Saving Reloadable Data

Most of the data loaded into or generated with yt can be saved to a format that can be reloaded as a first-class dataset. This includes the following:

  • geometric data containers (regions, spheres, disks, rays, etc.)

  • grid data containers (covering grids, arbitrary grids, fixed resolution buffers)

  • spatial plots (projections, slices, cutting planes)

  • profiles

  • generic array data

In the case of projections, slices, and profiles, reloaded data can be used to remake plots. For information on this, see Remaking Figures from Plot Datasets.

Geometric Data Containers

Data from geometric data containers can be saved with the save_as_dataset() function.

This function will return the name of the file to which the dataset was saved. The filename will be a combination of the name of the original dataset and the type of data container. Optionally, a specific filename can be given with the filename keyword. If no fields are given, the fields that have previously been queried will be saved.

The newly created dataset can be loaded like all other supported data through yt.load. Once loaded, field data can be accessed through the traditional data containers or through the data attribute, which will be a data container configured like the original data container used to make the dataset. Grid data is accessed by the grid data type and particle data is accessed with the original particle type. As with the original dataset, grid positions and cell sizes are accessible with, for example, (“grid”, “x”) and (“grid”, “dx”). Particle positions are accessible as (<particle_type>, “particle_position_x”). All original simulation parameters are accessible in the parameters dictionary, normally associated with all datasets.

sphere_ds = yt.load("DD0046_sphere.h5")

# use the original data container
print(["grid", "density"])

# create a new data container
ad = sphere_ds.all_data()

# grid data
print(ad["grid", "density"])
print(ad["grid", "x"])
print(ad["grid", "dx"])

# particle data
print(ad["all", "particle_mass"])
print(ad["all", "particle_position_x"])

Note that because field data queried from geometric containers is returned as unordered 1D arrays, data container datasets are treated, effectively, as particle data. Thus, 3D indexing of grid data from these datasets is not possible.

Grid Data Containers

Data containers that return field data as multidimensional arrays can be saved so as to preserve this type of access. This includes covering grids, arbitrary grids, and fixed resolution buffers. Saving data from these containers works just as with geometric data containers. Field data can be accessed through geometric data containers.

cg = ds.covering_grid(level=0, left_edge=[0.25] * 3, dims=[16] * 3)
fn = cg.save_as_dataset(fields=[("gas", "density"), ("all", "particle_mass")])

cg_ds = yt.load(fn)
ad = cg_ds.all_data()
print(ad["grid", "density"])

Multidimensional indexing of field data is also available through the data attribute.

print(["grid", "density"])

Fixed resolution buffers work just the same.

my_proj = ds.proj(("gas", "density"), "x", weight_field=("gas", "density"))
frb = my_proj.to_frb(1.0, (800, 800))
fn = frb.save_as_dataset(fields=[("gas", "density")])
frb_ds = yt.load(fn)
print(["gas", "density"])

Spatial Plots

Spatial plots, such as projections, slices, and off-axis slices (cutting planes) can also be saved and reloaded.

proj = ds.proj(("gas", "density"), "x", weight_field=("gas", "density"))

Once reloaded, they can be handed to their associated plotting functions to make images.

proj_ds = yt.load("DD0046_proj.h5")
p = yt.ProjectionPlot(proj_ds, "x", ("gas", "density"), weight_field=("gas", "density"))


Profiles created with create_profile(), ProfilePlot, and PhasePlot can be saved with the save_as_dataset() function, which works just as above. Profile datasets are a type of non-spatial grid datasets. Geometric selection is not possible, but data can be accessed through the .data attribute.

The x, y (if at least 2D), and z (if 3D) bin fields can be accessed as 1D arrays with “x”, “y”, and “z”.

print(["gas", "x"])

The bin fields can also be returned with the same shape as the profile data by accessing them with their original names. This allows for boolean masking of profile data using the bin fields.

# density is the x bin field
print(["gas", "density"])

For 1, 2, and 3D profile datasets, a fake profile object will be constructed by accessing the “.profile” attribute. This is used primarily in the case of 1 and 2D profiles to create figures using ProfilePlot and PhasePlot.

p = yt.PhasePlot(,
    ("gas", "density"),
    ("gas", "temperature"),
    ("gas", "mass"),

Generic Array Data

Generic arrays can be saved and reloaded as non-spatial data using the save_as_dataset() function, also available as yt.save_as_dataset. As with profiles, geometric selection is not possible, but the data can be accessed through the .data attribute.

Array data can be saved with or without a dataset loaded. If no dataset has been loaded, as fake dataset can be provided as a dictionary.