anml.parameter package

Submodules

Likelihood

The Likelihood keeps track of all information related to probability distributions that can be used for priors or data distributions.

class GaussianLikelihood(mean=None, std=None)[source]

Bases: anml.parameter.likelihood.Likelihood

class Likelihood[source]

Bases: object

get_grad(vals, parameters=None)[source]
get_neg_log_likelihood(vals, parameters=None)[source]

Gets the objective value based on parameters of the distribution and current values in vals.

Parameters

  • vals (Union[float, ndarray]) – location at which to evaluate the oracle

  • parameters (Union[List[float], List[ndarray], None]) – parameters of the oracle, optional. If not passed it will use the parameters attached to self, otherwise return an error if there are no parameters.

Returns

Return type

float representing the objective value evaluated based on vals and parameters.

exception LikelihoodError[source]

Bases: anml.exceptions.ANMLError

Raised when there is a mismatch in a distribution’s parameters.

Prior Distributions

Gives prior specifications that can be attached to fixed and/or random effect components of a variable and may be used in the solver optimization.

The default prior only includes upper and lower bounds for box constraints and defaults to [-np.inf, np.inf]. Alternative priors include GaussianPrior.

In order to get the error that should be added to the objective function value, each prior is associated with an Likelihood.

class GaussianPrior(lower_bound=<factory>, upper_bound=<factory>, mean=<factory>, std=<factory>)[source]

Bases: anml.parameter.prior.Prior

error_value(vals)[source]
grad(vals)[source]
mean: List[float]
std: List[float]
exception GaussianPriorError[source]

Bases: anml.parameter.prior.PriorError

class Prior(lower_bound=<factory>, upper_bound=<factory>)[source]

Bases: object

error_value(val)[source]
grad(val)[source]
lower_bound: List[float]
upper_bound: List[float]
exception PriorError[source]

Bases: anml.exceptions.ANMLError

Variables

Variables are the most granular object for constructing a model specification. At the simplest level, a variable is just Intercept, which is a column of ones (indicating that it does not change based on the data row, except through an optional random effect).

Each Variable has a collection of methods (e.g., build_design_matrix_fe()) that gets the design matrices, constraint matrices and bounds for that single covariate.

class Intercept(var_link_fun=<function Variable.<lambda>>, fe_prior=Prior(lower_bound=[-inf], upper_bound=[inf], _likelihood=<anml.parameter.likelihood.Likelihood object>), add_re=False, col_group=None, re_var_prior=Prior(lower_bound=[-inf], upper_bound=[inf], _likelihood=<anml.parameter.likelihood.Likelihood object>), re_prior=Prior(lower_bound=[0.0], upper_bound=[inf], _likelihood=<anml.parameter.likelihood.Likelihood object>))[source]

Bases: anml.parameter.variables.Variable

An intercept variable.

class ParameterBlock[source]

Bases: object

constr_lb_fe: Optional[numpy.ndarray] = None
constr_lb_re: Optional[numpy.ndarray] = None
constr_lb_re_var: Optional[numpy.ndarray] = None
constr_matrix_fe: Optional[numpy.ndarray] = None
constr_matrix_re: Optional[numpy.ndarray] = None
constr_matrix_re_var: Optional[numpy.ndarray] = None
constr_ub_fe: Optional[numpy.ndarray] = None
constr_ub_re: Optional[numpy.ndarray] = None
constr_ub_re_var: Optional[numpy.ndarray] = None
design_matrix_fe: Optional[numpy.ndarray] = None
design_matrix_re: Optional[numpy.ndarray] = None
num_fe: int = 0
property num_re
num_re_var: int = 0
re_priors: Optional[numpy.ndarray] = None
re_var_padding: Optional[numpy.ndarray] = None
reset()[source]
variables: List[anml.parameter.variables.Variable]
class Variable(covariate=None, var_link_fun=<function Variable.<lambda>>, fe_prior=Prior(lower_bound=[-inf], upper_bound=[inf], _likelihood=<anml.parameter.likelihood.Likelihood object>), add_re=False, col_group=None, re_var_prior=Prior(lower_bound=[-inf], upper_bound=[inf], _likelihood=<anml.parameter.likelihood.Likelihood object>), re_prior=Prior(lower_bound=[0.0], upper_bound=[inf], _likelihood=<anml.parameter.likelihood.Likelihood object>))[source]

Bases: object

A class that stores information about a variable.

Parameters

  • covariate (str) – name of the covariate for this variable.

  • var_link_fun (callable) – link function for this variable.

  • fe_prior (Prior, optional) – a prior of class Prior for fixed effects coefficient

  • add_re (bool, optional) – whether to add random effects to this variable

  • col_group (str, optional) – name for group column

  • re_var_prior (Prior, optional) – a prior of class Prior for random effect variance

  • re_prior (Prior, optional) – a prior of class Prior for random effects.

add_re: bool = False
build_bounds_fe()[source]

Build bounds for fixed effects

build_bounds_re()[source]

Build bounds for random effects.

build_bounds_re_var()[source]

Build bounds for random effects covariance.

build_constraint_matrix_fe()[source]

Build constraint matrix for fixed effects

build_constraint_matrix_re()[source]

Build constraint matrix for random effects

build_constraint_matrix_re_var()[source]

Build constraint matrix for random effects covariance.

build_design_matrix_fe(df, **kwargs)[source]

Build design matrix corresponding to fixed effects.

Parameters

df (pd.DataFrame) – input dataframe

build_design_matrix_re(df)[source]

Build design matrix corresponding to random effects covariances.

Parameters

df (pd.DataFrame) – input dataframe

col_group: str = None
covariate: str = None
design_matrix_fe: Optional[numpy.ndarray]
design_matrix_re: Optional[numpy.ndarray]
encode_groups(df)[source]

Convert a categorical column into ordinal numbers.

Parameters

df (pd.DataFrame) – input dataframe

Returns

a list of ints indicating category of each datapoint.

Return type

List[int]

Raises

ValueError – Only one group in the entire input dataframe.

fe_prior: anml.parameter.prior.Prior = Prior(lower_bound=[-inf], upper_bound=[inf], _likelihood=<anml.parameter.likelihood.Likelihood object>)
re_prior: anml.parameter.prior.Prior = Prior(lower_bound=[0.0], upper_bound=[inf], _likelihood=<anml.parameter.likelihood.Likelihood object>)
re_var_prior: anml.parameter.prior.Prior = Prior(lower_bound=[-inf], upper_bound=[inf], _likelihood=<anml.parameter.likelihood.Likelihood object>)
reset()[source]
exception VariableError[source]

Bases: anml.exceptions.ANMLError

collect_blocks(param_block, attr_name, build_func=None, should_include=<function <lambda>>, reset_params=False, inputs=None)[source]

Spline Variable

A subclass of anml.parameter.variables.Variable that handles spline related computations.

class Spline(covariate=None, var_link_fun=<function Variable.<lambda>>, col_group=None, re_var_prior=Prior(lower_bound=[-inf], upper_bound=[inf], _likelihood=<anml.parameter.likelihood.Likelihood object>), re_prior=Prior(lower_bound=[0.0], upper_bound=[inf], _likelihood=<anml.parameter.likelihood.Likelihood object>), knots_type='frequency', knots_num=3, degree=3, l_linear=False, r_linear=False, include_intercept=False, derivative_constr=<factory>, constr_grid_size_global=None)[source]

Bases: anml.parameter.variables.Variable

Spline variable.

Parameters

  • knots_type (str) – type of knots. can only be ‘frequency’ or ‘domain’

  • knots_num (int) – number of knots

  • degree (int) – degree of spines

  • l_linear (bool) – whether left tail is linear

  • r_linear (bool) – whether right tail is linear

  • include_intercept (bool) – whether to include intercept in design matrix

  • derivative_constr (List[~anml.parameter.spline_variable.SplineLinearConstr]) – constraints on derivatives

  • constr_grid_size_global (int, optional) – number of points to use when building constraint matrix. used only when grid_size for individual ~anml.parameter.spline_variable.SplineLinearConstr is not available

Raises
build_constraint_matrix_fe()[source]

build constrain matrix and bounds for constr_lb <= constr_matrix <= constr_ub.

Returns

constraint matrix, lower bounds and upper bounds.

Return type

List[np.ndarray]

constr_grid_size_global: int = None
constr_lb_fe: Optional[Union[List[float], numpy.ndarray]]
constr_matrix_fe: Optional[numpy.ndarray]
constr_ub_fe: Optional[Union[List[float], numpy.ndarray]]
create_spline(df)[source]
degree: int = 3
derivative_constr: List[anml.parameter.spline_variable.SplineLinearConstr]
include_intercept: bool = False
knots_num: int = 3
knots_type: str = 'frequency'
l_linear: bool = False
r_linear: bool = False
set_fe_prior(prior)[source]
spline: Optional[xspline.core.XSpline]
x: Optional[numpy.ndarray]
class SplineLinearConstr(order, y_bounds, x_domain=<factory>, grid_size=None)[source]

Bases: object

Constraints on spline derivatives. The general form is lb <= Ax <= ub where x is in some interval domain x_domain, and A can be 0th, 1st or 2nd order derivative matrix of the splines evaluated at some discretization points. A is not known at the initialization of this object, but will have dimension grid_size by number of spline basis. lb and ub are vectors of multiples of ones.

This type of constraints can be used to impose monotonicity and convexity constraints. For instance, for splines defined on [0, 5], one can specify monotonically decreasing on [0,1] with constr = SplineLinearConstr(x_domain=[0, 1], y_bounds=[-np.inf, 0.0], order=1), monotonically increasing on [4, 5] with constr = SplineLinearConstr(x_domain=[4, 5], y_bounds=[0.0, np.inf], order=1), and overall convexity with constr = SplineLinearConstr(x_domain=[0, 5], y_bounds=[0.0, np.inf], order=2).

Parameters

  • order (int) – order of the derivative

  • y_bounds (List[float, float]) – bounds for y = Ax

  • x_domain (List[float, float], optional) – domain for x, default to be -inf to inf

  • grid_size (int, optional) – size of grid

Raises

  • ValueError – domain for x is not valid.

  • ValueError – bounds for y = Ax is not valid.

  • ValueError – invalid derivative order

  • ValueError – invalid grid size

grid_size: int = None
order: int
x_domain: List[float]
y_bounds: List[float]
class Parameter(variables, param_name, link_fun=<function Parameter.<lambda>>)[source]

Bases: anml.parameter.variables.ParameterBlock

A class for parameters.

Parameters

  • param_name (str) – name of the parameter

  • link_fun (callable) – link function for the parameter

  • variables (List[Variable]) – a list of variables

All attributes from :class:`~anml.parameter.parameter.Variable`s in `variables`
are carried over but are put into a list.
property num_re
param_name: str
variables: List[anml.parameter.variables.Variable]
exception ParameterError[source]

Bases: anml.exceptions.ANMLError

class ParameterFunction(param_function_name, param_function, param_function_fe_prior=Prior(lower_bound=[-inf], upper_bound=[inf], _likelihood=<anml.parameter.likelihood.Likelihood object>))[source]

Bases: object

A class for function on parameters.

Parameters

  • param_function_name (str) – name of the parameter function

  • param_function (callable) – parameter function

  • param_function_fe_prior (List[float]) – a list of two floats specifying mean and std for Gaussian prior on the function.

param_function: Callable
param_function_fe_prior: anml.parameter.prior.Prior = Prior(lower_bound=[-inf], upper_bound=[inf], _likelihood=<anml.parameter.likelihood.Likelihood object>)
param_function_name: str
class ParameterSet(parameters, parameter_functions=None)[source]

Bases: anml.parameter.variables.ParameterBlock

A class for a set of parameters.

Parameters

  • parameters (List[Parameter]) – a list of parameters.

  • parameter_functions (List[ParameterFunction]) – a list of parameter functions.

get_param_function_index(param_function_name)[source]

A function that returns index of a given parameter function.

Parameters

param_function_name (str) – name of the parameter function

Returns

index of the parameter function

Return type

int

Raises

RuntimeError – parameter function not found in the parameter set.

get_param_index(param_name)[source]

A function that returns index of a given parameter.

Parameters

param_name (str) – name of the parameter

Returns

index of the parameter

Return type

int

Raises

RuntimeError – parameter not found in the parameter set.

property num_re
param_name: List[str]
parameter_functions: List[anml.parameter.parameter.ParameterFunction] = None
parameters: List[anml.parameter.parameter.Parameter]
variables: List[anml.parameter.variables.Variable]
exception ParameterSetError[source]

Bases: anml.exceptions.ANMLError

process_all(param_block, df)[source]
process_for_betas(param_block, df, reset=True)[source]
process_for_gammas(param_block, df, reset=False)[source]
process_for_us(param_block, df, reset=False)[source]
build_linear_constraint(constraints)[source]
build_re_matrix(matrix, group_assign_ord, n_groups)[source]
collect_priors(priors)[source]
combine_constraints(constr_matrix, constr_lb, constr_ub)[source]
encode_groups(group_assign_cat)[source]

Module contents