"""Functions to predict properties related to hysteresis loops."""
from __future__ import annotations
from pathlib import Path
from typing import TYPE_CHECKING
if TYPE_CHECKING:
import astropy.units
import mammos_entity
import numpy
import mammos_analysis
import mammos_entity as me
import mammos_units as u
import numpy as np
import onnxruntime as ort
MODEL_DIR = Path(__file__).parent
MODELS = {
"classifier": MODEL_DIR / "classifier_cube50_singlegrain_random_forest_v0.1.onnx",
"soft": MODEL_DIR / "soft_cube50_singlegrain_random_forest_v0.1.onnx",
"hard": MODEL_DIR / "hard_cube50_singlegrain_random_forest_v0.1.onnx",
}
_SESSION_OPTIONS = ort.SessionOptions()
_SESSION_OPTIONS.log_severity_level = 3
[docs]
def is_hard_magnet_from_Ms_A_K(
Ms: mammos_entity.Entity | astropy.units.Quantity | numpy.typing.ArrayLike,
A: mammos_entity.Entity | astropy.units.Quantity | numpy.typing.ArrayLike,
K1: mammos_entity.Entity | astropy.units.Quantity | numpy.typing.ArrayLike,
model: str = "cube50_singlegrain_random_forest_v0.1",
) -> bool | np.ndarray:
"""Classify material as soft or hard magnetic from micromagnetic parameters.
This function classifies a magnetic material as either soft or hard magnetic
based on its micromagnetic parameters spontaneous magnetization Ms, exchange
stiffness constant A and uniaxial anisotropy constant K1.
The shape of the input parameters needs to be the same. If single values are
provided, a single classification is returned. If arrays are provided, a
numpy array with the same shape is returned.
The following models are available for the prediction:
- ``cube50_singlegrain_random_forest_v0.1``: Random forest model trained on
simulated data for single grain cubic particles with 50 nm edge length with
the external field applied parallel to the anisotropy axis. These are both
aligned along an edge of the cube. Further details on the training data
and model can be found in the
`model repository <https://github.com/MaMMoS-project/ML-models/tree/main/beyond-stoner-wohlfarth/single-grain-easy-axis-model>`_.
Args:
Ms: Spontaneous magnetization.
A: Exchange stiffness constant.
K1: Uniaxial anisotropy constant.
model: AI model used for the classification
Returns:
Classification as False (soft) or True (hard).
Returns a boolean for scalar inputs, or a numpy array
with the same shape as the input for array inputs.
Examples:
>>> import mammos_ai
>>> import mammos_entity as me
>>> mammos_ai.is_hard_magnet_from_Ms_A_K(me.Ms(1e6), me.A(1e-12), me.Ku(1e6))
True
"""
Ms = me.Ms(Ms, unit=u.A / u.m)
A = me.A(A, unit=u.J / u.m)
K1 = me.Ku(K1, unit=u.J / u.m**3)
Ms_arr = np.atleast_1d(Ms.value)
A_arr = np.atleast_1d(A.value)
K1_arr = np.atleast_1d(K1.value)
if not (Ms_arr.shape == A_arr.shape == K1_arr.shape):
raise ValueError(
f"Input arrays must have the same shape. Shapes are Ms: {Ms_arr.shape}, "
f"A: {A_arr.shape}, Ku: {K1_arr.shape}"
)
original_shape = Ms_arr.shape
is_scalar = Ms_arr.ndim == 0 or (Ms_arr.ndim == 1 and Ms_arr.size == 1)
match model:
case "cube50_singlegrain_random_forest_v0.1":
classifier_path = MODELS["classifier"]
case _:
raise ValueError(f"Unknown model {model}")
session = ort.InferenceSession(classifier_path, _SESSION_OPTIONS)
X = np.column_stack([Ms_arr.ravel(), A_arr.ravel(), K1_arr.ravel()]).astype(
np.float32
)
# input name obtained from model expects a shape
# (n_samples, 3) containing [Ms, A, K1], returns 1D
# array with shape (n_samples,) containing class labels
# (0=soft magnetic, 1=hard magnetic)
results = session.run(None, {session.get_inputs()[0].name: X})[0]
labels = np.where(results == 0, False, True)
if is_scalar:
return labels.item()
else:
return labels.reshape(original_shape)
def _predict_cube50_singlegrain_random_forest_v0_1(
Ms, A, K1, original_shape, is_scalar
):
Ms_arr = np.atleast_1d(Ms.value)
A_arr = np.atleast_1d(A.value)
K1_arr = np.atleast_1d(K1.value)
# 1. Determine class
mat_class = is_hard_magnet_from_Ms_A_K(
Ms, A, K1, model="cube50_singlegrain_random_forest_v0.1"
)
# 2. Preprocess
X_log = np.log1p(
np.column_stack([Ms_arr.ravel(), A_arr.ravel(), K1_arr.ravel()]).astype(
np.float32
)
)
y_log = np.empty((X_log.shape[0], 3), dtype=np.float32)
classes = np.atleast_1d(mat_class).ravel()
for cls in [False, True]:
mask = classes == cls
if np.any(mask):
# 3. Load regression model
model_key = "hard" if cls else "soft"
session = ort.InferenceSession(MODELS[model_key], _SESSION_OPTIONS)
X_subset = X_log[mask]
# 4. Predict: input name obtained from model expects a shape
# (n_samples_in_class, 3) containing [Ms, A, K1], returns 2D
# array with shape (n_samples_in_class, 3) containing
# [Hc, Mr, BHmax] predictions
res = session.run(None, {session.get_inputs()[0].name: X_subset})[0]
y_log[mask] = res
# 5. Postprocess
y = np.expm1(y_log)
# Reshape output as y.shape = (n_samples_in_class, 3)
return y[0] if is_scalar else y.reshape(original_shape + (3,))
[docs]
def Hc_Mr_BHmax_from_Ms_A_K(
Ms: mammos_entity.Entity | astropy.units.Quantity | numpy.typing.ArrayLike,
A: mammos_entity.Entity | astropy.units.Quantity | numpy.typing.ArrayLike,
K1: mammos_entity.Entity | astropy.units.Quantity | numpy.typing.ArrayLike,
model: str = "cube50_singlegrain_random_forest_v0.1",
) -> mammos_analysis.hysteresis.ExtrinsicProperties:
"""Predict Hc, Mr and BHmax from micromagnetic properties Ms, A and K1.
This function predicts extrinsic properties coercive field Hc, remanent
magnetization Mr and maximum energy product BHmax given a set of micromagnetic
material parameters.
The following models are available for the prediction:
- ``cube50_singlegrain_random_forest_v0.1``: Random forest model trained on
simulated data for single grain cubic particles with 50 nm edge length with
the external field applied parallel to the anisotropy axis. These are both
aligned along an edge of the cube. Further details on the training data
and model can be found in the
`model repository <https://github.com/MaMMoS-project/ML-models/tree/main/beyond-stoner-wohlfarth/single-grain-easy-axis-model>`_.
Args:
Ms: Spontaneous magnetization.
A: Exchange stiffness constant.
K1: Uniaxial anisotropy constant.
model: AI model used for the prediction
Returns:
An object containing extrinsic properties Hc, Mr, BHmax
Examples:
>>> import mammos_ai
>>> import mammos_entity as me
>>> mammos_ai.Hc_Mr_BHmax_from_Ms_A_K(me.Ms(1e6), me.A(1e-12), me.Ku(1e6))
ExtrinsicProperties(Hc=..., Mr=..., BHmax=...)
"""
Ms = me.Ms(Ms, unit=u.A / u.m)
A = me.A(A, unit=u.J / u.m)
K1 = me.Ku(K1, unit=u.J / u.m**3)
Ms_arr = np.atleast_1d(Ms.value)
A_arr = np.atleast_1d(A.value)
K1_arr = np.atleast_1d(K1.value)
if not (Ms_arr.shape == A_arr.shape == K1_arr.shape):
raise ValueError(
f"Input arrays must have the same shape. Shapes are Ms: {Ms_arr.shape}, "
f"A: {A_arr.shape}, Ku: {K1_arr.shape}"
)
original_shape = Ms_arr.shape
is_scalar = Ms_arr.ndim == 0 or (Ms_arr.ndim == 1 and Ms_arr.size == 1)
match model:
case "cube50_singlegrain_random_forest_v0.1":
y = _predict_cube50_singlegrain_random_forest_v0_1(
Ms, A, K1, original_shape, is_scalar
)
case _:
raise ValueError(f"Unknown model {model}")
if is_scalar:
Hc_val = y[0]
Mr_val = y[1]
BHmax_val = y[2]
else:
Hc_val = y[..., 0]
Mr_val = y[..., 1]
BHmax_val = y[..., 2]
Hc = me.Hc(Hc_val, "A/m")
Mr = me.Mr(Mr_val, "A/m")
BHmax = me.BHmax(BHmax_val, "J/m3")
return mammos_analysis.hysteresis.ExtrinsicProperties(Hc=Hc, Mr=Mr, BHmax=BHmax)
