dustapprox.models package

dustapprox.models package#

Submodules#

dustapprox.models.basemodel module#

Base class for deriving various kinds of models.

class BaseModel[source]#

Bases: object

A model object other approximations derive from

meta#

meta information about the model

Type:: dict

__init__(**kwargs)[source]#

fit(*args, **kwargs)[source]#

Fit the model to the data

Parameters:

*args – positional arguments
**kwargs – keyword arguments

predict(*args, **kwargs)[source]#

Predict the extinction in a given passband.

Parameters:

*args – positional arguments
**kwargs – keyword arguments

to_pandas()[source]#

Export the model to a pandas DataFrame.

Returns:: The model as a pandas DataFrame.
Return type:: pd.DataFrame

dustapprox.models.polynomial module#

Polynomial approximation of extinction effect per passband.

In this library, we provide tools and pre-computed models to obtain the extinction coefficient \(k_x = A_x / A_0\) for various passbands.

Extinction coefficients depend primarily on the source spectral energy distribution and on the extinction itself (e.g., Gordon et al., 2016; Jordi et al., 2010) but also other parameters such as \(\log g\), \([\alpha/Fe]\), and \([Fe/H]\).

We define the extinction coefficient \(k_x\) as

\[k_x = A_x / A_0,\]

with \(A_0\) the extinction parameter at \(550 nm\), and \(x\) the passband.

We use a L1-regularized regression model (Lasso-Lars model) using BIC or AIC for model/complexity selection.

The optimization objective for Lasso is:

\[\frac{1}{2 n_{samples}} \cdot \|y - X \cdot w\|^2_2 + \alpha \times \|w\|_1\]

AIC is the Akaike information criterion and BIC is the Bayes Information criterion. AIC or BIC are useful criteria to select the value of the regularization parameter \(\alpha\) by making a trade-off between the goodness-of-fit and the complexity of the model.

This allows is to follow the principle of parsimony (aka Occam’s razor): a good model should explain well the data while being simple.

(Source code)

class PolynomialModel[source]#

Bases: BaseModel

A polynomial model object

meta#

meta information about the model

Type:: dict

transformer_#

polynomial transformer

Type:: PolynomialFeatures

coeffs_#

coefficients of the regression on the polynomial expended features

Type:: pd.Series

__init__(**kwargs)[source]#

property degree_: int | None#: Degree of the polynomial transformation

property feature_names: List[str] | None#: Input feature dimensions of the model

fit(df, features=None, label='Ax', degree=3, interaction_only=False)[source]#

Parameters:

df (DataFrame) – DataFrame with the passband grid data.
features (Sequence[str]) – input features from df (note: if used, teff will be normalized to teff/5040)
label (str) – which field contains the label values
degree (int) – The degree of the polynomial model.
interaction_only (bool) – If True, only the interaction terms are used.
input_parameters (Sequence[str]) – The input parameters to use. If None, ‘teff logg feh A0 alpha’ parameters are used.

classmethod from_file(filename, passband)[source]#

Restore a model from a file

Parameters:

filename (str) –
the ECSV filepath containing the model definition

The file should contain the various parameters associated with the model in its metadata.
passband (str) – name of the model to load (passband column in the ecsv file)

Returns:

model – model object

Return type:

PolynomialModel

get_transformed_feature_names()[source]#

get the feature names of the internal transformation

Return type:: ndarray[tuple[Any, …], dtype[Any]] | None

property name: str | None#: Get the model name also stored in the coeffs series

predict(X)[source]#

Predict the extinction in the specific passband

Note

if X is a Dataframe, teffnorm could be automatically added

Parameters:: X (Union[np.ndarray, pd.DataFrame]) – input features
Returns:: y – predicted values
Return type:: np.ndarray

to_ecsv(fname, **meta)[source]#

Export model into an ECSV file

Parameters:: fname (str)

to_pandas()[source]#

Export the model to a pandas array, useful for storage

Return type:: DataFrame

approx_model(r, *, passband='GAIA_GAIA3.G', degree=3, interaction_only=False, input_parameters=None, verbose=False)[source]#

Fit the passband grid data with a polynomial model.

Parameters:

r (DataFrame) – DataFrame with the passband grid data.
passband (str) – The passband to fit.
degree (int) – The degree of the polynomial model.
interaction_only (bool) – If True, only the interaction terms are used.
input_parameters (Sequence[str]) – The input parameters to use. If None, ‘teff logg feh A0 alpha’ parameters are used.
verbose (bool) – If True, print the model parameters and statistics

Returns:

Dictionary with the model parameters and statistics.

Return type:

dict

quick_plot_models(r, **kwargs)[source]#

Plot diagnostic plots for the models.

Parameters:

r (DataFrame) – DataFrame with the passband grid data.

Returns:

DataFrame – DataFrame with the all model parameters and statistics.
.. seealso:: – approx_model()

Return type:

Generator[DataFrame, None, None]

Module contents#

We provide various modeling schemes for extinction in a given photometric band.

Todo

add script to generate grid of models
add polynomial training.
compare literature values to ours.

class ModelInfo[source]#

Bases: object

Information about a precomputed model

__init__(atmosphere, extinction, comment, model, passbands, filename, _source_library=None)#

Parameters:

atmosphere (dict)
extinction (dict)
comment (Sequence[str])
model (dict)
passbands (Sequence[str])
filename (str)
_source_library (PrecomputedModel | None)

Return type:

None

atmosphere: dict = <dataclasses._MISSING_TYPE object>#

comment: Sequence[str] = <dataclasses._MISSING_TYPE object>#

copy()[source]#

Create a copy of this ModelInfo

Return type:: ModelInfo

extinction: dict = <dataclasses._MISSING_TYPE object>#

filename: str = <dataclasses._MISSING_TYPE object>#

load_model(passband=None)[source]#

Load the model described by this info

Parameters:: passband (str) – The passband to be loaded. If None, loads all available passband models.
Returns:: model
Return type:: dustapprox.models.polynomial.PolynomialModel

model: dict = <dataclasses._MISSING_TYPE object>#

passbands: Sequence[str] = <dataclasses._MISSING_TYPE object>#

class PrecomputedModel[source]#

Bases: object

Access to precomputed models

from dustapprox.models import PrecomputedModel

lib = PrecomputedModel()
# search for GALEX passbands if present
r = lib.find(passband="galex")
print(r)
# load both available models
models = []
for source in r.values():
    models.extend(
        [
            lib.load_model(r, passband=pbname)
            for pbname in source["passbands"]
        ]
    )

result from PrecomputedModel.find()#

[{'atmosphere': {'source': 'Kurucz (ODFNEW/NOVER 2003)',
    'teff': [3500.0, 50000.0],
    'logg': [0.0, 5.0],
    'feh': [-4, 0.5],
    'alpha': [0, 0.4]},
  'extinction': {'source': 'Fitzpatrick (1999)', 'R0': 3.1, 'A0': [0, 10]},
  'comment': ['teffnorm = teff / 5040', 'predicts kx = Ax / A0'],
  'model': {'kind': 'polynomial',
  'degree': 3,
  'interaction_only': False,
  'include_bias': True,
  'feature_names': ['A0', 'teffnorm']},
  'passbands': ['GALEX_GALEX.FUV', 'GALEX_GALEX.NUV'],
  'filename': 'dustapprox/data/precomputed/polynomial/f99/kurucz/kurucz_f99_a0_teff.ecsv'}]

result when loading models with from PrecomputedModel.load_model()#

[PolynomialModel: GALEX_GALEX.FUV
<dustapprox.models.polynomial.PolynomialModel object at 0x12917b6a0>
    from: A0, teffnorm   polynomial degree: 3,
PolynomialModel: GALEX_GALEX.NUV
<dustapprox.models.polynomial.PolynomialModel object at 0x129170820>
    from: A0, teffnorm   polynomial degree: 3]

Constructor

__init__(location=None)[source]#

Constructor

Parameters:: location (str | None)

find(*, passband=None, extinction=None, atmosphere=None, kind=None)[source]#

Find all the computed models that match the given parameters.

The search is case insentive and returns all matches.

Parameters:

passband (str) – The passband to be used.
extinction (str) – The extinction model to be used. (e.g., ‘Fitzpatrick’)
atmosphere (str) – The atmosphere model to be used. (e.g., ‘kurucz’)
kind (str) – The kind of model to be used (e.g., polynomial).

Return type:

Sequence[ModelInfo]

get_models_info(glob_pattern='/**/*.ecsv')[source]#

Retrieve the information for all models available and files :param glob_pattern: The glob pattern to use to search for model files. :type glob_pattern: str

Returns:: info – The list of model information structures.
Return type:: list of ModelInfo
Parameters:: glob_pattern (str)

load_model(fname, passband=None)[source]#

Load a model from a file or description (PrecomputedModel.find())

Parameters:

fname (str or dict) – The filename of the model to be loaded or a description of the model returned by PrecomputedModel.find()
passband (str) – The passband to be loaded. If None, loads all available passband models.

Returns:

model

Return type:

dustapprox.models.polynomial.PolynomialModel

dustapprox.models package

Contents

dustapprox.models package#

Submodules#

dustapprox.models.basemodel module#

dustapprox.models.polynomial module#

Module contents#