Active learning for TSH

Efficient and robust active learning for surface-hopping dynamics: often, you can do surface-hopping dynamics from start to finish within a couple of days on a single GPU! See for details:

• Mikołaj Martyka, Lina Zhang, Fuchun Ge, Yi-Fan Hou, Joanna Jankowska, Mario Barbatti, Pavlo O. Dral. Charting electronic-state manifolds across molecules with multi-state learning and gap-driven dynamics via efficient and robust active learning. npj Comput. Mater. 2025, 11, 132. DOI: 10.1038/s41524-025-01636-z. Preprint on ChemRxiv: https://doi.org/10.26434/chemrxiv-2024-dtc1w.

Note

It is more efficient to do active learning with the pre-trained, universal excited-state OMNI-P2x model rather than doing it from scratch with the MS-ANI model. See the tutorial ATL-NAMD: Active transfer learning for NAMD via fine-tuning of OMNI-P2x.

Zip with tutorial materials including the Jupyter notebook:

In [ ]:

import mlatom as ml 

In [ ]:

# Setting up the reference method. Note: you might want a better method than AIQM1/CIS for NAMD.
# Here, it is used for demo purposes.
aiqm1= ml.models.methods(method="AIQM1")
aiqm1_kwargs = {'nstates':2}

In [ ]:

# Load the molecule for AL. Needs to have frequencies computed.
# It will be used in the normal mode sampling (Wigner sampling)
# you can generate this file with frequency calculations using MLatom
# i.e., ml.simulations.freq(molecule=..., ...) - see the dedicated tutorial
eqmol = ml.molecule.load('eqmol.json',format='json')

In [ ]:

# Initial conditions for ML-NAMD in the AL loop
init_sampler_kwargs = {
        'eqmol':eqmol,
        'initial_temperature':0,
        'number_of_initial_conditions': 16 # defines number of trajectories here
    }
# Arguments for the AL ML-NAMD sampler
# you might want to use the defaults or stricter criteria
# this arguments are chosen to run the tutorial faster
sampler_kwargs = {
        'maximum_propagation_time':10.0, # fs
        'time_step':0.1,                 # fs
        'nstates':2,                     # number of electronic state for TSH
        'initial_state':1,               # initial state for ML-NAMD
        'prevent_back_hop':False,        # prevent back hops?
        'dump_trajectory_interval':10,   # interval to dump trajectory.
        'reduce_memory_usage':True,      # do not keep entire trajectories in memory (dumps and cuts the trajectory as defined in dump_trajectory_interval, if dump_trajectory_interval=None then the trajectory is dumped at the end). If reduce_memory_usage=False then AL will not dump anything but you may run out of memory.
        'min_surfuq_points': 12,         # minimum points for AL-NAMD to be considered converged.
        'ngapmd_trajs':16,               # total number of gapMD trajectories. Default the same as the number of trajectories. If the specified value is greater than the nuber of ML-NAMD trajectories (number_of_initial_conditions), then it is changed to number_of_initial_conditions.
        'max_hopprobuq_points':4,        # maximum number of uncertain hopping probability points, default - 15.
        'prob_uq_threshold':0.1,         # threshold for probability-based sampling.
        'plot_populations':True,         # plot populations each iteration? It can take extra time in longer dynamics.
        'stop_uncertain_md':False,       # stop MD when it is uncertain, otherwise keep propagating (has no impact on sampling, is slower but allows for plotting populations)
        'initcond_sampler_kwargs':init_sampler_kwargs
    }

In [ ]:

#Arguments for the initial points sampling procedure.
init_dataset_sampler_kwargs = {
    'molecule':eqmol,
    'initial_temperature': 0,
    'number_of_initial_conditions': 50
}

In [ ]:

# ML model arguments.
# again, you might want to use the default for gap_weight
ml_model_kwargs = {'nstates':   2,   # number of states in the MS-ANI model
                   'gap_weight':1.0} # weight of the gap loss 

In [ ]:

# Running active learning.
# Some settings better leave default (like nthreads, initial_point_refinement).
# Here they are changed for faster calculations.
ml.al(molecule                 = eqmol,
      reference_method         = aiqm1,          # reference method for labeling
      ml_model                 = 'msani',        # specify the MS-ANI model
      sampler                  = 'lzsh',         # use LZSH for sampling, with the AL-NAMD loop
      initdata_sampler_kwargs=init_dataset_sampler_kwargs,
      initial_points_refinement= 'one-shot',     # use one-shot method for initial points sampling
      initdata_sampler         = 'wigner',       # sample initial data from Wigner
      sampler_kwargs           = sampler_kwargs, 
      refmethod_kwargs         = aiqm1_kwargs,   # arguments of the reference method
      nthreads                 = 2,              # number of threads
      ml_model_kwargs          = ml_model_kwargs,# arguments for setting up the ML model
      new_points               = 50,             # number of new points each iteration
     )

Initial points sampling time: 16.064443078357726 s
Active learning iteration 0
Points to label: 50
New labeled points: 50
0 points are abandoned due to failed calculation
Number of points in the labeled data set: 50
New threshold: 0.02166365133151412
New threshold: 0.021648109401762487
    Number of training points: 50
        Number of subtraining points: 45
        Number of validation points: 5
        Main model
            Subtraining set:
                RMSE of values = 0.001697495306656166
                Correlation coefficient = 0.9976927354489449
                RMSE of gradients = 0.004054396915404656
                Correlation coefficient = 0.9982806152425137
            Validation set:
                RMSE of values = 4.964569583224559e-08
                Correlation coefficient = 0.9999999999969875
                RMSE of gradients = 3.5428167685899343e-08
                Correlation coefficient = 0.9999999999998586
        Auxiliary model
            Subtraining set:
                RMSE of values = 0.0026502502558855
                Correlation coefficient = 0.9610827810297459
            Validation set:
                RMSE of values = 0.01078232560595051
                Correlation coefficient = 0.674217189362713
Trajectory 1 number of steps: 101
Adding molecule from trajectory 1 at time 5.00 fs
Starting gapMD from trajectory 1 at time 4.00 fs
Trajectory 2 number of steps: 101
Adding molecule from trajectory 2 at time 2.70 fs
Starting gapMD from trajectory 2 at time 2.60 fs
Trajectory 3 number of steps: 101
Adding molecule from trajectory 3 at time 0.00 fs
Starting gapMD from trajectory 3 at time 0.00 fs
Trajectory 4 number of steps: 101
Adding molecule from trajectory 4 at time 5.60 fs
Starting gapMD from trajectory 4 at time 1.60 fs
Trajectory 5 number of steps: 101
Adding molecule from trajectory 5 at time 0.00 fs
Starting gapMD from trajectory 5 at time 0.00 fs
Trajectory 6 number of steps: 101
Adding molecule from trajectory 6 at time 6.10 fs
Starting gapMD from trajectory 6 at time 4.50 fs
Trajectory 7 number of steps: 101
Adding molecule from trajectory 7 at time 5.20 fs
Starting gapMD from trajectory 7 at time 3.80 fs
Trajectory 8 number of steps: 101
Adding molecule from trajectory 8 at time 4.00 fs
Starting gapMD from trajectory 8 at time 1.40 fs
Trajectory 9 number of steps: 101
Adding molecule from trajectory 9 at time 1.10 fs
Starting gapMD from trajectory 9 at time 0.20 fs
Trajectory 10 number of steps: 101
Adding molecule from trajectory 10 at time 0.00 fs
Starting gapMD from trajectory 10 at time 0.00 fs
Trajectory 11 number of steps: 101
Adding molecule from trajectory 11 at time 5.50 fs
Starting gapMD from trajectory 11 at time 3.80 fs
Trajectory 12 number of steps: 101
Adding molecule from trajectory 12 at time 4.90 fs
Starting gapMD from trajectory 12 at time 0.50 fs
Trajectory 13 number of steps: 101
Adding molecule from trajectory 13 at time 0.00 fs
Starting gapMD from trajectory 13 at time 0.00 fs
Trajectory 14 number of steps: 101
Adding molecule from trajectory 14 at time 0.00 fs
Starting gapMD from trajectory 14 at time 0.00 fs
Trajectory 15 number of steps: 101
Adding molecule from trajectory 15 at time 5.30 fs
Starting gapMD from trajectory 15 at time 2.00 fs
Trajectory 16 number of steps: 101
Adding molecule from trajectory 16 at time 4.70 fs
Starting gapMD from trajectory 16 at time 2.70 fs
The surface stopping times are:
5.0
2.7
0.0
5.6000000000000005
0.0
6.1000000000000005
5.2
4.0
1.1
0.0
5.5
4.9
0.0
0.0
5.300000000000001
4.7
The model is 0.0 % converged
Running gapMD trajs on lower surfaces
Trajectory 1 number of steps: 20
Adding molecule from trajectory 1 at time 1.00 fs
Trajectory 2 number of steps: 20
Adding molecule from trajectory 2 at time 1.00 fs
Trajectory 3 number of steps: 20
Adding molecule from trajectory 3 at time 1.50 fs
Trajectory 4 number of steps: 20
Adding molecule from trajectory 4 at time 1.00 fs
Trajectory 5 number of steps: 20
Adding molecule from trajectory 5 at time 1.50 fs
Trajectory 6 number of steps: 20
Adding molecule from trajectory 6 at time 1.50 fs
Trajectory 7 number of steps: 20
Adding molecule from trajectory 7 at time 1.50 fs
Trajectory 8 number of steps: 20
Adding molecule from trajectory 8 at time 1.00 fs
Running gapMD trajs on upper surfaces
Trajectory 1 number of steps: 20
Adding molecule from trajectory 1 at time 0.00 fs
Trajectory 2 number of steps: 20
Adding molecule from trajectory 2 at time 4.50 fs
Trajectory 3 number of steps: 20
Adding molecule from trajectory 3 at time 2.00 fs
Trajectory 4 number of steps: 20
Adding molecule from trajectory 4 at time 1.00 fs
Trajectory 5 number of steps: 20
Adding molecule from trajectory 5 at time 2.00 fs
Trajectory 6 number of steps: 20
Adding molecule from trajectory 6 at time 1.50 fs
Trajectory 7 number of steps: 20
Adding molecule from trajectory 7 at time 1.50 fs
Trajectory 8 number of steps: 20
Adding molecule from trajectory 8 at time 0.50 fs
32
Trajectory 1 number of steps: 101
Adding molecule from trajectory 1 at time 0.00 fs
Starting gapMD from trajectory 1 at time 0.00 fs
Trajectory 2 number of steps: 101
Adding molecule from trajectory 2 at time 6.40 fs
Starting gapMD from trajectory 2 at time 3.70 fs
Trajectory 3 number of steps: 101
Adding molecule from trajectory 3 at time 6.60 fs
Starting gapMD from trajectory 3 at time 0.70 fs
Trajectory 4 number of steps: 101
Adding molecule from trajectory 4 at time 4.50 fs
Starting gapMD from trajectory 4 at time 3.60 fs
Trajectory 5 number of steps: 101
Adding molecule from trajectory 5 at time 0.00 fs
Starting gapMD from trajectory 5 at time 0.00 fs
Trajectory 6 number of steps: 101
Adding molecule from trajectory 6 at time 2.80 fs
Starting gapMD from trajectory 6 at time 2.80 fs
Trajectory 7 number of steps: 101
Adding molecule from trajectory 7 at time 1.90 fs
Starting gapMD from trajectory 7 at time 0.70 fs
Trajectory 8 number of steps: 101
Adding molecule from trajectory 8 at time 4.80 fs
Starting gapMD from trajectory 8 at time 4.40 fs
Trajectory 9 number of steps: 101
Adding molecule from trajectory 9 at time 4.10 fs
Starting gapMD from trajectory 9 at time 1.80 fs
Trajectory 10 number of steps: 101
Adding molecule from trajectory 10 at time 0.00 fs
Starting gapMD from trajectory 10 at time 0.00 fs
Trajectory 11 number of steps: 101
Adding molecule from trajectory 11 at time 2.10 fs
Starting gapMD from trajectory 11 at time 0.10 fs
Trajectory 12 number of steps: 101
Adding molecule from trajectory 12 at time 0.00 fs
Starting gapMD from trajectory 12 at time 0.00 fs
Trajectory 13 number of steps: 101
Adding molecule from trajectory 13 at time 4.80 fs
Starting gapMD from trajectory 13 at time 0.70 fs
Trajectory 14 number of steps: 101
Adding molecule from trajectory 14 at time 6.60 fs
Starting gapMD from trajectory 14 at time 6.50 fs
Trajectory 15 number of steps: 101
Adding molecule from trajectory 15 at time 5.20 fs
Starting gapMD from trajectory 15 at time 1.00 fs
Trajectory 16 number of steps: 101
Adding molecule from trajectory 16 at time 0.00 fs
Starting gapMD from trajectory 16 at time 0.00 fs
The surface stopping times are:
0.0
6.4
6.6000000000000005
4.5
0.0
2.8000000000000003
1.9000000000000001
4.800000000000001
4.1000000000000005
0.0
2.1
0.0
4.800000000000001
6.6000000000000005
5.2
0.0
The model is 0.0 % converged
Running gapMD trajs on lower surfaces
Trajectory 1 number of steps: 20
Adding molecule from trajectory 1 at time 0.50 fs
Trajectory 2 number of steps: 20
Adding molecule from trajectory 2 at time 0.00 fs
Trajectory 3 number of steps: 20
Adding molecule from trajectory 3 at time 0.00 fs
Trajectory 4 number of steps: 20
Adding molecule from trajectory 4 at time 3.00 fs
Trajectory 5 number of steps: 20
Adding molecule from trajectory 5 at time 1.50 fs
Trajectory 6 number of steps: 20
Adding molecule from trajectory 6 at time 1.00 fs
Trajectory 7 number of steps: 20
Adding molecule from trajectory 7 at time 0.50 fs
Trajectory 8 number of steps: 20
Adding molecule from trajectory 8 at time 0.00 fs
Running gapMD trajs on upper surfaces
Trajectory 1 number of steps: 20
Adding molecule from trajectory 1 at time 2.50 fs
Trajectory 2 number of steps: 20
Adding molecule from trajectory 2 at time 0.00 fs
Trajectory 3 number of steps: 20
Adding molecule from trajectory 3 at time 0.00 fs
Trajectory 4 number of steps: 20
Adding molecule from trajectory 4 at time 0.50 fs
Trajectory 5 number of steps: 20
Adding molecule from trajectory 5 at time 0.00 fs
Trajectory 6 number of steps: 20
Adding molecule from trajectory 6 at time 2.50 fs
Trajectory 7 number of steps: 20
Adding molecule from trajectory 7 at time 0.00 fs
Trajectory 8 number of steps: 20
Adding molecule from trajectory 8 at time 6.00 fs
32
Iteration 0 takes 565.5004676911049 s
  Labeling time: 15.793768778908998 s
  Training time: 420.94353390298784 s
  Sampling time: 128.76316500920802 s




Active learning iteration 1
Points to label: 50
New labeled points: 50
0 points are abandoned due to failed calculation
Number of points in the labeled data set: 100
Current threshold for state 0: 0.02166365133151412
Current threshold for state 1: 0.021648109401762487
    Number of training points: 100
        Number of subtraining points: 90
        Number of validation points: 10
        Main model
            Subtraining set:
                RMSE of values = 0.0009359529276005658
                Correlation coefficient = 0.9994262114468375
                RMSE of gradients = 0.003836567826145676
                Correlation coefficient = 0.9989704130475721
            Validation set:
                RMSE of values = 0.0019189207731975563
                Correlation coefficient = 0.9977948236666376
                RMSE of gradients = 0.0073077039742743466
                Correlation coefficient = 0.9948677670569402
        Auxiliary model
            Subtraining set:
                RMSE of values = 0.002208234503521621
                Correlation coefficient = 0.9948439852171983
            Validation set:
                RMSE of values = 0.004628894208882872
                Correlation coefficient = 0.9825574728467348
Trajectory 1 number of steps: 101
Adding molecule from trajectory 1 at time 7.60 fs
Starting gapMD from trajectory 1 at time 4.50 fs
Trajectory 2 number of steps: 101
Adding molecule from trajectory 2 at time 0.00 fs
Starting gapMD from trajectory 2 at time 0.00 fs
Trajectory 3 number of steps: 101
Starting gapMD from trajectory 3 at time 1.90 fs
Trajectory 4 number of steps: 101
Adding molecule from trajectory 4 at time 7.90 fs
Starting gapMD from trajectory 4 at time 2.50 fs
Trajectory 5 number of steps: 101
Adding molecule from trajectory 5 at time 7.20 fs
Starting gapMD from trajectory 5 at time 0.20 fs
Trajectory 6 number of steps: 101
Adding molecule from trajectory 6 at time 7.30 fs
Starting gapMD from trajectory 6 at time 0.40 fs
Trajectory 7 number of steps: 101
Adding molecule from trajectory 7 at time 1.30 fs
Starting gapMD from trajectory 7 at time 0.10 fs
Trajectory 8 number of steps: 101
Starting gapMD from trajectory 8 at time 4.80 fs
Trajectory 9 number of steps: 101
Adding molecule from trajectory 9 at time 8.30 fs
Starting gapMD from trajectory 9 at time 2.90 fs
Trajectory 10 number of steps: 101
Adding molecule from trajectory 10 at time 6.30 fs
Starting gapMD from trajectory 10 at time 1.30 fs
Trajectory 11 number of steps: 101
Adding molecule from trajectory 11 at time 8.30 fs
Starting gapMD from trajectory 11 at time 4.30 fs
Trajectory 12 number of steps: 101
Starting gapMD from trajectory 12 at time 4.30 fs
Trajectory 13 number of steps: 101
Adding molecule from trajectory 13 at time 6.30 fs
Starting gapMD from trajectory 13 at time 6.00 fs
Trajectory 14 number of steps: 101
Adding molecule from trajectory 14 at time 8.90 fs
Starting gapMD from trajectory 14 at time 2.90 fs
Trajectory 15 number of steps: 101
Adding molecule from trajectory 15 at time 6.00 fs
Starting gapMD from trajectory 15 at time 2.10 fs
Trajectory 16 number of steps: 101
Adding molecule from trajectory 16 at time 7.30 fs
Starting gapMD from trajectory 16 at time 1.50 fs
The surface stopping times are:
7.6000000000000005
0.0
7.9
7.2
7.300000000000001
1.3
8.3
6.300000000000001
8.3
6.300000000000001
8.9
6.0
7.300000000000001
The model is 18.75 % converged
Running gapMD trajs on lower surfaces
Trajectory 1 number of steps: 20
Adding molecule from trajectory 1 at time 2.50 fs
Trajectory 2 number of steps: 20
Adding molecule from trajectory 2 at time 4.00 fs
Trajectory 3 number of steps: 20
Adding molecule from trajectory 3 at time 2.00 fs
Trajectory 4 number of steps: 20
Adding molecule from trajectory 4 at time 4.00 fs
Trajectory 5 number of steps: 20
Adding molecule from trajectory 5 at time 4.00 fs
Trajectory 6 number of steps: 20
Adding molecule from trajectory 6 at time 4.50 fs
Trajectory 7 number of steps: 20
Adding molecule from trajectory 7 at time 5.50 fs
Trajectory 8 number of steps: 20
Adding molecule from trajectory 8 at time 1.00 fs
Running gapMD trajs on upper surfaces
Trajectory 1 number of steps: 20
Adding molecule from trajectory 1 at time 4.50 fs
Trajectory 2 number of steps: 20
Adding molecule from trajectory 2 at time 4.00 fs
Trajectory 3 number of steps: 20
Adding molecule from trajectory 3 at time 3.00 fs
Trajectory 4 number of steps: 20
Adding molecule from trajectory 4 at time 1.50 fs
Trajectory 5 number of steps: 20
Adding molecule from trajectory 5 at time 6.00 fs
Trajectory 6 number of steps: 20
Adding molecule from trajectory 6 at time 6.50 fs
Trajectory 7 number of steps: 20
Adding molecule from trajectory 7 at time 4.00 fs
Trajectory 8 number of steps: 20
Adding molecule from trajectory 8 at time 3.00 fs
29
Trajectory 1 number of steps: 101
Adding molecule from trajectory 1 at time 7.70 fs
Starting gapMD from trajectory 1 at time 2.80 fs
Trajectory 2 number of steps: 101
Adding molecule from trajectory 2 at time 8.00 fs
Starting gapMD from trajectory 2 at time 2.70 fs
Trajectory 3 number of steps: 101
Adding molecule from trajectory 3 at time 0.10 fs
Starting gapMD from trajectory 3 at time 0.00 fs
Trajectory 4 number of steps: 101
Adding molecule from trajectory 4 at time 6.70 fs
Starting gapMD from trajectory 4 at time 5.80 fs
Trajectory 5 number of steps: 101
Adding molecule from trajectory 5 at time 6.60 fs
Starting gapMD from trajectory 5 at time 0.90 fs
Trajectory 6 number of steps: 101
Adding molecule from trajectory 6 at time 7.10 fs
Starting gapMD from trajectory 6 at time 6.50 fs
Trajectory 7 number of steps: 101
Starting gapMD from trajectory 7 at time 2.90 fs
Trajectory 8 number of steps: 101
Adding molecule from trajectory 8 at time 8.50 fs
Starting gapMD from trajectory 8 at time 8.10 fs
Trajectory 9 number of steps: 101
Adding molecule from trajectory 9 at time 0.00 fs
Starting gapMD from trajectory 9 at time 0.00 fs
Trajectory 10 number of steps: 101
Adding molecule from trajectory 10 at time 9.50 fs
Starting gapMD from trajectory 10 at time 7.20 fs
Trajectory 11 number of steps: 101
Adding molecule from trajectory 11 at time 9.50 fs
Starting gapMD from trajectory 11 at time 1.50 fs
Trajectory 12 number of steps: 101
Adding molecule from trajectory 12 at time 0.00 fs
Starting gapMD from trajectory 12 at time 0.00 fs
Trajectory 13 number of steps: 101
Adding molecule from trajectory 13 at time 6.50 fs
Starting gapMD from trajectory 13 at time 0.10 fs
Trajectory 14 number of steps: 101
Adding molecule from trajectory 14 at time 9.10 fs
Starting gapMD from trajectory 14 at time 0.40 fs
Trajectory 15 number of steps: 101
Adding molecule from trajectory 15 at time 8.00 fs
Starting gapMD from trajectory 15 at time 4.90 fs
Trajectory 16 number of steps: 101
Adding molecule from trajectory 16 at time 0.00 fs
Starting gapMD from trajectory 16 at time 0.00 fs
The surface stopping times are:
7.7
8.0
0.1
6.7
6.6000000000000005
7.1000000000000005
8.5
0.0
9.5
9.5
0.0
6.5
9.1
8.0
0.0
The model is 6.25 % converged
Running gapMD trajs on lower surfaces
Trajectory 1 number of steps: 20
Adding molecule from trajectory 1 at time 7.00 fs
Trajectory 2 number of steps: 20
Adding molecule from trajectory 2 at time 0.00 fs
Trajectory 3 number of steps: 20
Adding molecule from trajectory 3 at time 2.50 fs
Trajectory 4 number of steps: 20
Adding molecule from trajectory 4 at time 0.50 fs
Trajectory 5 number of steps: 20
Adding molecule from trajectory 5 at time 0.50 fs
Trajectory 6 number of steps: 20
Adding molecule from trajectory 6 at time 2.50 fs
Trajectory 7 number of steps: 20
Adding molecule from trajectory 7 at time 6.00 fs
Trajectory 8 number of steps: 20
Adding molecule from trajectory 8 at time 1.00 fs
Running gapMD trajs on upper surfaces
Trajectory 1 number of steps: 20
Adding molecule from trajectory 1 at time 4.50 fs
Trajectory 2 number of steps: 20
Adding molecule from trajectory 2 at time 1.00 fs
Trajectory 3 number of steps: 20
Adding molecule from trajectory 3 at time 3.00 fs
Trajectory 4 number of steps: 20
Adding molecule from trajectory 4 at time 2.50 fs
Trajectory 5 number of steps: 20
Adding molecule from trajectory 5 at time 3.00 fs
Trajectory 6 number of steps: 20
Adding molecule from trajectory 6 at time 1.50 fs
Trajectory 7 number of steps: 20
Adding molecule from trajectory 7 at time 6.00 fs
Trajectory 8 number of steps: 20
Adding molecule from trajectory 8 at time 0.00 fs
31
Iteration 1 takes 683.7381067443639 s
  Labeling time: 12.741916863247752 s
  Training time: 540.8486425001174 s
  Sampling time: 130.14754738099873 s




Active learning iteration 2
Points to label: 50
New labeled points: 50
0 points are abandoned due to failed calculation
Number of points in the labeled data set: 150
Current threshold for state 0: 0.02166365133151412
Current threshold for state 1: 0.021648109401762487
    Number of training points: 150
        Number of subtraining points: 135
        Number of validation points: 15
        Main model
            Subtraining set:
                RMSE of values = 0.0005569757242605153
                Correlation coefficient = 0.999893743202432
                RMSE of gradients = 0.002572013976181595
                Correlation coefficient = 0.9996511422987345
            Validation set:
                RMSE of values = 0.0007755574094383158
                Correlation coefficient = 0.9997931428803224
                RMSE of gradients = 0.004419000198206317
                Correlation coefficient = 0.9989716001762989
        Auxiliary model
            Subtraining set:
                RMSE of values = 0.0019819420967923454
                Correlation coefficient = 0.99851033686998
            Validation set:
                RMSE of values = 0.005016435923752971
                Correlation coefficient = 0.9912751692411135
Trajectory 1 number of steps: 101
Starting gapMD from trajectory 1 at time 0.90 fs
Trajectory 2 number of steps: 101
Adding molecule from trajectory 2 at time 9.60 fs
Starting gapMD from trajectory 2 at time 2.00 fs
probUQ diverged
probUQ diverged
Trajectory 3 number of steps: 101
Starting gapMD from trajectory 3 at time 1.00 fs
Trajectory 4 number of steps: 101
Adding molecule from trajectory 4 at time 8.90 fs
Starting gapMD from trajectory 4 at time 1.40 fs
Trajectory 5 number of steps: 101
Adding molecule from trajectory 5 at time 9.90 fs
Starting gapMD from trajectory 5 at time 5.60 fs
Trajectory 6 number of steps: 101
Starting gapMD from trajectory 6 at time 4.30 fs
Trajectory 7 number of steps: 101
Starting gapMD from trajectory 7 at time 8.70 fs
Trajectory 8 number of steps: 101
Starting gapMD from trajectory 8 at time 5.70 fs
Trajectory 9 number of steps: 101
Starting gapMD from trajectory 9 at time 4.40 fs
Trajectory 10 number of steps: 101
Starting gapMD from trajectory 10 at time 5.00 fs
Trajectory 11 number of steps: 101
Starting gapMD from trajectory 11 at time 6.20 fs
Trajectory 12 number of steps: 101
Starting gapMD from trajectory 12 at time 5.80 fs
Trajectory 13 number of steps: 101
Starting gapMD from trajectory 13 at time 3.90 fs
Trajectory 14 number of steps: 101
Adding molecule from trajectory 14 at time 9.10 fs
Starting gapMD from trajectory 14 at time 2.30 fs
Trajectory 15 number of steps: 101
Starting gapMD from trajectory 15 at time 8.30 fs
Trajectory 16 number of steps: 101
Adding molecule from trajectory 16 at time 8.00 fs
Starting gapMD from trajectory 16 at time 6.80 fs
The surface stopping times are:
9.600000000000001
8.9
9.9
9.1
8.0
The model is 68.75 % converged
12
5
The ML-NAMD trajectories are converged, AL-NAMD converged.
0
Number of points to be labeled is less than 5, active learning converged
Iteration 2 takes 1201.5959568759426 s
  Labeling time: 13.451935027725995 s
  Training time: 1136.9589645522647 s
  Sampling time: 51.18505729595199 s

Out[ ]:

<mlatom_alnamd.mlatom.al.al at 0x2b38020aa310>

<Figure size 1500x1200 with 0 Axes>

<Figure size 1500x1200 with 0 Axes>

<Figure size 1500x1200 with 0 Axes>

<Figure size 1500x1200 with 0 Axes>

<Figure size 1500x1200 with 0 Axes>

<Figure size 1500x1200 with 0 Axes>

Results above do not show any population as the settings were not strict enough to get to any hoppings and the trajectories are not propagated long enough