ipsuite.analysis.model package¶

Submodules¶

ipsuite.analysis.model.dynamics module¶

class ipsuite.analysis.model.dynamics.BoxHeatUp(*args, **kwargs)[source]¶

Bases: ProcessSingleAtom

Perform a heat-up analysis by gradually increasing the temperature of the system and monitoring the measured temperature and energy.

Parameters¶

start_temperaturefloat: The initial temperature (in Kelvin) to start the analysis.
stop_temperaturefloat: The upper bound of the temperature range (in Kelvin).
stepsint: Number of temperature increments between start and stop temperature.
time_stepfloat, optional: Time step of the simulation in femtoseconds. Default is 0.5 fs.
frictionfloat: Langevin friction coefficient.
repeattuple of int, optional: Number of repetitions of the unit cell in each direction. Default is (1, 1, 1).
max_temperaturefloat or None, optional: Maximum allowed temperature before stopping the simulation. If None, set to 1.5 * stop_temperature.
modeltyping.Any: The MLModel node that implements the ‘get_calculator’ method.
model_outspathlib.Path: Output directory for model results.
plotspathlib.Path: Path to save the temperature plot.

Attributes¶

flux_datapd.DataFrame: DataFrame containing measured temperature, total energy, and set temperature at each step.
steps_before_explosionint: Number of steps completed before exceeding max_temperature, or -1 if not exceeded.
frameslist of ase.Atoms: List of atomic configurations at each step.

data: List[ase.Atoms]¶

flux_data: DataFrame = NOT_AVAILABLE¶

friction: float¶

get_atoms() → Atoms[source]¶

max_temperature: float | None = None¶

model: Any¶

model_outs: Path = PosixPath('$nwd$/model')¶

plot_temperature()[source]¶

plots: Path = PosixPath('$nwd$/temperature.png')¶

repeat: tuple[int, int, int] = (1, 1, 1)¶

run()[source]¶

start_temperature: float¶

steps: int¶

stop_temperature: float¶

time_step: float = 0.5¶

class ipsuite.analysis.model.dynamics.BoxScale(*args, **kwargs)[source]¶

Bases: ProcessSingleAtom

Scale all particles and predict energies.

Attributes¶

model: The MLModel node that implements the ‘predict’ method atoms: list[Atoms] to predict properties for start: int, default = None

The initial box scale, default value is the original box size.

stop: float, default = 1.0: The stop value for the generated space of stdev points
num: int, default = 100: The size of the generated space of stdev points

data: List[ase.Atoms]¶

energies: DataFrame = NOT_AVAILABLE¶

mapping: Any | None = None¶

model: MLModel¶

model_outs: Path = PosixPath('$nwd$/model')¶

num: int = 100¶

plot: Path = PosixPath('$nwd$/energy.png')¶

run()[source]¶

start: float = 1¶

stop: float = 2.0¶

class ipsuite.analysis.model.dynamics.MDStability(*args, **kwargs)[source]¶

Bases: IPSNode

Perform NVE molecular dynamics for all supplied atoms using a trained model. Several stability checks can be supplied to judge whether a particular trajectory is stable. If the check fails, the trajectory is terminated. After all trajectories are done, a histogram of the duration of stability is created.

Attributes¶

model: A node which implements the calc property. Typically an MLModel instance. data: list[Atoms] to run MD for for max_steps: Maximum number of steps for each trajectory time_step: MD integration time step initial_temperature: Initial velocities are drawn from a maxwell boltzman distribution. save_last_n: how many configurations before the instability should be saved bins: number of bins in the histogram seed: seed for the MaxwellBoltzmann distribution

bins: int = None¶

checks: list[Node] = None¶

data: list[Atoms]¶

property frames: List[Atoms]¶

get_plots(stable_steps: int) → None[source]¶: Create figures for all available data.

initial_temperature: float = 300¶

max_steps: int¶

model: Any¶

model_outs: Path = PosixPath('$nwd$/model_outs')¶

plots_dir: Path = PosixPath('$nwd$/plots')¶

run() → None[source]¶

save_last_n: int = 1¶

seed: int = 0¶

stable_steps_df: DataFrame = NOT_AVAILABLE¶

time_step: float = 0.5¶

traj_file: Path = PosixPath('$nwd$/structures.h5')¶

class ipsuite.analysis.model.dynamics.RattleAnalysis(*args, **kwargs)[source]¶

Bases: ProcessSingleAtom

Move particles with a given stdev from a starting configuration and predict.

Attributes¶

model: The MLModel node that implements the ‘predict’ method atoms: list[Atoms] to predict properties for logspace: bool, default=True

Increase the stdev of rattle with ‘np.logspace’ instead of ‘np.linspace’

stop: float, default = 1.0: The stop value for the generated space of stdev points
num: int, default = 100: The size of the generated space of stdev points
factor: float, default = 0.001: The ‘np.linspace(0.0, stop, num) * factor’
atom_id: int, default = 0: The atom to pick from self.atoms as a starting point

data: List[ase.Atoms]¶

energies: DataFrame = NOT_AVAILABLE¶

factor: float = 0.001¶

logspace: bool = True¶

model: MLModel¶

model_outs: Path = PosixPath('$nwd$/model')¶

num: int = 100¶

run()[source]¶

seed: int = 1234¶

stop: float = 3.0¶

ipsuite.analysis.model.dynamics.run_stability_nve(atoms: Atoms, time_step: float, max_steps: int, init_temperature: float, checks: list[Check], save_last_n: int, rng: Generator | None = None) → Tuple[int, list[Atoms]][source]¶: Runs an NVE trajectory for a single configuration until either max_steps is reached or one of the checks fails.

ipsuite.analysis.model.dynamics_checks module¶

ipsuite.analysis.model.math module¶

ipsuite.analysis.model.math.comptue_rll(inputs, std, target)[source]¶: Compute relative log likelihood Adapted from https://github.com/bananenpampe/DPOSE

ipsuite.analysis.model.math.compute_intertia_tensor(centered_positions, masses)[source]¶

ipsuite.analysis.model.math.compute_rmse(errors)[source]¶

ipsuite.analysis.model.math.compute_rot_forces(mol, key: str = 'forces')[source]¶

ipsuite.analysis.model.math.compute_trans_forces(mol, key: str = 'forces')[source]¶: Compute translational forces of a molecule.

ipsuite.analysis.model.math.compute_uncertainty_metrics(y_pred, y_std, y_true)[source]¶

ipsuite.analysis.model.math.decompose_stress_tensor(stresses)[source]¶

ipsuite.analysis.model.math.force_decomposition(atom, mapping, key: str = 'forces')[source]¶

ipsuite.analysis.model.math.nll(pred, std, true)[source]¶

ipsuite.analysis.model.math.nlls(pred, std, true)[source]¶

ipsuite.analysis.model.plots module¶

ipsuite.analysis.model.plots.density_scatter(ax, x, y, bins, **kwargs) → None[source]¶: Create a scatter plot colored by 2d histogram density.

Parameters¶

ax: matplotlib.axes.Axes x: np.ndarray y: np.ndarray bins: int kwargs

any kwargs passed to ‘ax.scatter’

Returns¶

References¶

Adapted from https://stackoverflow.com/a/53865762/10504481

ipsuite.analysis.model.plots.gauss(x, *p)[source]¶

ipsuite.analysis.model.plots.get_calibration_figure(error, std, markersize: float = 3.0, datalabel=None, forces=False, figsize: tuple = (10, 7))[source]¶: Log-log plot of errors vs predicted standard deviations with quantiles for a linearly increasing noise level.

ipsuite.analysis.model.plots.get_figure(true, prediction, datalabel: str, xlabel: str, ylabel: str, ymax: float | None = None, figsize: tuple = (10, 7), density=True) → Figure[source]¶: Create a correlation plot for true, prediction values.

Parameters¶

true: the true values prediction: the predicted values datalabel: str, the label for the prediction, e.g. ‘MAE: 0.123 eV’ xlabel: str, the xlabel ylabel: str, the xlabel figsize: tuple, size of the figure

Returns¶

plt.Figure

ipsuite.analysis.model.plots.get_gaussianicity_figure(error_true, error_pred, forces=True)[source]¶: Plots empirical and predicted error distributions. If possible, it also tries to fit a gaussian to the empirical distribution.

ipsuite.analysis.model.plots.get_hist(data, label, xlabel, ylabel) → Tuple[Figure, Axes][source]¶

ipsuite.analysis.model.plots.get_histogram_figure(bin_edges, counts, datalabel: str, xlabel: str, ylabel: str, x_lim: Tuple[float, float] = None, y_lim: Tuple[float, float] = None, logy_scale=True, figsize: tuple = (10, 7)) → Figure[source]¶

Creates a Matplotlib figure based on precomputed bin edges and counts.

Parameters¶

bin_edges: np.array: Edges of the histogram bins.
counts: np.array: Number of occurrences in each bin.
datalabel: str: Labels for the figure legend.
xlabel: str: X-axis label.
ylabel: str: Y-axis label.
x_lim: tuple: X-axis limits.
y_lim: tuple: Y-axis limits.
figsize: tuple: Size of the Matplotlib figure

ipsuite.analysis.model.plots.slice_ensemble_uncertainty(true, pred_ens, slice_start, slice_end)[source]¶

ipsuite.analysis.model.plots.slice_uncertainty(true, pred_mean, pred_std, slice_start, slice_end)[source]¶

ipsuite.analysis.model.predict module¶

class ipsuite.analysis.model.predict.CalibrationMetrics(*args, **kwargs)[source]¶

Bases: ComparePredictions

Analyse the calibration of a models uncertainty estimate. Plots the empirical vs predicted error distribution, a log-log calibration plot and the miscalibration area. Further, various UQ metrics are computed: - Mean absolute calibration error - Root mean square miscalibration error - Miscalibration area - NLL - RLL

For more information checkout the uncertainty toolbox or the following paper: 10.1088/2632-2153/ad594a

Parameters¶

force_dist_slices: List[tuple]: Interval in which to analyse the gassianity of error distributions.

data_file: Path = PosixPath('$nwd$/data.npz')¶

energy: dict = NOT_AVAILABLE¶

force_dist_slices: List[tuple] | None = None¶

forces: dict = NOT_AVAILABLE¶

get_data()[source]¶: Create dict of all data.

get_metrics()[source]¶: Update the metrics.

get_plots(save=False)[source]¶: Create figures for all available data.

plots_dir: Path = PosixPath('$nwd$/plots')¶

run()[source]¶

x: list[ase.Atoms]¶

y: list[ase.Atoms]¶

class ipsuite.analysis.model.predict.ForceAngles(*, name: str | None = None, always_changed: bool = False, x: list[ase.atoms.Atoms], y: list[ase.atoms.Atoms], plot: pathlib.Path = PosixPath('$nwd$/angle.png'), log_plot: pathlib.Path = PosixPath('$nwd$/angle_ylog.png'))[source]¶

Bases: ComparePredictions

angles: dict = NOT_AVAILABLE¶

log_plot: Path = PosixPath('$nwd$/angle_ylog.png')¶

plot: Path = PosixPath('$nwd$/angle.png')¶

run()[source]¶

x: list[ase.Atoms]¶

y: list[ase.Atoms]¶

class ipsuite.analysis.model.predict.ForceDecomposition(*args, **kwargs)[source]¶

Bases: ComparePredictions

Node for decomposing forces in a system of molecular units into translational, rotational and vibrational components.

The implementation follows the method described in https://doi.org/10.26434/chemrxiv-2022-l4tb9

Currently, single atoms and diatomic molecules are simply filtered out. Please raise an issue if you need those cases to be treated correctly in your work.

Attributes¶

wasserstein_distance: float: Compute the wasserstein distance between the distributions of the predicted and true forces for each trans, rot, vib component.

get_histogram()[source]¶

get_metrics()[source]¶: Update the metrics.

get_plots()[source]¶

histogram_plt: Path = PosixPath('$nwd$/histogram.png')¶

rot_force_plt: Path = PosixPath('$nwd$/rot_force.png')¶

rot_forces: dict = NOT_AVAILABLE¶

run()[source]¶

trans_force_plt: Path = PosixPath('$nwd$/trans_force.png')¶

trans_forces: dict = NOT_AVAILABLE¶

vib_force_plt: Path = PosixPath('$nwd$/vib_force.png')¶

vib_forces: dict = NOT_AVAILABLE¶

wasserstein_distance: dict = NOT_AVAILABLE¶

x: list[ase.Atoms]¶

y: list[ase.Atoms]¶

class ipsuite.analysis.model.predict.ForceUncertaintyDecomposition(*args, **kwargs)[source]¶

Bases: ComparePredictions

Node for decomposing force uncertainties in a system of molecular units into translational, rotational and vibrational components.

The implementation follows the method described in https://doi.org/10.26434/chemrxiv-2022-l4tb9

Currently, single atoms and diatomic molecules are simply filtered out. Please raise an issue if you need those cases to be treated correctly in your work.

get_metrics()[source]¶: Update the metrics.

get_plots()[source]¶

plots_dir: Path = PosixPath('$nwd$/plots')¶

rot_forces: dict = NOT_AVAILABLE¶

run()[source]¶

trans_forces: dict = NOT_AVAILABLE¶

vib_forces: dict = NOT_AVAILABLE¶

x: list[ase.Atoms]¶

y: list[ase.Atoms]¶

class ipsuite.analysis.model.predict.Prediction(*args, **kwargs)[source]¶

Bases: ProcessAtoms

Create and Save the predictions from model on atoms.

Attributes¶

model: The MLModel node that implements the ‘predict’ method atoms: list[Atoms] to predict properties for

predictions: list[Atoms] the atoms that have the predicted properties from model

data: list[ase.Atoms]¶

model: MLModel¶

run()[source]¶

class ipsuite.analysis.model.predict.PredictionMetrics(*args, **kwargs)[source]¶

Bases: ComparePredictions

Compare model predictions against reference data with comprehensive metrics.

Computes and visualizes prediction accuracy for energy, forces, and stress with statistical measures including MAE, RMSE, and correlation coefficients. Useful for benchmarking different models and analyzing prediction quality.

Parameters¶

xlist[ase.Atoms]: Reference (ground truth) atomic configurations with calculated properties.
ylist[ase.Atoms]: Model predictions on the same configurations.
figure_ymaxdict[str, float], optional: Y-axis limits for plots by property type (energy, forces, stress).

Attributes¶

energydict: Energy prediction metrics (MAE, RMSE, R², etc.) in meV/atom.
forcesdict: Force prediction metrics (MAE, RMSE, R², etc.) in meV/Å.
stressdict: Stress prediction metrics in eV/Å³.
plots_dirPath: Directory containing generated comparison plots.

Examples¶

>>> medium_model = ips.MACEMPModel(model="medium")
>>> small_model = ips.MACEMPModel(model="small")
>>> with project:
...     data = ips.AddData(file="ethanol.xyz")
...     medium_data = ips.ApplyCalculator(data=data.frames, model=medium_model)
...     small_data = ips.ApplyCalculator(data=data.frames, model=small_model)
...     metrics = ips.PredictionMetrics(x=medium_data.frames, y=small_data.frames)
>>> project.repro()
>>> print(f"Energy MAE: {metrics.energy['mae']:.2f} meV/atom")
Energy MAE: 52.23 meV/atom
>>> print(f"Force MAE: {metrics.forces['mae']:.2f} meV/Å")
Force MAE: 213.22 meV/Å

data_file: Path = PosixPath('$nwd$/data.npz')¶

energy: dict = NOT_AVAILABLE¶

figure_ymax: dict[str, float]¶

forces: dict = NOT_AVAILABLE¶

get_content()[source]¶

get_data()[source]¶: Create dict of all data.

get_metrics()[source]¶: Update the metrics.

get_plots(save=False)[source]¶: Create figures for all available data.

plots_dir: Path = PosixPath('$nwd$/plots')¶

run()[source]¶

stress: dict = NOT_AVAILABLE¶

stress_deviat: dict = NOT_AVAILABLE¶

stress_hydro: dict = NOT_AVAILABLE¶

x: list[ase.Atoms]¶

y: list[ase.Atoms]¶

ipsuite.analysis.model.predict.decompose_force_uncertainty(atom_true, atom_pred)[source]¶

Module contents¶

As in all ML applications, analysing both the dataset and model predictions are of paramount importance. For dataset exploration it is often convenient to visualize the distribution of labels. Most Nodes are concerned with analysing trained models and often compare to a reference calculator. This ranges from simple prediction correlation plots to force decompositions and energy-volume curves.

class ipsuite.analysis.model.BoxHeatUp(*args, **kwargs)[source]¶

Bases: ProcessSingleAtom

Perform a heat-up analysis by gradually increasing the temperature of the system and monitoring the measured temperature and energy.

Parameters¶

start_temperaturefloat: The initial temperature (in Kelvin) to start the analysis.
stop_temperaturefloat: The upper bound of the temperature range (in Kelvin).
stepsint: Number of temperature increments between start and stop temperature.
time_stepfloat, optional: Time step of the simulation in femtoseconds. Default is 0.5 fs.
frictionfloat: Langevin friction coefficient.
repeattuple of int, optional: Number of repetitions of the unit cell in each direction. Default is (1, 1, 1).
max_temperaturefloat or None, optional: Maximum allowed temperature before stopping the simulation. If None, set to 1.5 * stop_temperature.
modeltyping.Any: The MLModel node that implements the ‘get_calculator’ method.
model_outspathlib.Path: Output directory for model results.
plotspathlib.Path: Path to save the temperature plot.

Attributes¶

flux_datapd.DataFrame: DataFrame containing measured temperature, total energy, and set temperature at each step.
steps_before_explosionint: Number of steps completed before exceeding max_temperature, or -1 if not exceeded.
frameslist of ase.Atoms: List of atomic configurations at each step.

data: List[ase.Atoms]¶

flux_data: DataFrame = NOT_AVAILABLE¶

friction: float¶

get_atoms() → Atoms[source]¶

max_temperature: float | None = None¶

model: Any¶

model_outs: Path = PosixPath('$nwd$/model')¶

plot_temperature()[source]¶

plots: Path = PosixPath('$nwd$/temperature.png')¶

repeat: tuple[int, int, int] = (1, 1, 1)¶

run()[source]¶

start_temperature: float¶

steps: int¶

stop_temperature: float¶

time_step: float = 0.5¶

class ipsuite.analysis.model.BoxScale(*args, **kwargs)[source]¶

Bases: ProcessSingleAtom

Scale all particles and predict energies.

Attributes¶

model: The MLModel node that implements the ‘predict’ method atoms: list[Atoms] to predict properties for start: int, default = None

The initial box scale, default value is the original box size.

stop: float, default = 1.0: The stop value for the generated space of stdev points
num: int, default = 100: The size of the generated space of stdev points

data: List[ase.Atoms]¶

energies: DataFrame = NOT_AVAILABLE¶

mapping: Any | None = None¶

model: MLModel¶

model_outs: Path = PosixPath('$nwd$/model')¶

num: int = 100¶

plot: Path = PosixPath('$nwd$/energy.png')¶

run()[source]¶

start: float = 1¶

stop: float = 2.0¶

class ipsuite.analysis.model.CalibrationMetrics(*args, **kwargs)[source]¶

Bases: ComparePredictions

For more information checkout the uncertainty toolbox or the following paper: 10.1088/2632-2153/ad594a

Parameters¶

force_dist_slices: List[tuple]: Interval in which to analyse the gassianity of error distributions.

data_file: Path = PosixPath('$nwd$/data.npz')¶

energy: dict = NOT_AVAILABLE¶

force_dist_slices: List[tuple] | None = None¶

forces: dict = NOT_AVAILABLE¶

get_data()[source]¶: Create dict of all data.

get_metrics()[source]¶: Update the metrics.

get_plots(save=False)[source]¶: Create figures for all available data.

plots_dir: Path = PosixPath('$nwd$/plots')¶

run()[source]¶

x: list[ase.Atoms]¶

y: list[ase.Atoms]¶

class ipsuite.analysis.model.ForceAngles(*, name: str | None = None, always_changed: bool = False, x: list[ase.atoms.Atoms], y: list[ase.atoms.Atoms], plot: pathlib.Path = PosixPath('$nwd$/angle.png'), log_plot: pathlib.Path = PosixPath('$nwd$/angle_ylog.png'))[source]¶

Bases: ComparePredictions

angles: dict = NOT_AVAILABLE¶

log_plot: Path = PosixPath('$nwd$/angle_ylog.png')¶

plot: Path = PosixPath('$nwd$/angle.png')¶

run()[source]¶

x: list[ase.Atoms]¶

y: list[ase.Atoms]¶

class ipsuite.analysis.model.ForceDecomposition(*args, **kwargs)[source]¶

Bases: ComparePredictions

Node for decomposing forces in a system of molecular units into translational, rotational and vibrational components.

The implementation follows the method described in https://doi.org/10.26434/chemrxiv-2022-l4tb9

Currently, single atoms and diatomic molecules are simply filtered out. Please raise an issue if you need those cases to be treated correctly in your work.

Attributes¶

wasserstein_distance: float: Compute the wasserstein distance between the distributions of the predicted and true forces for each trans, rot, vib component.

get_histogram()[source]¶

get_metrics()[source]¶: Update the metrics.

get_plots()[source]¶

histogram_plt: Path = PosixPath('$nwd$/histogram.png')¶

rot_force_plt: Path = PosixPath('$nwd$/rot_force.png')¶

rot_forces: dict = NOT_AVAILABLE¶

run()[source]¶

trans_force_plt: Path = PosixPath('$nwd$/trans_force.png')¶

trans_forces: dict = NOT_AVAILABLE¶

vib_force_plt: Path = PosixPath('$nwd$/vib_force.png')¶

vib_forces: dict = NOT_AVAILABLE¶

wasserstein_distance: dict = NOT_AVAILABLE¶

x: list[ase.Atoms]¶

y: list[ase.Atoms]¶

class ipsuite.analysis.model.ForceUncertaintyDecomposition(*args, **kwargs)[source]¶

Bases: ComparePredictions

Node for decomposing force uncertainties in a system of molecular units into translational, rotational and vibrational components.

The implementation follows the method described in https://doi.org/10.26434/chemrxiv-2022-l4tb9

Currently, single atoms and diatomic molecules are simply filtered out. Please raise an issue if you need those cases to be treated correctly in your work.

get_metrics()[source]¶: Update the metrics.

get_plots()[source]¶

plots_dir: Path = PosixPath('$nwd$/plots')¶

rot_forces: dict = NOT_AVAILABLE¶

run()[source]¶

trans_forces: dict = NOT_AVAILABLE¶

vib_forces: dict = NOT_AVAILABLE¶

x: list[ase.Atoms]¶

y: list[ase.Atoms]¶

class ipsuite.analysis.model.MDStability(*args, **kwargs)[source]¶

Bases: IPSNode

Attributes¶

bins: int = None¶

checks: list[Node] = None¶

data: list[Atoms]¶

property frames: List[Atoms]¶

get_plots(stable_steps: int) → None[source]¶: Create figures for all available data.

initial_temperature: float = 300¶

max_steps: int¶

model: Any¶

model_outs: Path = PosixPath('$nwd$/model_outs')¶

plots_dir: Path = PosixPath('$nwd$/plots')¶

run() → None[source]¶

save_last_n: int = 1¶

seed: int = 0¶

stable_steps_df: DataFrame = NOT_AVAILABLE¶

time_step: float = 0.5¶

traj_file: Path = PosixPath('$nwd$/structures.h5')¶

class ipsuite.analysis.model.Prediction(*args, **kwargs)[source]¶

Bases: ProcessAtoms

Create and Save the predictions from model on atoms.

Attributes¶

model: The MLModel node that implements the ‘predict’ method atoms: list[Atoms] to predict properties for

predictions: list[Atoms] the atoms that have the predicted properties from model

data: list[ase.Atoms]¶

model: MLModel¶

run()[source]¶

class ipsuite.analysis.model.PredictionMetrics(*args, **kwargs)[source]¶

Bases: ComparePredictions

Compare model predictions against reference data with comprehensive metrics.

Parameters¶

xlist[ase.Atoms]: Reference (ground truth) atomic configurations with calculated properties.
ylist[ase.Atoms]: Model predictions on the same configurations.
figure_ymaxdict[str, float], optional: Y-axis limits for plots by property type (energy, forces, stress).

Attributes¶

energydict: Energy prediction metrics (MAE, RMSE, R², etc.) in meV/atom.
forcesdict: Force prediction metrics (MAE, RMSE, R², etc.) in meV/Å.
stressdict: Stress prediction metrics in eV/Å³.
plots_dirPath: Directory containing generated comparison plots.

Examples¶

>>> medium_model = ips.MACEMPModel(model="medium")
>>> small_model = ips.MACEMPModel(model="small")
>>> with project:
...     data = ips.AddData(file="ethanol.xyz")
...     medium_data = ips.ApplyCalculator(data=data.frames, model=medium_model)
...     small_data = ips.ApplyCalculator(data=data.frames, model=small_model)
...     metrics = ips.PredictionMetrics(x=medium_data.frames, y=small_data.frames)
>>> project.repro()
>>> print(f"Energy MAE: {metrics.energy['mae']:.2f} meV/atom")
Energy MAE: 52.23 meV/atom
>>> print(f"Force MAE: {metrics.forces['mae']:.2f} meV/Å")
Force MAE: 213.22 meV/Å

data_file: Path = PosixPath('$nwd$/data.npz')¶

energy: dict = NOT_AVAILABLE¶

figure_ymax: dict[str, float]¶

forces: dict = NOT_AVAILABLE¶

get_content()[source]¶

get_data()[source]¶: Create dict of all data.

get_metrics()[source]¶: Update the metrics.

get_plots(save=False)[source]¶: Create figures for all available data.

plots_dir: Path = PosixPath('$nwd$/plots')¶

run()[source]¶

stress: dict = NOT_AVAILABLE¶

stress_deviat: dict = NOT_AVAILABLE¶

stress_hydro: dict = NOT_AVAILABLE¶

x: list[ase.Atoms]¶

y: list[ase.Atoms]¶

class ipsuite.analysis.model.RattleAnalysis(*args, **kwargs)[source]¶

Bases: ProcessSingleAtom

Move particles with a given stdev from a starting configuration and predict.

Attributes¶

model: The MLModel node that implements the ‘predict’ method atoms: list[Atoms] to predict properties for logspace: bool, default=True

Increase the stdev of rattle with ‘np.logspace’ instead of ‘np.linspace’

stop: float, default = 1.0: The stop value for the generated space of stdev points
num: int, default = 100: The size of the generated space of stdev points
factor: float, default = 0.001: The ‘np.linspace(0.0, stop, num) * factor’
atom_id: int, default = 0: The atom to pick from self.atoms as a starting point

data: List[ase.Atoms]¶

energies: DataFrame = NOT_AVAILABLE¶

factor: float = 0.001¶

logspace: bool = True¶

model: MLModel¶

model_outs: Path = PosixPath('$nwd$/model')¶

num: int = 100¶

run()[source]¶

seed: int = 1234¶

stop: float = 3.0¶

ipsuite.analysis.model package¶

Submodules¶

ipsuite.analysis.model.dynamics module¶

Parameters¶

Attributes¶

Attributes¶

Attributes¶

Attributes¶

ipsuite.analysis.model.dynamics_checks module¶

ipsuite.analysis.model.math module¶

ipsuite.analysis.model.plots module¶

Parameters¶

Returns¶

References¶

Parameters¶

Returns¶

Parameters¶

ipsuite.analysis.model.predict module¶

Parameters¶

Attributes¶

Attributes¶

Parameters¶

Attributes¶

Examples¶

Module contents¶

Parameters¶

Attributes¶

Attributes¶

Parameters¶

Attributes¶

Attributes¶

Attributes¶

Parameters¶

Attributes¶

Examples¶

Attributes¶