# MODULE: grass.jupyter.timeseriesmap
#
# AUTHOR(S): Caitlin Haedrich <caitlin DOT haedrich AT gmail>
#
# PURPOSE: This module contains functions for visualizing raster and vector
# space-time datasets in Jupyter Notebooks
#
# COPYRIGHT: (C) 2022 Caitlin Haedrich, and by the GRASS Development Team
#
# This program is free software under the GNU General Public
# License (>=v2). Read the file COPYING that comes with GRASS
# for details.
"""Create and display visualizations for space-time datasets."""
import tempfile
import os
import weakref
import shutil
import grass.script as gs
from .map import Map
from .region import RegionManagerForTimeSeries
[docs]def fill_none_values(names):
"""Replace `None` values in array with previous item"""
for i, name in enumerate(names):
if name == "None":
names[i] = names[i - 1]
else:
pass
return names
[docs]def collect_layers(timeseries, element_type, fill_gaps):
"""Create lists of layer names and start_times for a
space-time raster or vector dataset.
For datasets with variable time steps, makes step regular with
"gran" method for t.rast.list or t.vect.list then fills in
missing layers with previous time step layer.
:param str timeseries: name of space-time dataset
:param str element_type: element type, "stvds" or "strds"
:param bool fill_gaps: fill empty time steps with data from previous step
"""
# NEW WAY: Comment in after json output for t.rast.list and t.vect.list is merged
# import json
# if element_type == "strds":
# result = json.loads(
# gs.read_command(
# "t.rast.list", method="gran", input=timeseries, format="json"
# )
# )
# elif element_type == "stvds":
# result = json.loads(
# gs.read_command(
# "t.vect.list", method="gran", input=timeseries, format="json"
# )
# )
# else:
# raise NameError(
# _("Dataset {} must be element type 'strds' or 'stvds'").format(timeseries)
# )
#
# # Get layer names and start time from json
# names = [item["name"] for item in result["data"]]
# dates = [item["start_time"] for item in result["data"]]
if element_type == "strds":
rows = gs.read_command(
"t.rast.list", method="gran", input=timeseries
).splitlines()
elif element_type == "stvds":
rows = gs.read_command(
"t.vect.list", method="gran", input=timeseries
).splitlines()
else:
raise NameError(
_("Dataset {} must be element type 'strds' or 'stvds'").format(timeseries)
)
# Parse string
# Create list of list
new_rows = [row.split("|") for row in rows]
# Transpose into columns where the first value is the name of the column
new_array = [list(row) for row in zip(*new_rows)]
# Collect layer name and start time
for column in new_array:
if column[0] == "name":
names = column[1:]
if column[0] == "start_time":
dates = column[1:]
# For datasets with variable time steps, fill in gaps with
# previous time step value, if fill_gaps==True.
if fill_gaps:
names = fill_none_values(names)
return names, dates
[docs]def check_timeseries_exists(timeseries, element_type):
"""Check that timeseries is time space dataset"""
test = gs.read_command("t.list", type=element_type, where=f"name='{timeseries}'")
if not test:
raise NameError(
_("Could not find space time dataset named {} of type {}").format(
timeseries, element_type
)
)
[docs]class TimeSeriesMap:
"""Creates visualizations of time-space raster and vector datasets in Jupyter
Notebooks.
Basic usage::
>>> img = TimeSeriesMap("series_name")
>>> img.d_legend() # Add legend
>>> img.show() # Create TimeSlider
>>> img.save("image.gif")
This class of grass.jupyter is experimental and under development. The API can
change at anytime.
"""
# pylint: disable=too-many-instance-attributes
# Need more attributes to build timeseriesmap visuals
def __init__(
self,
width=None,
height=None,
env=None,
use_region=False,
saved_region=None,
):
"""Creates an instance of the TimeSeriesMap visualizations class.
:param int width: width of map in pixels
:param int height: height of map in pixels
:param str env: environment
:param use_region: if True, use either current or provided saved region,
else derive region from rendered layers
:param saved_region: if name of saved_region is provided,
this region is then used for rendering
"""
# Copy Environment
if env:
self._env = env.copy()
else:
self._env = os.environ.copy()
self.timeseries = None
self._element_type = None
self._fill_gaps = None
self._legend = None
self._base_layer_calls = []
self._overlay_calls = []
self._timeseries_added = False
self._layers_rendered = False
self._layers = None
self._dates = None
self._date_layer_dict = {}
self._date_filename_dict = {}
self._width = width
self._height = height
# Create a temporary directory for our PNG images
# Resource managed by weakref.finalize.
self._tmpdir = (
# pylint: disable=consider-using-with
tempfile.TemporaryDirectory()
)
def cleanup(tmpdir):
tmpdir.cleanup()
weakref.finalize(self, cleanup, self._tmpdir)
# Handle Regions
self._region_manager = RegionManagerForTimeSeries(
use_region, saved_region, self._env
)
[docs] def add_raster_series(self, timeseries, fill_gaps=False):
"""
:param str timeseries: name of space-time dataset
:param bool fill_gaps: fill empty time steps with data from previous step
"""
if self._timeseries_added and self.timeseries != timeseries:
raise AttributeError("Cannot add more than one space time dataset")
self._element_type = "strds"
check_timeseries_exists(timeseries, self._element_type)
self.timeseries = timeseries
self._fill_gaps = fill_gaps
self._timeseries_added = True
# create list of layers to render and date/times
self._layers, self._dates = collect_layers(
self.timeseries, self._element_type, self._fill_gaps
)
self._date_layer_dict = {
self._dates[i]: self._layers[i] for i in range(len(self._dates))
}
# Update Region
self._region_manager.set_region_from_timeseries(self.timeseries)
[docs] def add_vector_series(self, timeseries, fill_gaps=False):
"""
:param str timeseries: name of space-time dataset
:param bool fill_gaps: fill empty time steps with data from previous step
"""
if self._timeseries_added and self.timeseries != timeseries:
raise AttributeError("Cannot add more than one space time dataset")
self._element_type = "stvds"
check_timeseries_exists(timeseries, self._element_type)
self.timeseries = timeseries
self._fill_gaps = fill_gaps
self._timeseries_added = True
# create list of layers to render and date/times
self._layers, self._dates = collect_layers(
self.timeseries, self._element_type, self._fill_gaps
)
self._date_layer_dict = {
self._dates[i]: self._layers[i] for i in range(len(self._dates))
}
# Update Region
self._region_manager.set_region_from_timeseries(self.timeseries)
def __getattr__(self, name):
"""Parse attribute to GRASS display module. Attribute should be in
the form 'd_module_name'. For example, 'd.rast' is called with 'd_rast'.
"""
# Check to make sure format is correct
if not name.startswith("d_"):
raise AttributeError(_("Module must begin with 'd_'"))
# Reformat string
grass_module = name.replace("_", ".")
# Assert module exists
if not shutil.which(grass_module):
raise AttributeError(_("Cannot find GRASS module {}").format(grass_module))
def wrapper(**kwargs):
if not self._timeseries_added:
self._base_layer_calls.append((grass_module, kwargs))
if self._timeseries_added:
self._overlay_calls.append((grass_module, kwargs))
return wrapper
[docs] def d_legend(self, **kwargs):
"""Display legend.
Wraps d.legend and uses same keyword arguments.
"""
if "raster" in kwargs and not self._timeseries_added:
self._base_layer_calls.append(("d.legend", kwargs))
if "raster" in kwargs and self._timeseries_added:
self._overlay_calls.append(("d.legend", kwargs))
else:
self._legend = kwargs
# If d_legend has been called, we need to re-render layers
self._layers_rendered = False
def _render_baselayers(self, img):
"""Add collected baselayers to Map instance"""
for grass_module, kwargs in self._base_layer_calls:
img.run(grass_module, **kwargs)
def _render_legend(self, img):
"""Add legend to Map instance"""
info = gs.parse_command(
"t.info", input=self.timeseries, flags="g", env=self._env
)
min_min = info["min_min"]
max_max = info["max_max"]
img.d_legend(
raster=self._layers[0],
range=f"{min_min}, {max_max}",
**self._legend,
)
def _render_overlays(self, img):
"""Add collected overlays to Map instance"""
for grass_module, kwargs in self._overlay_calls:
img.run(grass_module, **kwargs)
def _render_blank_layer(self, filename):
"""Write blank image for gaps in time series.
Adds overlays and legend to base map.
"""
img = Map(
width=self._width,
height=self._height,
filename=filename,
use_region=True,
env=self._env,
read_file=True,
)
# Add overlays
self._render_overlays(img)
# Add legend if needed
if self._legend:
self._render_legend(img)
def _render_layer(self, layer, filename):
"""Render layer to file with overlays and legend"""
img = Map(
width=self._width,
height=self._height,
filename=filename,
use_region=True,
env=self._env,
read_file=True,
)
if self._element_type == "strds":
img.d_rast(map=layer)
elif self._element_type == "stvds":
img.d_vect(map=layer)
# Add overlays
self._render_overlays(img)
# Add legend if needed
if self._legend:
self._render_legend(img)
[docs] def render(self):
"""Renders image for each time-step in space-time dataset.
Save PNGs to temporary directory. Must be run before creating a visualization
(i.e. show or save). Can be time-consuming to run with large
space-time datasets.
"""
if not self._timeseries_added:
raise RuntimeError(
"Cannot render space time dataset since none has been added."
"Use TimeSeriesMap.add_raster_series() or "
"TimeSeriesMap.add_vector_series() to add dataset"
)
# Make base image (background and baselayers)
# Random name needed to avoid potential conflict with layer names
random_name_base = gs.append_random("base", 8) + ".png"
base_file = os.path.join(self._tmpdir.name, random_name_base)
img = Map(
width=self._width,
height=self._height,
filename=base_file,
use_region=True,
env=self._env,
read_file=True,
)
# We have to call d_erase to ensure the file is created. If there are no
# base layers, then there is nothing to render in random_base_name
img.d_erase()
# Add baselayers
self._render_baselayers(img)
# Create name for empty layers
# Random name needed to avoid potential conflict with layer names
# A new random_name_none is created each time the render function is run,
# and any existing random_name_none file will be ignored
random_name_none = gs.append_random("none", 8) + ".png"
# Render each layer
for date, layer in self._date_layer_dict.items():
if layer == "None":
# Create file
filename = os.path.join(self._tmpdir.name, random_name_none)
self._date_filename_dict[date] = filename
# Render blank layer if it hasn't been done already
if not os.path.exists(filename):
shutil.copyfile(base_file, filename)
self._render_blank_layer(filename)
else:
# Create file
filename = os.path.join(self._tmpdir.name, f"{layer}.png")
# Copying the base_file ensures that previous results are overwritten
shutil.copyfile(base_file, filename)
self._date_filename_dict[date] = filename
# Render image
self._render_layer(layer, filename)
self._layers_rendered = True
[docs] def show(self, slider_width=None):
"""Create interactive timeline slider.
param str slider_width: width of datetime selection slider
The slider_width parameter sets the width of the slider in the output cell.
It should be formatted as a percentage (%) between 0 and 100 of the cell width
or in pixels (px). Values should be formatted as strings and include the "%"
or "px" suffix. For example, slider_width="80%" or slider_width="500px".
slider_width is passed to ipywidgets in ipywidgets.Layout(width=slider_width).
"""
# Lazy Imports
import ipywidgets as widgets # pylint: disable=import-outside-toplevel
# Render images if they have not been already
if not self._layers_rendered:
self.render()
# Set default slider width
if not slider_width:
slider_width = "70%"
# Datetime selection slider
slider = widgets.SelectionSlider(
options=self._dates,
value=self._dates[0],
description=_("Date/Time"),
disabled=False,
continuous_update=True,
orientation="horizontal",
readout=True,
layout=widgets.Layout(width=slider_width),
)
play = widgets.Play(
interval=500,
value=0,
min=0,
max=len(self._dates) - 1,
step=1,
description="Press play",
disabled=False,
)
out_img = widgets.Image(value=b"", format="png")
def change_slider(change):
slider.value = slider.options[change.new]
play.observe(change_slider, names="value")
# Display image associated with datetime
def change_image(date):
# Look up layer name for date
filename = self._date_filename_dict[date]
with open(filename, "rb") as rfile:
out_img.value = rfile.read()
# Return interact widget with image and slider
widgets.interactive_output(change_image, {"date": slider})
layout = widgets.Layout(
width="100%", display="inline-flex", flex_flow="row wrap"
)
return widgets.HBox([play, slider, out_img], layout=layout)
[docs] def save(
self,
filename,
duration=500,
label=True,
font="DejaVuSans.ttf",
text_size=12,
text_color="gray",
):
"""
Creates a GIF animation of rendered layers.
Text color must be in a format accepted by PIL ImageColor module. For supported
formats, visit:
https://pillow.readthedocs.io/en/stable/reference/ImageColor.html#color-names
param str filename: name of output GIF file
param int duration: time to display each frame; milliseconds
param bool label: include date/time stamp on each frame
param str font: font file
param int text_size: size of date/time text
param str text_color: color to use for the text.
"""
# Create a GIF from the PNG images
import PIL.Image # pylint: disable=import-outside-toplevel
import PIL.ImageDraw # pylint: disable=import-outside-toplevel
import PIL.ImageFont # pylint: disable=import-outside-toplevel
# Render images if they have not been already
if not self._layers_rendered:
self.render()
# filepath to output GIF
if not filename.endswith(".gif"):
raise ValueError(_("filename must end in '.gif'"))
images = []
for date in self._dates:
img_path = self._date_filename_dict[date]
img = PIL.Image.open(img_path)
img = img.convert("RGBA", dither=None)
draw = PIL.ImageDraw.Draw(img)
if label:
draw.text(
(0, 0),
date,
fill=text_color,
font=PIL.ImageFont.truetype(font, text_size),
)
images.append(img)
images[0].save(
fp=filename,
format="GIF",
append_images=images[1:],
save_all=True,
duration=duration,
loop=0,
)
# Display the GIF
return filename
