_signals = list(hd5[path].keys())
if flag_filtered:
desired_signals = set(DEFINED_TMAPS[sig_type])
common = set(list_signals).intersection(desired_signals)
list_signals = sorted(common)
tensor[i][:len(list_signals)] = list_signals
return tensor
def _tensor_from_file(tm, hd5, **kwargs):
visits = get_visits(tm, hd5, **kwargs)
max_size = max(hd5[tm.path_prefix.replace("*", v)].size
for v in visits)
shape = (len(visits), max_size)
flag_dates = kwargs.get("readable_dates")
if flag_dates:
tensor = np.zeros(shape, dtype=object)
else:
tensor = np.zeros(shape)
for i, visit in enumerate(visits):
path = tm.path_prefix.replace("*", visit)
indices = hd5[path][()]
is_nan_indices = np.isnan(indices)
no_nan_indices = indices[~is_nan_indices].astype(int)
ecg_tm = get_signal_tmap(f"{lead}_time")
data = ecg_tm.tensor_from_file(ecg_tm, hd5,
visit=visit)[0][no_nan_indices]
nan_indices = np.where(is_nan_indices)[0]
nan_indices -= np.array(range(nan_indices.shape[0]))
data = np.append(data, np.array([np.nan] * nan_indices.shape[0]))
data = np.insert(data, nan_indices, np.nan)[:indices.shape[0]]
if flag_dates:
tensor[i][:data.shape[0]] = get_local_timestamps(data)
else:
tensor[i][:data.shape[0]] = data
return tensor
def _tensor_from_file(tm, hd5, **kwargs):
visits = get_visits(tm, hd5, **kwargs)
flag_dates = kwargs.get("readable_dates")
if tm.is_timeseries:
max_size = max(
hd5[f"{tm.path_prefix}/value".replace("*", v)].size for v in visits
)
shape = (len(visits), 2, max_size)
if flag_dates:
tensor = np.zeros(shape, dtype=object)
else:
tensor = np.zeros(shape)
for i, visit in enumerate(visits):
path = tm.path_prefix.replace("*", visit)
data = hd5[f"{path}/value"][()]
tensor[i][0][: data.shape[0]] = data
unix_array = hd5[f"{path}/time"][()]
if flag_dates:
tensor[i][1][: unix_array.shape[0]] = get_local_timestamps(
unix_array,
)
else:
tensor[i][1][: unix_array.shape[0]] = unix_array
elif tm.is_event or tm.is_continuous:
max_size = max(hd5[tm.path_prefix.replace("*", v)].size for v in visits)
shape = (len(visits), max_size)
tensor = np.zeros(shape)
for i, visit in enumerate(visits):
path = tm.path_prefix.replace("*", visit)
data = hd5[path][()]
if "|S" in str(data.dtype):
data = data.astype("<U32")
for element in data:
try:
float(element)
except ValueError:
if element.startswith("<") or element.startswith(">"):
try:
float(element[1:])
data[data == element] = element[1:]
except ValueError:
data[data == element] = np.nan
else:
data[data == element] = np.nan
data = data.astype(float)
tensor[i][: data.shape[0]] = data
else:
raise ValueError(
f"Incorrect interpretation '{tm.interpretation}' for measurement tmap.",
)
return tensor
def _tensor_from_file(tm, hd5, **kwargs):
visits = get_visits(tm, hd5, **kwargs)
shape = (len(visits),) + tm.shape[1:]
tensor = np.full(shape, "", object)
for i, visit in enumerate(visits):
path = tm.path_prefix.replace("*", visit)
tensor[i] = hd5[path].attrs[key]
tensor.astype(int)
return tensor
def _tensor_from_file(tm, hd5, **kwargs):
visits = get_visits(tm, hd5, **kwargs)
flag_dates = kwargs.get("readable_dates")
max_size = max(hd5[tm.path_prefix.replace("*", v)].size for v in visits)
shape = (len(visits), max_size)
if tm.path_prefix.endswith("start_date") and flag_dates:
tensor = np.zeros(shape, dtype=object)
else:
tensor = np.zeros(shape)
for i, visit in enumerate(visits):
path = tm.path_prefix.replace("*", visit)
data = hd5[path][()]
if path.endswith("start_date") and flag_dates:
tensor[i][: data.shape[0]] = get_local_timestamps(data)
else:
tensor[i][: data.shape[0]] = data
return tensor
def _tensor_from_file(tm, hd5, **kwargs):
visits = get_visits(tm, hd5, **kwargs)
flag_dates = kwargs.get("readable_dates")
if tm.is_timeseries:
max_size = max(
hd5[f"{tm.path_prefix}/dose".replace("*", v)].size for v in visits
)
shape = (len(visits), 2, max_size)
if flag_dates:
tensor = np.zeros(shape, dtype=object)
else:
tensor = np.zeros(shape)
for i, visit in enumerate(visits):
path = tm.path_prefix.replace("*", visit)
data = hd5[f"{path}/dose"][()]
tensor[i][0][: data.shape[0]] = data
unix_array = hd5[f"{path}/start_date"][()]
if flag_dates:
tensor[i][1][: unix_array.shape[0]] = get_local_timestamps(
unix_array,
)
else:
tensor[i][1][: unix_array.shape[0]] = unix_array
elif tm.is_event or tm.is_continuous:
max_size = max(hd5[tm.path_prefix.replace("*", v)].size for v in visits)
shape = (len(visits), max_size)
tensor = np.zeros(shape)
for i, visit in enumerate(visits):
path = tm.path_prefix.replace("*", visit)
data = hd5[path][()]
tensor[i][: data.shape[0]] = data
else:
raise ValueError(
f"Incorrect interpretation '{tm.interpretation}' "
"for medication tmap {tm.name}.",
)
return tensor
def _tensor_from_file(tm, hd5, **kwargs):
visits = get_visits(tm, hd5, **kwargs)
flag_dates = kwargs.get("readable_dates")
sizes = []
for visit in visits:
size = 0
for event_name in events_names_list:
try:
size += hd5[
tm.path_prefix.replace("*", visit).replace("{}", event_name)
].size
except KeyError:
continue
sizes.append(size)
max_size = max(sizes)
shape = (len(visits), max_size)
if flag_dates:
tensor = np.zeros(shape, dtype=object)
else:
tensor = np.zeros(shape)
for i, visit in enumerate(visits):
subpath = tm.path_prefix.replace("*", visit)
data = []
for event_name in events_names_list:
try:
path = subpath.replace("{}", event_name)
data = np.append(data, hd5[path][()])
except KeyError:
continue
data = np.sort(data)
if flag_dates:
tensor[i][: data.shape[0]] = get_local_timestamps(data)
else:
tensor[i][: data.shape[0]] = data
return tensor
def _tensor_from_file(tm, hd5, **kwargs):
visits = get_visits(tm, hd5, **kwargs)
max_size = max(hd5[tm.path_prefix.replace("*", v)].size
for v in visits)
if tm.is_timeseries:
shape = (len(visits), 2, max_size)
flag_dates = kwargs.get("readable_dates")
if flag_dates:
tensor = np.zeros(shape, dtype=object)
else:
tensor = np.zeros(shape)
for i, visit in enumerate(visits):
path = tm.path_prefix.replace("*", visit)
time = get_tmap(ref_peak).tensor_from_file(
get_tmap(ref_peak),
hd5,
visit=visit,
)[0]
data = hd5[path][()]
tensor[i][0][:data.shape[0]] = data
if flag_dates:
tensor[i][1][:time.shape[0]] = get_local_timestamps(time)
else:
tensor[i][1][:time.shape[0]] = time
elif tm.is_continuous:
shape = (len(visits), max_size)
tensor = np.zeros(shape)
for i, visit in enumerate(visits):
path = tm.path_prefix.replace("*", visit)
data = hd5[path][()]
tensor[i][:data.shape[0]] = data
else:
raise ValueError(
"Incorrect interpretation "
f"'{tm.interpretation}' for ecg_feature tmap {tm.name}.", )
return tensor
def _tensor_from_file(tm, hd5):
visits = get_visits(tm, hd5)
shape = tm.shape
tensor = np.zeros(shape)
for visit in visits:
path = tm.path_prefix.replace("*", visit)
dates = hd5[path][()]
if dates.shape[0] > 0:
dates = get_local_timestamps(dates)
for date in dates:
date = str(pd.to_datetime(date))
move_in = np.where(
hd5[f"edw/{visit}"].attrs["move_time"].astype(str) <
date, )[0][-1]
department = (hd5[f"edw/{visit}"].attrs["department_nm"].
astype(str)[move_in])
if both_sources and not has_data(hd5, visit, "bedmaster"):
continue
try:
tensor[tm.channel_map[department.lower()]] += 1
except KeyError:
tensor[tm.channel_map["other"]] += 1
return tensor
def _tensor_from_file(tm, hd5, **kwargs):
visits = get_visits(tm, hd5, **kwargs)
base_path = tm.path_prefix
storage_type = type(hd5[base_path.replace("*", visits[0])].attrs[field])
if storage_type == np.ndarray:
max_size = max(
hd5[base_path.replace("*", v)].attrs[field].size for v in visits
)
shape = (len(visits), max_size)
else:
shape = (len(visits),)
tensor = np.zeros(shape) if numeric else np.full(shape, "", object)
for i, visit in enumerate(visits):
path = base_path.replace("*", visit)
if storage_type == np.ndarray:
data = hd5[path].attrs[field]
tensor[i][: data.shape[0]] = data
else:
tensor[i] = hd5[path].attrs[field]
return tensor if numeric else tensor.astype("U")
def _tensor_from_file(tm, hd5, **kwargs):
visits = get_visits(tm, hd5, **kwargs)
flag_dates = kwargs.get("readable_dates")
if tm.interpretation == Interpretation.TIMESERIES:
max_size = max(
hd5[f"{tm.path_prefix}/value".replace("*", v)].size for v in visits
)
shape = (len(visits), 2, max_size)
if flag_dates:
tensor = np.zeros(shape, dtype=object)
else:
tensor = np.zeros(shape)
for i, visit in enumerate(visits):
path = tm.path_prefix.replace("*", visit)
values = hd5[f"{path}/value"][()].astype(str)
for index, value in enumerate(values):
if value == "nan":
tensor[i][0][index] = value
elif position is not None:
pressures = [float(pressure) for pressure in value.split("/")]
tensor[i][0][index] = pressures[position]
else:
pressures = [float(pressure) for pressure in value.split("/")]
tensor[i][0][index] = (
pressures[0] - pressures[-1]
) / pressures[0]
unix_array = hd5[f"{path}/time"][()]
if flag_dates:
tensor[i][1][: unix_array.shape[0]] = get_local_timestamps(
unix_array,
)
else:
tensor[i][1][: unix_array.shape[0]] = unix_array
elif (
tm.interpretation == Interpretation.EVENT
or tm.interpretation == Interpretation.CONTINUOUS
):
max_size = max(hd5[tm.path_prefix.replace("*", v)].size for v in visits)
shape = (len(visits), max_size)
tensor = np.zeros(shape)
for i, visit in enumerate(visits):
path = tm.path_prefix.replace("*", visit)
values = hd5[path][()].astype(str)
for index, value in enumerate(values):
if value == "nan" or tm.interpretation == Interpretation.EVENT:
tensor[i][index] = value
elif position is not None:
pressures = [float(pressure) for pressure in value.split("/")]
tensor[i][index] = pressures[position]
else:
pressures = [float(pressure) for pressure in value.split("/")]
if pressures[0] and pressures[0] != 0:
tensor[i][index] = (
pressures[0] - pressures[-1]
) / pressures[0]
else:
tensor[i][index] = pressures[0]
else:
raise ValueError(
f"Incorrect interpretation '{tm.interpretation}' for measurement tmap.",
)
return tensor
def _tensor_from_file(tm, hd5, **kwargs) -> np.ndarray:
visit_ids = get_visits(tm, hd5, **kwargs)
base_path = tm.path_prefix
if tm.is_continuous:
max_size = max(
hd5[f"{base_path}/{field}".replace("*", v)].size for v in visit_ids
)
shape: Tuple = (len(visit_ids), max_size)
tensor = np.zeros(shape, dtype=dtype)
for i, visit in enumerate(visit_ids):
path = base_path.replace("*", visit)
data = (
hd5[f"{path}/{field}"]
if field != "value"
else get_value(hd5, path, **kwargs)
)
tensor[i][: data.shape[0]] = data[()]
elif tm.is_timeseries:
flag_dates = kwargs.get("readable_dates")
max_size = max(
hd5[f"{base_path}/{field}".replace("*", v)].size
if kwargs.get("interpolation") != "complete_nans_end"
else get_timeseries(hd5, base_path.replace("*", v), field, **kwargs)[
0
].size
for v in visit_ids
)
shape = (len(visit_ids), 2, max_size)
if flag_dates:
tensor = np.zeros(shape, dtype=object)
else:
tensor = np.zeros(shape)
for i, visit in enumerate(visit_ids):
path = base_path.replace("*", visit)
data, unix_array = get_timeseries(hd5, path, field, **kwargs)
tensor[i][0][: data.shape[0]] = data
if flag_dates:
tensor[i][1][: unix_array.shape[0]] = get_local_timestamps(
unix_array,
)
else:
tensor[i][1][: unix_array.shape[0]] = unix_array
elif tm.is_event:
flag_dates = kwargs.get("readable_dates")
max_size = max(
hd5[f"{base_path}/{field}".replace("*", v)].size
if kwargs.get("interpolation") != "complete_nans_end"
else get_timeseries(hd5, base_path.replace("*", v), field, **kwargs)[
0
].size
for v in visit_ids
)
shape = (len(visit_ids), max_size)
if flag_dates:
tensor = np.zeros(shape, dtype=object)
else:
tensor = np.zeros(shape)
for i, visit in enumerate(visit_ids):
path = base_path.replace("*", visit)
_, unix_array = get_timeseries(hd5, path, field, **kwargs)
if flag_dates:
tensor[i][: unix_array.shape[0]] = get_local_timestamps(unix_array)
else:
tensor[i][: unix_array.shape[0]] = unix_array
else:
raise ValueError(
f"Incorrect interpretation "
f"'{tm.interpretation}' for Bedmaster tmap {tm.name}.",
)
return tensor
