import logging
import os
import traceback
import warnings
from collections import defaultdict
from pathlib import Path
import numpy as np
import pandas as pd
from .objects import Epithelium
from .sheet import Sheet
logger = logging.getLogger(name=__name__)
def _filter_columns(cols_hist, cols_in, element):
if not set(cols_hist).issubset(cols_in):
warnings.warn(
f"""
Columns {set(cols_hist).difference(cols_in)} are in the history
{element} dataframe but not in the sheet {element} dataframe.
These non existent columns will not be saved."""
)
cols_hist = set(cols_hist).intersection(cols_in)
return list(cols_hist)
[docs]class History:
"""This class handles recording and retrieving time series
of sheet objects.
"""
def __init__(
self,
sheet,
save_every=None,
dt=None,
save_only=None,
extra_cols=None,
save_all=True,
):
"""Creates a `SheetHistory` instance.
Parameters
----------
sheet : a :class:`Sheet` object which we want to record
save_every : float, set every time interval to save the sheet
dt : float, time step
save_only: dict : dictionnary with sheet.datasets as keys and list of
columns as values. Default None
extra_cols : dictionnary with sheet.datasets as keys and list of
columns as values. Default None
save_all : bool
if True, saves all the data at each time point
"""
if extra_cols is not None:
warnings.warn(
"extra_cols and save_all parameters are deprecated."
" Use save_only instead. "
)
if save_only is not None:
extra_cols = defaultdict(list, **save_only)
else:
extra_cols = {k: list(sheet.datasets[k].columns) for k in sheet.datasets}
self.sheet = sheet
self.time = 0.0
self.index = 0
if save_every is not None:
self.save_every = save_every
self.dt = dt
else:
self.save_every = None
self.datasets = {}
self.columns = {}
vcols = sheet.coords + extra_cols["vert"]
vcols = list(set(vcols))
self.vcols = _filter_columns(vcols, sheet.vert_df.columns, "vertex")
_vert_h = sheet.vert_df[self.vcols].reset_index(drop=False)
if "time" not in self.vcols:
_vert_h["time"] = 0
self.datasets["vert"] = _vert_h
self.columns["vert"] = self.vcols
fcols = extra_cols["face"]
self.fcols = _filter_columns(fcols, sheet.face_df.columns, "face")
_face_h = sheet.face_df[self.fcols].reset_index(drop=False)
if "time" not in self.fcols:
_face_h["time"] = 0
self.datasets["face"] = _face_h
self.columns["face"] = self.fcols
if sheet.cell_df is not None:
ccols = extra_cols["cell"]
self.ccols = _filter_columns(ccols, sheet.cell_df.columns, "cell")
_cell_h = sheet.cell_df[self.ccols].reset_index(drop=False)
if "time" not in self.ccols:
_cell_h["time"] = 0
self.datasets["cell"] = _cell_h
self.columns["cell"] = self.ccols
extra_cols["edge"].append("cell")
ecols = ["srce", "trgt", "face"] + extra_cols["edge"]
ecols = list(set(ecols))
self.ecols = _filter_columns(ecols, sheet.edge_df.columns, "edge")
_edge_h = sheet.edge_df[self.ecols].reset_index(drop=False)
if "time" not in self.ecols:
_edge_h["time"] = 0
self.datasets["edge"] = _edge_h
self.columns["edge"] = self.ecols
def __len__(self):
return self.time_stamps.__len__()
[docs] def to_archive(self, hf5file):
"""Saves the history to a HDF file
This file can later be accessed again with the `HistoryHdf5.from_archive`
class method
"""
with pd.HDFStore(hf5file, "a") as store:
for key, df in self.datasets.items():
kwargs = {"data_columns": ["time"]}
if "segment" in df.columns:
kwargs["min_itemsize"] = {"segment": 7}
store.append(key=key, value=df, **kwargs)
@property
def time_stamps(self):
return self.datasets["vert"]["time"].unique()
@property
def vert_h(self):
return self.datasets["vert"]
@property
def edge_h(self):
return self.datasets["edge"]
@property
def face_h(self):
return self.datasets["face"]
@property
def cell_h(self):
return self.datasets.get("cell", None)
[docs] def record(self, time_stamp=None):
"""Appends a copy of the sheet datasets to the history instance.
Parameters
----------
time_stamp : float, save specific timestamp
"""
if time_stamp is not None:
self.time = time_stamp
else:
self.time += 1
if (self.save_every is None) or (
self.index % (int(self.save_every / self.dt)) == 0
):
for element in self.datasets:
hist = self.datasets[element]
cols = self.columns[element]
df = self.sheet.datasets[element][cols].reset_index(drop=False)
if "time" not in cols:
times = pd.Series(np.ones((df.shape[0],)) * self.time, name="time")
df = pd.concat([df, times], ignore_index=False, axis=1, sort=False)
else:
df["time"] = self.time
if self.time in hist["time"]:
# erase previously recorded time point
hist = hist[hist["time"] != self.time]
hist = pd.concat([hist, df], ignore_index=True, axis=0, sort=False)
self.datasets[element] = hist
self.index += 1
[docs] def retrieve(self, time):
"""Return datasets at time `time`.
If a specific dataset was not recorded at time time,
the closest record before that time is used.
"""
if time > self.datasets["vert"]["time"].values[-1]:
warnings.warn(
"""
The time argument you requested is bigger than the maximum recorded time,
are you sure you passed the time stamp as parameter, and not an index ?
"""
)
sheet_datasets = {}
for element in self.datasets:
hist = self.datasets[element]
cols = self.columns[element]
df = _retrieve(hist, time)
df = df.set_index(element)[cols]
sheet_datasets[element] = df
return type(self.sheet)(
f"{self.sheet.identifier}_{time:04.3f}", sheet_datasets, self.sheet.specs
)
def __iter__(self):
"""Iterates over all the time points of the history"""
for t in self.time_stamps:
sheet = self.retrieve(t)
yield t, sheet
[docs] def slice(self, start=0, stop=None, size=None, endpoint=True):
"""Returns a slice of the history's time_stamps array
The slice is over or under sampled to have exactly size point
between start and stop
"""
if size is not None:
if stop is not None:
time_stamps = self.time_stamps[start : stop + int(endpoint)]
else:
time_stamps = self.time_stamps
indices = np.round(
np.linspace(0, time_stamps.size + 1, size, endpoint=True)
).astype(int)
times = time_stamps.take(indices.clip(max=time_stamps.size - 1))
elif stop is not None:
times = self.time_stamps[start : stop + int(endpoint)]
else:
times = self.time_stamps
return times
[docs] def browse(self, start=0, stop=None, size=None, endpoint=True):
"""Returns an iterator over part of the history
Parameters
----------
start: int, index of the first time point
stop: int, index of the last time point
size: int, the number of time points to return
endpoint: bool, wether to include the stop time point (default True)
Returns
-------
iterator over (t, sheet(t)) for the retrieved time points
"""
for t in self.slice(start=start, stop=stop, size=size, endpoint=endpoint):
yield t, self.retrieve(t)
[docs]class HistoryHdf5(History):
"""This class handles recording and retrieving time series
of sheet objects.
"""
def __init__(
self,
sheet=None,
save_every=None,
dt=None,
save_only=None,
hf5file="",
overwrite=False,
):
"""Creates a `HistoryHdf5` instance.
Use the `from_archive` class method to re-open a saved history file
Parameters
----------
sheet : a :class:`Sheet` object which we want to record
save_every : float, set every time interval to save the sheet
dt : float, time step
save_only : dictionnary with sheet.datasets as keys and list of
columns as values. Default None
hf5file : string, define the path of the HDF5 file
overwrite : bool, Overwrite or not the file if it is already exist.
Default False
"""
if not hf5file:
warnings.warn(
"No directory is given. The HDF5 file will be saved"
" in the working directory as out.hf5."
)
self.hf5file = Path(os.getcwd()) / "out.hf5"
else:
self.hf5file = Path(hf5file)
if self.hf5file.exists():
if overwrite:
tb = traceback.extract_stack(limit=2)
if "from_archive" not in tb[0].name:
warnings.warn(
"The file already exist and will be overwritten."
" This is normal if you reopened an archive"
)
else:
expand = 0
while True:
new_hf5file = self.hf5file.parent / self.hf5file.name.replace(
self.hf5file.suffix, f"{expand}{self.hf5file.suffix}"
)
expand += 1
if new_hf5file.exists():
continue
else:
self.hf5file = new_hf5file
warnings.warn(
"The file already exist and the new filename is {}".format(
new_hf5file
)
)
break
if sheet is None:
last = self.time_stamps[-1]
with pd.HDFStore(self.hf5file, "r") as file:
keys = file.keys()
if r"\cell" in keys:
sheet = Epithelium(last)
History.__init__(self, sheet, save_every, dt, save_only)
self.dtypes = {
k: df[self.columns[k]].dtypes for k, df in sheet.datasets.items()
}
[docs] @classmethod
def from_archive(cls, hf5file, columns=None, eptm_class=None):
datasets = {}
settings = {}
hf5file = Path(hf5file)
with pd.HDFStore(hf5file, "r") as store:
keys = [k.strip("/") for k in store.keys()]
if columns is None:
# read everything
columns = {k: None for k in keys}
if eptm_class is None:
eptm_class = Epithelium if "cell" in keys else Sheet
last = store.select("vert", columns=["time"]).iloc[-1]["time"]
for key in keys:
if key == "settings":
settings = store[key]
continue
df = store.select(key, where=f"time == {last}", columns=columns[key])
datasets[key] = df
eptm = eptm_class(hf5file.name, datasets)
eptm.settings.update(settings)
return cls(sheet=eptm, hf5file=hf5file, overwrite=True)
@property
def time_stamps(self, element="vert"):
with pd.HDFStore(self.hf5file, "r") as file:
times = file.select(element, columns=["time"])["time"].unique()
return times
[docs] def record(self, time_stamp=None, sheet=None):
"""Appends a copy of the sheet datasets to the history HDF file.
Parameters
----------
sheet: a :class:`Sheet` object which we want to record. This argument can
be used if the sheet object is different at each time point.
"""
if sheet is not None:
self.sheet = sheet
if time_stamp is not None:
self.time = time_stamp
else:
self.time += 1.0
dtypes_ = {k: df.dtypes for k, df in self.sheet.datasets.items()}
for element, df in self.sheet.datasets.items():
old_types = self.dtypes[element].to_dict()
new_types = {k: dtypes_[element].to_dict()[k] for k in old_types.keys()}
if new_types != old_types:
changed_type = {
k: old_types[k]
for k in old_types
if k in new_types and old_types[k] != new_types[k]
}
raise ValueError(
f"There is a change of datatype in {element} table"
f" in {changed_type} columns"
)
if (self.save_every is None) or (
self.index % (int(self.save_every / self.dt)) == 0
):
for element, df in self.sheet.datasets.items():
times = pd.Series(np.ones((df.shape[0],)) * self.time, name="time")
df = df[self.columns[element]]
df = pd.concat([df, times], ignore_index=False, axis=1, sort=False)
kwargs = {"data_columns": ["time"]}
if "segment" in df.columns:
kwargs["min_itemsize"] = {"segment": 8}
with pd.HDFStore(self.hf5file, "a") as store:
if (
element in store
and store.select(element, where=f"time == {self.time}")[
"time"
].shape[0]
> 0
):
store.remove(key=element, where=f"time == {self.time}")
store.append(key=element, value=df, **kwargs)
self.index += 1
[docs] def retrieve(self, time):
"""Returns datasets at time `time`.
If a specific dataset was not recorded at time time,
the closest record before that time is used.
"""
times = self.time_stamps
if time > times[-1]:
warnings.warn(
"The time argument you passed is bigger than the maximum recorded time,"
" are you sure you pass time in parameter and not an index ? "
)
time = times[np.argmin(np.abs(times - time))]
with pd.HDFStore(self.hf5file, "r") as store:
sheet_datasets = {}
for element in self.datasets:
sheet_datasets[element] = store.select(element, where=f"time == {time}")
sheet = type(self.sheet)(
f"{self.sheet.identifier}_{time:04.3f}", sheet_datasets, self.sheet.specs
)
sheet.coords = self.sheet.coords
sheet.edge_df.index.rename("edge", inplace=True)
return sheet
[docs] def retrieve_columns(self, element, columns):
"""
Return a table with the selected columns from the given element
Parameters
----------
element: str
either 'vert', 'edge', 'face' or 'cell'
columns: list of str
a list of columns to retrieve
"""
with pd.HDFStore(self.hf5file, "r") as store:
data = store.select(
element,
columns=columns,
)
return data
def _retrieve(dset, time):
times = dset["time"].values
t = times[np.argmin(np.abs(times - time))]
return dset[dset["time"] == t]
