PyLaDy

Usage

  import pylady as pl

Descriptor selection

  g2 = pl.descriptors.G2(n_g2_eta=3, eta_max_g2=1.1)
  g3 = pl.descriptors.Descriptor(desc_type=2)

Compute descriptor (no fitting)

    d = pylady.Descriptor(desc_type=1, r_cut=r)
    d.compute([pylady.System(f) for f in f_list])

Database definition

• Each atomic configuration is stored in a pl.System object
• Related pl.System objects are gathered in pl.Collection
• The pl.Database gathers a set of Collection objects.

  from glob import glob

  mydb = pl.Database(weight_per_element=[1.0, 2.0])  
  
  mydb.add(pl.Collection(name='bulk_300K', 
                             systems=[pl.System(poscar="bulk_300K_01.poscar"), 
                                      pl.System(poscar="bulk_300K_02.poscar")],
                             w_energy = 1.e6, # enforce a value of weight instead of optimizing
                             w_force  = 0,    # do not fit forces
                             test_size=0.33)) # size of the test set

  mydb.add(pl.Collection(name='some_defect',
                             systems=[pl.System(poscar=p) for p in glob("some/pattern.poscar")],
                             w_energy = [1.e1, 1.e7], # range for fitting weight
                             w_force  = [1.e3, 1.e6], 
                             w_stress = [1.e2, 1.e6])) # efs for 'energy, force, stress' 
 
  # shows the db as a pandas.DataFrame with 1 system per line, and columns: 
  # name, system filename, w_energy_min, w_energy_max, ..., fit_with, weight_per_element 
  print(mydb.as_df())
  
  # save and reload DB
  mydb.as_df().to_json("mydb.json")
  df = pylady.Database.from_json("mydb.json").as_df()

Model training

  for desc in (g2, g3):

    mymodel = pl.Model(ml_type=-1, descriptor=desc, database=mydb)
  
    # Run milady with -np 4, in a tmp dir and with additional kwargs
    mymodel.fit(njobs=4, save_directory="milady/run_01", n_g3_lambda=3, seed=24)
  
    # results are accessible through the Model object.  
    # To access descriptors of first system of the "some_defect" collection :
    print(mymodel.db["some_defect"][0].descriptors) 
    print(mymodel.design_matrix)

Tools

Available as CLI & python funcs

• Configure the path to Milady binary in ~/.config/milady_config.json

  pylady create-config

  pylady.cli.create_config(set_env_cmd, run_milady_cmd, force, location)

• Check a valid path to Milady binary is set in ~/.config/milady_config.json and the file is accessible

  pylady check-config

  pylady.cli.check_config()

• Plot train/test accuracy:

  pylady plot -db_in_file db_model.in --test "test_*.out" --train "train*.out" --save "graph.pdf"
  
  pylady plot --db db.json --fit_data fit.h5 --save "graph.pdf"

  import pylady.plot
  pylady.plot(dbfile="db_model.in", test_files="test*.out", train_files="train*.out")

• Save configurations found in MD outcar files matching the glob pattern "file_*.outcar". Only save one configuration every 60 timesteps (default is 0, i.e. only save the first step). Saved files have the name file_*_{step}.mlposcar.

  pylady convert --in "file_*.outcar" --out "dir/" --read_every 60  

  import pylady.convert
  pylady.convert(in="file_*.outcar", out="dir/", read_every=60)

CI/CD

• Tested on python 3.6 to 3.9