regressproxy

Proxy model classes for regression analysis

This module contains regression proxy models with lag and finite lifetime.

regressproxy.models_cel

Proxy classes for regression analysis (celerite version)

Proxy model classes for regression analysis using the celerite.modeling modelling protocol [1].

[1]https://celerite.readthedocs.io/en/stable/python/modeling/
class regressproxy.models_cel.ConstantModel(*args, **kwargs)[source]

Bases: Model

A simple concrete model with a single parameter value

Parameters:

value (float) – The value of the model.
Attributes:
full_size

The total number of parameters (including frozen parameters)

parameter_vector

An array of all parameters (including frozen parameters)

vector_size

The number of active (or unfrozen) parameters

Methods

compute_gradient(x) Compute the "gradient" of the model for the current parameters
freeze_all_parameters() Freeze all parameters of the model
freeze_parameter(name) Freeze a parameter by name
get_parameter(name) Get a parameter value by name
get_parameter_bounds([include_frozen]) Get a list of the parameter bounds
get_parameter_dict([include_frozen]) Get an ordered dictionary of the parameters
get_parameter_names([include_frozen]) Get a list of the parameter names
get_parameter_vector([include_frozen]) Get an array of the parameter values in the correct order
get_value(x) Compute the "value" of the model for the current parameters
log_prior() Compute the log prior probability of the current parameters
set_parameter(name, value) Set a parameter value by name
set_parameter_vector(vector[, include_frozen]) Set the parameter values to the given vector
thaw_all_parameters() Thaw all parameters of the model
thaw_parameter(name) Thaw a parameter by name
get_gradient  
compute_gradient(x)[source]

Compute the “gradient” of the model for the current parameters

This method should be overloaded by subclasses to implement the actual functionality of the model. The output of this function should be an array where the first dimension is full_size.

get_value(x)[source]

Compute the “value” of the model for the current parameters

This method should be overloaded by subclasses to implement the actual functionality of the model.

parameter_names = ('value',)
class regressproxy.models_cel.HarmonicModelAmpPhase(*args, **kwargs)[source]

Bases: Model

Model for harmonic terms

Models a harmonic term using amplitude and phase of a sine. It hase the same phase as returned by HarmonicModelCosineSine.get_phase().

Parameters:
  • freq (float) – The frequency in years^-1
  • amp (float) – The amplitude of the harmonic term
  • phase (float) – The phase of the harmonic part

See also

HarmonicModelCosineSine

Attributes:
full_size

The total number of parameters (including frozen parameters)

parameter_vector

An array of all parameters (including frozen parameters)

vector_size

The number of active (or unfrozen) parameters

Methods

compute_gradient(t) Compute the "gradient" of the model for the current parameters
freeze_all_parameters() Freeze all parameters of the model
freeze_parameter(name) Freeze a parameter by name
get_parameter(name) Get a parameter value by name
get_parameter_bounds([include_frozen]) Get a list of the parameter bounds
get_parameter_dict([include_frozen]) Get an ordered dictionary of the parameters
get_parameter_names([include_frozen]) Get a list of the parameter names
get_parameter_vector([include_frozen]) Get an array of the parameter values in the correct order
get_value(t) Compute the "value" of the model for the current parameters
log_prior() Compute the log prior probability of the current parameters
set_parameter(name, value) Set a parameter value by name
set_parameter_vector(vector[, include_frozen]) Set the parameter values to the given vector
thaw_all_parameters() Thaw all parameters of the model
thaw_parameter(name) Thaw a parameter by name
get_amplitude  
get_gradient  
get_phase  
compute_gradient(t)[source]

Compute the “gradient” of the model for the current parameters

This method should be overloaded by subclasses to implement the actual functionality of the model. The output of this function should be an array where the first dimension is full_size.

get_amplitude()[source]
get_phase()[source]
get_value(t)[source]

Compute the “value” of the model for the current parameters

This method should be overloaded by subclasses to implement the actual functionality of the model.

parameter_names = ('freq', 'amp', 'phase')
class regressproxy.models_cel.HarmonicModelCosineSine(*args, **kwargs)[source]

Bases: Model

Model for harmonic terms

Models a harmonic term using a cosine and sine part. The amplitude and phase returned are that for a sine function, i.e. amplitude * sin(t + phase).

Parameters:
  • freq (float) – The frequency in years^-1
  • cos (float) – The amplitude of the cosine part
  • sin (float) – The amplitude of the sine part

See also

HarmonicModelAmpPhase

Attributes:
full_size

The total number of parameters (including frozen parameters)

parameter_vector

An array of all parameters (including frozen parameters)

vector_size

The number of active (or unfrozen) parameters

Methods

compute_gradient(t) Compute the "gradient" of the model for the current parameters
freeze_all_parameters() Freeze all parameters of the model
freeze_parameter(name) Freeze a parameter by name
get_parameter(name) Get a parameter value by name
get_parameter_bounds([include_frozen]) Get a list of the parameter bounds
get_parameter_dict([include_frozen]) Get an ordered dictionary of the parameters
get_parameter_names([include_frozen]) Get a list of the parameter names
get_parameter_vector([include_frozen]) Get an array of the parameter values in the correct order
get_value(t) Compute the "value" of the model for the current parameters
log_prior() Compute the log prior probability of the current parameters
set_parameter(name, value) Set a parameter value by name
set_parameter_vector(vector[, include_frozen]) Set the parameter values to the given vector
thaw_all_parameters() Thaw all parameters of the model
thaw_parameter(name) Thaw a parameter by name
get_amplitude  
get_gradient  
get_phase  
compute_gradient(t)[source]

Compute the “gradient” of the model for the current parameters

This method should be overloaded by subclasses to implement the actual functionality of the model. The output of this function should be an array where the first dimension is full_size.

get_amplitude()[source]
get_phase()[source]
get_value(t)[source]

Compute the “value” of the model for the current parameters

This method should be overloaded by subclasses to implement the actual functionality of the model.

parameter_names = ('freq', 'cos', 'sin')
class regressproxy.models_cel.ProxyModel(proxy_times, proxy_vals, center=False, phi_intp=None, fit_phase=False, lifetime_prior=None, lifetime_metric=1.0, days_per_time_unit=365.25, *args, **kwargs)[source]

Bases: Model

Model for proxy terms

Models proxy terms with a finite and (semi-)annually varying life time.

Parameters:
  • proxy_times ((N,) array_like) – The times of the proxy values according to days_per_time_unit.
  • proxy_vals ((N,) array_like) – The proxy values at proxy_times.
  • amp (float) – The amplitude of the proxy term.
  • lag (float) – The lag of the proxy value (in days, see days_per_time_unit).
  • tau0 (float) – The base life time of the proxy (in days, see days_per_time_unit).
  • taucos1 (float) – The amplitude of the cosine part of the annual life time variation (in days, see days_per_time_unit).
  • tausin1 (float) – The amplitude of the sine part of the annual life time variation (in days, see days_per_time_unit).
  • taucos2 (float) – The amplitude of the cosine part of the semi-annual life time variation (in days, see days_per_time_unit).
  • tausin2 (float) – The amplitude of the sine part of the semi-annual life time variation (in days, see days_per_time_unit).
  • ltscan (float) – The number of days to sum the previous proxy values. If it is negative, the value will be set to three times the maximal lifetime. No lifetime adjustemets are calculated when set to zero.
  • center (bool, optional) – Centers the proxy values by subtracting the overall mean. The mean is calculated from the whole proxy_vals array and is stored in the mean attribute. Default: False
  • phi_intp (scipy.interpolate.interp1d() instance, optional) – When not None, the interpolated angle phi (e.g. SZA) and cos(phi) and sin(phi) are used to model the variation of the lifetime instead of the time. Semi-annual variations are not used in this case. Default: None
  • fit_phase (bool, optional) – Fit the phase shift directly instead of using sine and cosine terms for the (semi-)annual lifetime variations. If True, the fitted cosine parameter is the amplitude and the sine parameter the phase. Default: False (= fit sine and cosine terms)
  • lifetime_prior (str, optional) – The prior probability density for each coefficient of the lifetime. Possible types are “flat” or None for a flat prior, “exp” for an exponential density ~ \(\text{exp}(-|\tau| / \text{metric})\), and “normal” for a normal distribution ~ \(\text{exp}(-\tau^2 / (2 * \text{metric}^2))\). Default: None (= flat prior).
  • lifetime_metric (float, optional) – The metric (scale) of the lifetime priors in days, see prior. Default 1.
  • days_per_time_unit (float, optional) – The number of days per time unit, used to normalize the lifetime units. Use 365.25 if the times are in fractional years, or 1 if they are in days. Default: 365.25
Attributes:
full_size

The total number of parameters (including frozen parameters)

parameter_vector

An array of all parameters (including frozen parameters)

vector_size

The number of active (or unfrozen) parameters

Methods

compute_gradient(t) Compute the "gradient" of the model for the current parameters
freeze_all_parameters() Freeze all parameters of the model
freeze_parameter(name) Freeze a parameter by name
get_parameter(name) Get a parameter value by name
get_parameter_bounds([include_frozen]) Get a list of the parameter bounds
get_parameter_dict([include_frozen]) Get an ordered dictionary of the parameters
get_parameter_names([include_frozen]) Get a list of the parameter names
get_parameter_vector([include_frozen]) Get an array of the parameter values in the correct order
get_value(t) Compute the "value" of the model for the current parameters
log_prior() Compute the log prior probability of the current parameters
set_parameter(name, value) Set a parameter value by name
set_parameter_vector(vector[, include_frozen]) Set the parameter values to the given vector
thaw_all_parameters() Thaw all parameters of the model
thaw_parameter(name) Thaw a parameter by name
get_gradient  
compute_gradient(t)[source]

Compute the “gradient” of the model for the current parameters

This method should be overloaded by subclasses to implement the actual functionality of the model. The output of this function should be an array where the first dimension is full_size.

get_value(t)[source]

Compute the “value” of the model for the current parameters

This method should be overloaded by subclasses to implement the actual functionality of the model.

log_prior()[source]

Compute the log prior probability of the current parameters

parameter_names = ('amp', 'lag', 'tau0', 'taucos1', 'tausin1', 'taucos2', 'tausin2', 'ltscan')
class regressproxy.models_cel.ProxyModelSet(models)[source]

Bases: ModelSet

Combined model class for, e.g. trace gases (and probably other data)

Inherited from celerite.ModelSet, provides get_value() and compute_gradient() methods.

Attributes:
dirty
full_size

The total number of parameters (including frozen parameters)

parameter_bounds
parameter_names
parameter_vector

An array of all parameters (including frozen parameters)

unfrozen_mask
vector_size

The number of active (or unfrozen) parameters

Methods

compute_gradient(t) Compute the "gradient" of the model for the current parameters
freeze_all_parameters() Freeze all parameters of the model
freeze_parameter(name) Freeze a parameter by name
get_parameter(name) Get a parameter value by name
get_parameter_bounds([include_frozen]) Get a list of the parameter bounds
get_parameter_dict([include_frozen]) Get an ordered dictionary of the parameters
get_parameter_names([include_frozen]) Get a list of the parameter names
get_parameter_vector([include_frozen]) Get an array of the parameter values in the correct order
get_value(t) Compute the "value" of the model for the current parameters
log_prior() Compute the log prior probability of the current parameters
set_parameter(name, value) Set a parameter value by name
set_parameter_vector(vector[, include_frozen]) Set the parameter values to the given vector
thaw_all_parameters() Thaw all parameters of the model
thaw_parameter(name) Thaw a parameter by name
get_gradient  
compute_gradient(t)[source]

Compute the “gradient” of the model for the current parameters

This method should be overloaded by subclasses to implement the actual functionality of the model. The output of this function should be an array where the first dimension is full_size.

get_value(t)[source]

Compute the “value” of the model for the current parameters

This method should be overloaded by subclasses to implement the actual functionality of the model.

regressproxy.models_cel.proxy_model_set(constant=True, freqs=None, proxy_config=None, **kwargs)[source]

Model set including proxies and harmonics

Sets up a proxy model for easy access. All parameters are optional, defaults to an offset, no harmonics, proxies uncentered and unscaled.

Parameters:
  • constant (bool, optional) – Whether or not to include a constant (offset) term, default is True.
  • freqs (list, optional) – Frequencies of the harmonic terms in 1 / a^-1 (inverse years).
  • proxy_config (dict, optional) – Proxy configuration if different from the standard setup.
  • **kwargs (optional) –

    Additional keyword arguments, all of them are also passed on to the proxy setup. For now, supported are the following which are also passed along to the proxy setup with setup_proxy_model_with_bounds():

    • fit_phasebool
      fit amplitude and phase instead of sine and cosine
    • scalefloat
      the factor by which the data is scaled, used to constrain the maximum and minimum amplitudes to be fitted.
    • time_formatstring
      The astropy.time.Time format string to setup the time axis.
    • days_per_time_unitfloat
      The number of days per time unit, used to normalize the frequencies for the harmonic terms. Use 365.25 if the times are in fractional years, 1 if they are in days. Default: 365.25
    • max_ampfloat
      Maximum magnitude of the coefficients, used to constrain the parameter search.
    • max_daysfloat
      Maximum magnitude of the lifetimes, used to constrain the parameter search.
Returns:

model
Return type:

ProxyModelSet (extends celerite.ModelSet)
regressproxy.models_cel.setup_proxy_model_with_bounds(times, values, max_amp=10000000000.0, max_days=100, **kwargs)[source]

regressproxy.models_pymc3

Proxy classes for regression analysis (pymc3/theano version)

Model classes for data regression proxies using theano for pymc3.

This interface is still experimental, it is available when installing the pymc3 (or theano) extra:

pip install "regressproxy[pymc3]"

The classes can be imported as usual, e.g. via:

from regressproxy.models_pymc3 import ProxyModel
class regressproxy.models_pymc3.HarmonicModelAmpPhase(freq, amp, phase)[source]

Bases: object

Model for harmonic terms

Models harmonic terms using amplitude and phase of a sine.

Parameters:
  • freq (float) – The frequency in years^-1
  • amp (float) – The amplitude of the harmonic term
  • phase (float) – The phase of the harmonic part

See also

HarmonicModelCosineSine

Methods

compute_gradient  
get_amplitude  
get_phase  
get_value  
compute_gradient(t)[source]
get_amplitude()[source]
get_phase()[source]
get_value(t)[source]
class regressproxy.models_pymc3.HarmonicModelCosineSine(freq, cos, sin)[source]

Bases: object

Model for harmonic terms

Models a harmonic term using a cosine and sine part. The amplitude and phase returned are that for a sine function, i.e. amplitude * sin(t + phase).

Parameters:
  • freq (float) – The frequency in years^-1
  • cos (float) – The amplitude of the cosine part
  • sin (float) – The amplitude of the sine part

See also

HarmonicModelAmpPhase

Methods

compute_gradient  
get_amplitude  
get_phase  
get_value  
compute_gradient(t)[source]
get_amplitude()[source]
get_phase()[source]
get_value(t)[source]
class regressproxy.models_pymc3.LifetimeModel(harmonics, lower=0.0)[source]

Bases: object

Model for variable lifetime

Returns the positive values of the sine/cosine.

Parameters:harmonics (HarmonicModelCosineSine or HarmonicModelAmpPhase or list) –

Methods

get_value  
get_value(t)[source]
class regressproxy.models_pymc3.ModelSet(models)[source]

Bases: object

Methods

get_value  
get_value(t)[source]
class regressproxy.models_pymc3.ProxyModel(ptimes, pvalues, amp, lag=0.0, tau0=0.0, tau_harm=None, tau_scan=0, days_per_time_unit=365.25)[source]

Bases: object

Model for proxy terms

Models proxy terms with a finite and (semi-)annually varying life time.

Parameters:
  • proxy_times ((N,) array_like) – The times of the proxy values according to days_per_time_unit.
  • proxy_vals ((N,) array_like) – The proxy values at proxy_times.
  • amp (float) – The amplitude of the proxy term.
  • lag (float, optional) – The lag of the proxy value (in days, see days_per_time_unit).
  • tau0 (float, optional) – The base life time of the proxy (in days, see days_per_time_unit).
  • tau_harm (LifetimeModel, optional) – The lifetime harmonic model with a lower limit.
  • tau_scan (float, optional) – The number of days to sum the previous proxy values. If it is negative, the value will be set to three times the maximal lifetime. No lifetime adjustemets are calculated when set to zero.
  • days_per_time_unit (float, optional) – The number of days per time unit, used to normalize the lifetime units. Use 365.25 if the times are in fractional years, or 1 if they are in days. Default: 365.25

Methods

get_value  
get_value(t, interpolate=True)[source]
regressproxy.models_pymc3.proxy_model_set(constant=True, freqs=None, proxy_config=None, **kwargs)[source]

Model set including proxies and harmonics

Sets up a proxy model for easy access. All parameters are optional, defaults to an offset, no harmonics, proxies uncentered and unscaled.

Parameters:
  • constant (bool, optional) – Whether or not to include a constant (offset) term, default is True.
  • freqs (list, optional) – Frequencies of the harmonic terms in 1 / a^-1 (inverse years).
  • proxy_config (dict, optional) – Proxy configuration if different from the standard setup.
  • **kwargs (optional) –

    Additional keyword arguments, all of them are also passed on to the proxy setup. For now, supported are the following which are also passed along to the proxy setup with setup_proxy_model_with_bounds():

    • fit_phasebool
      fit amplitude and phase instead of sine and cosine
    • scalefloat
      the factor by which the data is scaled, used to constrain the maximum and minimum amplitudes to be fitted.
    • time_formatstring
      The astropy.time.Time format string to setup the time axis.
    • days_per_time_unitfloat
      The number of days per time unit, used to normalize the frequencies for the harmonic terms. Use 365.25 if the times are in fractional years, 1 if they are in days. Default: 365.25
    • max_ampfloat
      Maximum magnitude of the coefficients, used to constrain the parameter search.
    • max_daysfloat
      Maximum magnitude of the lifetimes, used to constrain the parameter search.
    • model_kwargsdict
      Keyword arguments passed to pymc.Model(), e.g. name or coords.
Returns:

model, ModelSet, offset – The pymc3.model.Model containing the random variables, the ModelSet with entries of type ProxyModel as defined by proxy_config. The offset is included to keep it similar to the celerite model setup.
Return type:

tuple

See also

HarmonicModelCosineSine, ProxyModel

regressproxy.models_pymc3.setup_proxy_model_pymc3(model, name, times, values, max_amp=10000000000.0, max_days=100, **kwargs)[source]

regressproxy.models_pymc4

Proxy classes for regression analysis (pymc4/5 version)

Model classes for data regression proxies for use with pymc versions 4 and 5.

This interface is still experimental, it is available when installing either the pymc4 or pymc5 extra:

pip install "regressproxy[pymc4]"

Due to dependency reasons, versions 4 and 5 are not supported together. The classes can be imported as usual, e.g. via:

from regressproxy.models_pymc4 import ProxyModel
class regressproxy.models_pymc4.HarmonicModelAmpPhase(freq, amp, phase)[source]

Bases: object

Model for harmonic terms

Models harmonic terms using amplitude and phase of a sine.

Parameters:
  • freq (float) – The frequency in years^-1
  • amp (float) – The amplitude of the harmonic term
  • phase (float) – The phase of the harmonic part

Methods

compute_gradient  
get_amplitude  
get_phase  
get_value  
compute_gradient(t)[source]
get_amplitude()[source]
get_phase()[source]
get_value(t)[source]
class regressproxy.models_pymc4.HarmonicModelCosineSine(freq, cos, sin)[source]

Bases: object

Model for harmonic terms

Models harmonic terms using a cosine and sine part. The total amplitude and phase can be inferred from that.

Parameters:
  • freq (float) – The frequency in years^-1
  • cos (float) – The amplitude of the cosine part
  • sin (float) – The amplitude of the sine part

Methods

compute_gradient  
get_amplitude  
get_phase  
get_value  
compute_gradient(t)[source]
get_amplitude()[source]
get_phase()[source]
get_value(t)[source]
class regressproxy.models_pymc4.LifetimeModel(harmonics, lower=0.0)[source]

Bases: object

Model for variable lifetime

Returns the positive values of the sine/cosine.

Parameters:harmonics (HarmonicModelCosineSine or HarmonicModelAmpPhase or list) –

Methods

get_value  
get_value(t)[source]
class regressproxy.models_pymc4.ModelSet(models)[source]

Bases: object

Methods

get_value  
get_value(t)[source]
class regressproxy.models_pymc4.ProxyModel(ptimes, pvalues, amp, lag=0.0, tau0=0.0, tau_harm=None, tau_scan=0, days_per_time_unit=365.25)[source]

Bases: object

Model for proxy terms

Models proxy terms with a finite and (semi-)annually varying life time.

Parameters:
  • proxy_times ((N,) array_like) – The times of the proxy values according to days_per_time_unit.
  • proxy_vals ((N,) array_like) – The proxy values at proxy_times
  • amp (float) – The amplitude of the proxy term
  • lag (float, optional) – The lag of the proxy value (in days, see days_per_time_unit).
  • tau0 (float, optional) – The base life time of the proxy (in days, see days_per_time_unit).
  • tau_harm (LifetimeModel, optional) – The lifetime harmonic model with a lower limit.
  • tau_scan (float, optional) – The number of days to sum the previous proxy values. If it is negative, the value will be set to three times the maximal lifetime. No lifetime adjustemets are calculated when set to zero.
  • days_per_time_unit (float, optional) – The number of days per time unit, used to normalize the lifetime units. Use 365.25 if the times are in fractional years, or 1 if they are in days. Default: 365.25

Methods

get_value  
get_value(t, interpolate=True)[source]
regressproxy.models_pymc4.proxy_model_set(constant=True, freqs=None, proxy_config=None, **kwargs)[source]

Model set including proxies and harmonics

Sets up a proxy model for easy access. All parameters are optional, defaults to an offset, no harmonics, proxies uncentered and unscaled.

Parameters:
  • constant (bool, optional) – Whether or not to include a constant (offset) term, default is True.
  • freqs (list, optional) – Frequencies of the harmonic terms in 1 / a^-1 (inverse years).
  • proxy_config (dict, optional) – Proxy configuration if different from the standard setup.
  • **kwargs (optional) –

    Additional keyword arguments, all of them are also passed on to the proxy setup. For now, supported are the following which are also passed along to the proxy setup with setup_proxy_model_with_bounds():

    • fit_phasebool
      fit amplitude and phase instead of sine and cosine
    • scalefloat
      the factor by which the data is scaled, used to constrain the maximum and minimum amplitudes to be fitted.
    • time_formatstring
      The astropy.time.Time format string to setup the time axis.
    • days_per_time_unitfloat
      The number of days per time unit, used to normalize the frequencies for the harmonic terms. Use 365.25 if the times are in fractional years, 1 if they are in days. Default: 365.25
    • max_ampfloat
      Maximum magnitude of the coefficients, used to constrain the parameter search.
    • max_daysfloat
      Maximum magnitude of the lifetimes, used to constrain the parameter search.
    • model_kwargsdict
      Keyword arguments passed to pymc.Model(), e.g. name or coords.
Returns:

model, ModelSet, offset – The pymc.Model containing the random variables, the ModelSet with entries of type ProxyModel as defined by proxy_config. The offset is included to keep it similar to the celerite model setup.
Return type:

tuple

See also

HarmonicModelCosineSine, ProxyModel

regressproxy.models_pymc4.setup_proxy_model_pymc4(model, name, times, values, max_amp=10000000000.0, max_days=100, **kwargs)[source]