v0.0.4

What's Changed

🎉 New Features

• [6909b61] MonteCarlo supports restarting from a checkpoint that properly stores the structure and the state of the random number generator. The trajectory will be consistent when the simulation restarts from a checkpoint. Use ckpt_period to set the period to save a checkpoint.
• [e7789a3] Some trainers can train a group of models on the same dataset.
• [cacd3f2] The codes for the active session seems to be stable. Users can customise a session to run an active learning workflow.

🥳 Improvements

• [29934e5] AseDriver dumps its parameters for the simulation.
• [274df62] Support archiving computation folders for some drivers, namely, ase, lammps, vasp. As a default, the DriverBasedWorker/SingleWorker archives computation folders when retrieving results. This saves time when backing up data.

🐞 Bug Fixes

• [fe601c5] LmpDriver cannot correctly deal with largely tilted structures.
• [a63fa0e] [5860b84] LaspDriver cannot read trajectories with terrible structures (most of them have too short bonds).
• [8559eb8][e353f42][1335534] LaspDriver cannot read invalid forces of ****** due to bad structures.

📄 Other Changes

• [84f8863] PotentialManagers are registered in one single file __init__.py.
• [dbe7c88] Add a preprocess function check whether all input nodes are valid for an operation.