140-configuration displacement-force training set for the compressive-sensing (LASSO)
force-constant fit of Phys. Rev. Lett. 120, 105901 (2018).

  disp_random_0.dat / force_random_0.dat   140 configurations x 54 atoms
  disp_random.dat   / force_random.dat     identical copies

Sampling: ab initio molecular dynamics at 800 K on the +4 % expanded cell, 140 snapshots
uniformly sampled, then every atom displaced by a further 0.1 A in a random direction.

NOTE: the Supplemental Material of the paper gives this temperature as 300 K. That is a typo;
the correct value is 800 K, confirmed by the author.  See section 2.7 of the top-level README.

Only this one training set survives; the per-volume sets for the other four lattice constants
are not preserved.  See section 6 of the top-level README.

TWO sets survive, and they are different data (no shared configurations):
  disp_random_0.dat / force_random_0.dat   most likely the cell0 set
  disp_random.dat   / force_random.dat     most likely the set behind the "+4 %" fit,
                                           whose IFC file is named ..._random_lasso1 (no "2")
Refitting with disp_random_0 reproduces the published cell0 cubic force constants with
Pearson r = 0.9991 and a median ratio of 0.998 on significant terms.
See 04_force_constants/lasso_published/input_alm_6th_order.reconstructed.txt.
