Performing workflows on database structures

[becca9835]

I am writing a workflow that will depend on previous calculations performed with simmate. For example, I have performed relaxations on many structures and now want to perform static energy calculations on those structures that have already been relaxed and are in the simmate database. I am having trouble figuring out to: 1) pull the "structure" column from a database table to perform calculations on, 2) filter to only perform my new workflow on the wanted structures and 3) exclude duplicates in the table so I am not performing the static energy calculation on multiple of the same structure that is present in the relaxation table. I know that the fifth line of code, structures = workflows_relaxation.objects.filter( is incorrect because it does not call the specific "structure" column that I need, but I am not sure how to make it work with also filtering.

from simmate.database import connect
from simmate.workflows.utilities import get_workflow
##do I need to import anything else to be able to get structures from database tables?

    structures = workflows_relaxation.objects.filter(
    spacegroup=148,
    formula_anonymous="ABC6",
    ##how do we exclude duplicates with same formula_reduced?,
).all()

data = structures.to_dataframe()
data.to_csv("mydata.csv")

static_workflow = get_workflow("static-energy.vasp.matproj")
static_jobs = []
for structure in structures:
    status = static_workflow.run_cloud(
        structure=structure,
    tags = ["radomster"],
        command="mpirun vasp_std > vasp.out",
    )
    static_jobs.append(status)

[jacksund]

1. You don't need to pull the structure column! You can convert the search results directly to toolkit (pymatgen-like) objects using the to_toolkit method. But since you are using these for a new calculation, you can actually just give the database object to your workflow.
2. You have the filtering down. To see the columns that you can filter with or do some more complex queries, you can read through the docs here
3. If you only wanted 1 structure returned in your search, you can call first() on your search results. You might also want to do something like "give me the most recent one" in which case you would do order_by("-created_at").first(). Your case is slightly different where you want to remove duplicates in the formula_full column, in which case you'd add distinct("formula_full"). Be careful though because the distinct method only works with postgres, and not the default database sqlite. Read the distinct docs if you need more complex logic.

Try this updated script:

from simmate.database import connect
from simmate.workflows.utilities import get_workflow

# Grab the database table from our workflow object
relax_workflow = get_workflow("relaxation.vasp.matproj")
relax_results = relax_workflow.all_results

# Filter down to the structure you'd like
structures = (
    relax_results.filter(
        spacegroup=148,
        formula_anonymous="ABC6",
    )
    .distinct("formula_full")
    .all()
)

# Then submit each of these to the new workflow.
# There's no need to grab the structure column because Simmate does it for you
static_workflow = get_workflow("static-energy.vasp.matproj")
static_jobs = []
for structure in structures:
    status = static_workflow.run_cloud(
        structure=structure,  
        tags = ["radomster"],
        command="mpirun vasp_std > vasp.out",
    )
    static_jobs.append(status)

[becca9835]

Thank you so much, this was very helpful! The script seems to run well (yay!) but unfortunately it seems like it is still picking up duplicate structures from the database? Is there an easy fix for this?

[jacksund]

whoops. sorry I was trying to get too fancy with the distinct call. I updated the script in my comment above to fix it