samplefit
samplefit is a Python library to assess sample fit, as opposed to model fit,
via the Sample Fit Reliability algorithm as developed by Okasa & Younge (2022).
samplefit is linked to the statsmodels library (Seabold & Perktold, 2010)
and follows the same command workflow.
Description
samplefit is a Python library for the assessment of sample fit in
econometric models. In particular, samplefit implements the Sample Fit
Reliability (SFR) algorithm, a re-sampling procedure to estimate the
reliability of data and check the sensitivity of results. To that end,
SFR is a computational approach with three aspects: Scoring, to estimate a
point-wise reliability score for every observation in a sample based on the
expected estimation loss over sub-samples; Annealing, to test the sensitivity
of results to the sequential removal of unreliable data points; and Fitting,
to estimate a weighted regression that adjusts for the reliability of the data.
Installation
To install the samplefit library from PyPi run:
pip install samplefit
or alternatively, to clone the repo run:
git clone https://github.com/okasag/samplefit.git
samplefit relies on Python 3 and requires the following dependencies:
- numpy (>=1.22.0)
- pandas (>=1.3.5)
- scipy (>=1.7.2)
- statsmodels (>=0.12.2)
- matplotlib (>=3.4.2)
- joblib (>=1.0.1)
- psutil (>=5.8.0)
The required modules can be installed by navigating to the root of
the cloned project and executing the following command:
pip install -r requirements.txt.
Examples
The example below demonstrates the workflow of using the samplefit library
in conjunction with the well-known statsmodels library.
Import libraries:
import samplefit as sf
import statsmodels.api as sm
Get data:
boston = sm.datasets.get_rdataset("Boston", "MASS")
Y = boston.data['crim']
X = boston.data['lstat']
X = sm.add_constant(X)
Assess model fit:
model = sm.OLS(endog=Y, exog=X)
model_fit = model.fit()
model_fit.summary()
Assess sample fit:
sample = sf.SFR(model=model)
sample_fit = sample.fit()
sample_fit.summary()
Assess sample reliability:
sample_scores = sample.score()
sample_scores.plot()
Assess sample sensitivity:
sample_annealing = sample.anneal()
sample_annealing.plot()
Authors
Gabriel Okasa & Kenneth A. Younge
References
- Okasa, Gabriel, and Kenneth A. Younge. “Sample Fit Reliability.” arXiv preprint arXiv:2209.06631. 2022.
- Seabold, Skipper, and Josef Perktold. “statsmodels: Econometric and statistical modeling with python.” Proceedings of the 9th Python in Science Conference. 2010.
Expand source code
"""
`samplefit` is a Python library to assess sample fit, as opposed to model fit,
via the *Sample Fit Reliability* algorithm as developed by Okasa & Younge (2022).
`samplefit` is linked to the `statsmodels` library (Seabold & Perktold, 2010)
and follows the same command workflow.
Description
----------------------------
`samplefit` is a Python library for the assessment of sample fit in
econometric models. In particular, `samplefit` implements the Sample Fit
Reliability (SFR) algorithm, a re-sampling procedure to estimate the
reliability of data and check the sensitivity of results. To that end,
SFR is a computational approach with three aspects: *Scoring*, to estimate a
point-wise reliability score for every observation in a sample based on the
expected estimation loss over sub-samples; *Annealing*, to test the sensitivity
of results to the sequential removal of unreliable data points; and *Fitting*,
to estimate a weighted regression that adjusts for the reliability of the data.
Installation
----------------------------
To install the `samplefit` library from `PyPi` run:
```
pip install samplefit
```
or alternatively, to clone the repo run:
```
git clone https://github.com/okasag/samplefit.git
```
`samplefit` relies on Python 3 and requires the following dependencies:
* numpy (>=1.22.0)
* pandas (>=1.3.5)
* scipy (>=1.7.2)
* statsmodels (>=0.12.2)
* matplotlib (>=3.4.2)
* joblib (>=1.0.1)
* psutil (>=5.8.0)
The required modules can be installed by navigating to the root of
the cloned project and executing the following command:
`pip install -r requirements.txt`.
Examples
----------------------------
The example below demonstrates the workflow of using the `samplefit` library
in conjunction with the well-known `statsmodels` library.
Import libraries:
```python
import samplefit as sf
import statsmodels.api as sm
```
Get data:
```python
boston = sm.datasets.get_rdataset("Boston", "MASS")
Y = boston.data['crim']
X = boston.data['lstat']
X = sm.add_constant(X)
```
Assess model fit:
```python
model = sm.OLS(endog=Y, exog=X)
model_fit = model.fit()
model_fit.summary()
```
Assess sample fit:
```python
sample = sf.SFR(model=model)
sample_fit = sample.fit()
sample_fit.summary()
```
Assess sample reliability:
```python
sample_scores = sample.score()
sample_scores.plot()
```
Assess sample sensitivity:
```python
sample_annealing = sample.anneal()
sample_annealing.plot()
```
Authors
----------------------------
Gabriel Okasa & Kenneth A. Younge
References
----------------------------
- Okasa, Gabriel, and Kenneth A. Younge. “Sample Fit Reliability.”
arXiv preprint arXiv:2209.06631. 2022.
- Seabold, Skipper, and Josef Perktold. “statsmodels: Econometric and
statistical modeling with python.” Proceedings of the 9th Python in Science
Conference. 2010.
"""
from samplefit.Reliability import SFR
from samplefit.Reliability import SFRFitResults
from samplefit.Reliability import SFRAnnealResults
from samplefit.Reliability import SFRScoreResults
__all__ = ["SFR", "SFRFitResults", "SFRAnnealResults", "SFRScoreResults"]
__version__ = "0.3.1.9000"
__module__ = 'samplefit'
__author__ = "Gabriel Okasa & Kenneth A. Younge"
__copyright__ = "Copyright (c) 2022, Gabriel Okasa & Kenneth A. Younge"
__license__ = "MIT License"Sub-modules
samplefit.Reliability-
samplefit …
Classes
class SFR (linear_model=None, n_samples=1000, min_samples=None, loss=None, n_jobs=-1, random_state=None)-
Sample Fit Reliability class labeled
SFR. Initializes parameters for sample fit.Parameters
linear_model:statsmodels class- Linear model specified via statsmodels OLS or GLM class.
n_samples:int- The number of sub-samples in the re-sampling procedure. The default is 1000.
min_samples:int, floatorNoneType- Minimum number of observations for each sub-sample, i.e. number of observations to draw from the data without replacement. If integer supplied, exact number of observation is sampled. If float, share of full sample is considered (rounded up). If None, the minimum number of observations to estimate the model is selected, i.e p+1 (reccommended), where p is number of model parameters. The default is None.
loss:strorlambda function- Loss function for evaluation of the estimation errors. Loss must be either 'absolute_error' (reccommended) or 'squared_error'. For a user defined loss function, user can directly supply own lambda function of type: 'lambda y, yhat:'. Default is 'absolute_error'.
n_jobs:intorNoneType-
The number of parallel jobs to be used for multithreading in
.fit(),.score()and.anneal(). Followsjoblibsemantics:n_jobs=-1means all - 1 available cpu physical cores.n_jobs=Noneandn_jobs=1means no parallelism.
The default is -1.
random_state:int, NoneTypeornumpy.random.RandomState object- Random seed used to initialize the pseudo-random number
generator. See
numpydocumentation for details. If None specified, 0 is used. The default is None.
Returns
Initializes SFR class. Following methods are available:
.fit(), .score() and .anneal().
Notes
SFRincludes methods to.fit(),.score()and.anneal().For further details, see examples below.
Examples
# import libraries import samplefit as sf import statsmodels.api as sm # get data boston = sm.datasets.get_rdataset("Boston", "MASS") Y = boston.data['crim'] # per capita crime rate X = boston.data['lstat'] # % lower status population X = sm.add_constant(X) # assess model fit model = sm.OLS(endog=Y, exog=X) model_fit = model.fit() model_fit.summary() # assess sample fit sample = sf.SFR(linear_model=model) sample_fit = sample.fit() sample_fit.summary() # assess sample sensitivity sample_annealing = sample.anneal() sample_annealing.plot() # assess sample reliability sample_scores = sample.score() sample_scores.plot()Expand source code
class SFR(BaseSFR): """ Sample Fit Reliability class labeled `SFR()`. Initializes parameters for sample fit. Parameters ---------- linear_model : statsmodels class Linear model specified via statsmodels OLS or GLM class. n_samples : int The number of sub-samples in the re-sampling procedure. The default is 1000. min_samples : int, float or NoneType Minimum number of observations for each sub-sample, i.e. number of observations to draw from the data without replacement. If integer supplied, exact number of observation is sampled. If float, share of full sample is considered (rounded up). If None, the minimum number of observations to estimate the model is selected, i.e p+1 (reccommended), where p is number of model parameters. The default is None. loss : str or lambda function Loss function for evaluation of the estimation errors. Loss must be either 'absolute_error' (reccommended) or 'squared_error'. For a user defined loss function, user can directly supply own lambda function of type: 'lambda y, yhat:'. Default is 'absolute_error'. n_jobs : int or NoneType The number of parallel jobs to be used for multithreading in [`.fit()`](#samplefit.Reliability.SFR.fit), [`.score()`](#samplefit.Reliability.SFR.score) and [`.anneal()`](#samplefit.Reliability.SFR.anneal). Follows [`joblib`](https://joblib.readthedocs.io){:target="_blank"} semantics: - `n_jobs=-1` means all - 1 available cpu physical cores. - `n_jobs=None` and `n_jobs=1` means no parallelism. The default is -1. random_state : int, NoneType or numpy.random.RandomState object Random seed used to initialize the pseudo-random number generator. See [`numpy` documentation](https://numpy.org/doc/stable/reference/random/legacy.html){:target="_blank"} for details. If None specified, 0 is used. The default is None. Returns ------- Initializes SFR class. Following methods are available: .fit(), .score() and .anneal(). Notes ----- `SFR()` includes methods to [`.fit()`](#samplefit.Reliability.SFR.fit), [`.score()`](#samplefit.Reliability.SFR.score) and [`.anneal()`](#samplefit.Reliability.SFR.anneal). For further details, see examples below. Examples -------- ```py # import libraries import samplefit as sf import statsmodels.api as sm # get data boston = sm.datasets.get_rdataset("Boston", "MASS") Y = boston.data['crim'] # per capita crime rate X = boston.data['lstat'] # % lower status population X = sm.add_constant(X) # assess model fit model = sm.OLS(endog=Y, exog=X) model_fit = model.fit() model_fit.summary() # assess sample fit sample = sf.SFR(linear_model=model) sample_fit = sample.fit() sample_fit.summary() # assess sample sensitivity sample_annealing = sample.anneal() sample_annealing.plot() # assess sample reliability sample_scores = sample.score() sample_scores.plot() ``` """ # define init function def __init__(self, linear_model=None, n_samples=1000, min_samples=None, loss=None, n_jobs=-1, random_state=None): # access inherited methods super().__init__( linear_model=linear_model, n_samples=n_samples, min_samples=min_samples, loss=loss, n_jobs=n_jobs, random_state=random_state ) def fit(self, weights=None, n_boot=None): """ Sample fit based on the reliability scores via the SFR algorithm. Parameters ---------- weights : array-like of shape (n_obs, 1) or NoneType An array of weights for weighted regression. If None, squared reliability scores will be used as weights as a default. Note, that if bootstrapping is used for inference, the estimation of user-supplied weights is not reflected. Default is None. n_boot : int or NoneType Number of bootstrap replications for inference. If None specified, asymptotic approximation is used for inference instead. For valid inference, bootstrapping is recommended. Note that bootstrapping requires longer computation time. Default is None. Returns ------- Results of class SFRFitResults. Following methods are available: .summary(), .conf_int() and .predict(). Notes ----- [`.fit()`](#samplefit.Reliability.SFR.fit) estimates the reliability scores via the SFR algorithm in the first step and estimates weighted regression in the second step, with the squared reliability scores as weights if not specified otherwise. Examples -------- ```py # import libraries import samplefit as sf import statsmodels.api as sm # get data boston = sm.datasets.get_rdataset("Boston", "MASS") Y = boston.data['crim'] # per capita crime rate X = boston.data['lstat'] # % lower status population X = sm.add_constant(X) # specify model model = sm.OLS(endog=Y, exog=X) # specify sample sample = sf.SFR(linear_model=model) # sample fit with defaults sample_fit = sample.fit() # sample fit with bootstrapping sample_fit = sample.fit(n_boot=1000) # get summary of sample fit sample_fit.summary() # get confidence intervals ci_low, ci_up = sample_fit.conf_int() # get predictions (in-sample) preds = sample_fit.predict() ``` """ return super().fit( weights=weights, n_boot=n_boot ) def score(self): """ Estimation of reliability scores via the SFR algorithm. Parameters ---------- None. Returns ------- Results of class SFRScoreResults. Following methods are available: .plot(). Notes ----- [`.score()`](#samplefit.Reliability.SFR.score) estimates the reliability scores via the SFR algorithm. Each observation is scored for the reliability with 0 being the most unreliable observation and with 1 being the most reliable observation within a sample. Examples -------- ```py # import libraries import samplefit as sf import statsmodels.api as sm # get data boston = sm.datasets.get_rdataset("Boston", "MASS") Y = boston.data['crim'] # per capita crime rate X = boston.data['lstat'] # % lower status population X = sm.add_constant(X) # specify model model = sm.OLS(endog=Y, exog=X) # specify sample sample = sf.SFR(linear_model=model) # score reliability sample_scores = sample.score() # extract reliability scores scores = sample_scores.scores # plot reliability scores sample_scores.plot() ``` """ return super().score() def anneal(self, share=0.1, n_boot=None): """ Sample annealing based on the reliability scores via the SFR algorithm. Parameters ---------- share : float or NoneType Share of sample that gets annealed based on the most unreliable observations. Default is 0.1. n_boot : int or NoneType Number of bootstrap replications for inference. If None specified, asymptotic approximation is used for inference instead. For valid inference, bootstrapping is recommended. Note that bootstrapping requires longer computation time. Default is None. Returns ------- Results of class SFRAnnealResults. Following methods are available: .conf_int() and .plot(). Notes ----- [`.anneal()`](#samplefit.Reliability.SFR.anneal) re-estimates the model while sequentially dropping the most unreliable observations. Such annealing procedure helps to assess the sample sensitivity and detect how much the parameters depend on particularly unreliable observations. Examples -------- ```py # import libraries import samplefit as sf import statsmodels.api as sm # get data boston = sm.datasets.get_rdataset("Boston", "MASS") Y = boston.data['crim'] # per capita crime rate X = boston.data['lstat'] # % lower status population X = sm.add_constant(X) # specify model model = sm.OLS(endog=Y, exog=X) # specify sample sample = sf.SFR(linear_model=model) # sample annealing with defaults sample_annealing = sample.anneal() # sample annealing with specified share sample_annealing = sample.anneal(share=0.1) # sample annealing with bootstrapping sample_annealing = sample.anneal(n_boot=1000) # get confidence intervals ci_low, ci_up = sample_annealing.conf_int() # get annealing plot sample_annealing.plot() ``` """ return super().anneal( share=share, n_boot=n_boot )Ancestors
- samplefit._BaseReliability.BaseSFR
Methods
def anneal(self, share=0.1, n_boot=None)-
Sample annealing based on the reliability scores via the SFR algorithm.
Parameters
share:floatorNoneType- Share of sample that gets annealed based on the most unreliable observations. Default is 0.1.
n_boot:intorNoneType- Number of bootstrap replications for inference. If None specified, asymptotic approximation is used for inference instead. For valid inference, bootstrapping is recommended. Note that bootstrapping requires longer computation time. Default is None.
Returns
Resultsofclass SFRAnnealResults. Following methods are available:
.conf_int() and .plot().
Notes
.anneal()re-estimates the model while sequentially dropping the most unreliable observations. Such annealing procedure helps to assess the sample sensitivity and detect how much the parameters depend on particularly unreliable observations.Examples
# import libraries import samplefit as sf import statsmodels.api as sm # get data boston = sm.datasets.get_rdataset("Boston", "MASS") Y = boston.data['crim'] # per capita crime rate X = boston.data['lstat'] # % lower status population X = sm.add_constant(X) # specify model model = sm.OLS(endog=Y, exog=X) # specify sample sample = sf.SFR(linear_model=model) # sample annealing with defaults sample_annealing = sample.anneal() # sample annealing with specified share sample_annealing = sample.anneal(share=0.1) # sample annealing with bootstrapping sample_annealing = sample.anneal(n_boot=1000) # get confidence intervals ci_low, ci_up = sample_annealing.conf_int() # get annealing plot sample_annealing.plot()Expand source code
def anneal(self, share=0.1, n_boot=None): """ Sample annealing based on the reliability scores via the SFR algorithm. Parameters ---------- share : float or NoneType Share of sample that gets annealed based on the most unreliable observations. Default is 0.1. n_boot : int or NoneType Number of bootstrap replications for inference. If None specified, asymptotic approximation is used for inference instead. For valid inference, bootstrapping is recommended. Note that bootstrapping requires longer computation time. Default is None. Returns ------- Results of class SFRAnnealResults. Following methods are available: .conf_int() and .plot(). Notes ----- [`.anneal()`](#samplefit.Reliability.SFR.anneal) re-estimates the model while sequentially dropping the most unreliable observations. Such annealing procedure helps to assess the sample sensitivity and detect how much the parameters depend on particularly unreliable observations. Examples -------- ```py # import libraries import samplefit as sf import statsmodels.api as sm # get data boston = sm.datasets.get_rdataset("Boston", "MASS") Y = boston.data['crim'] # per capita crime rate X = boston.data['lstat'] # % lower status population X = sm.add_constant(X) # specify model model = sm.OLS(endog=Y, exog=X) # specify sample sample = sf.SFR(linear_model=model) # sample annealing with defaults sample_annealing = sample.anneal() # sample annealing with specified share sample_annealing = sample.anneal(share=0.1) # sample annealing with bootstrapping sample_annealing = sample.anneal(n_boot=1000) # get confidence intervals ci_low, ci_up = sample_annealing.conf_int() # get annealing plot sample_annealing.plot() ``` """ return super().anneal( share=share, n_boot=n_boot ) def fit(self, weights=None, n_boot=None)-
Sample fit based on the reliability scores via the SFR algorithm.
Parameters
weights:array-likeofshape (n_obs, 1)orNoneType- An array of weights for weighted regression. If None, squared reliability scores will be used as weights as a default. Note, that if bootstrapping is used for inference, the estimation of user-supplied weights is not reflected. Default is None.
n_boot:intorNoneType- Number of bootstrap replications for inference. If None specified, asymptotic approximation is used for inference instead. For valid inference, bootstrapping is recommended. Note that bootstrapping requires longer computation time. Default is None.
Returns
Resultsofclass SFRFitResults. Following methods are available:
.summary(), .conf_int() and .predict().
Notes
.fit()estimates the reliability scores via the SFR algorithm in the first step and estimates weighted regression in the second step, with the squared reliability scores as weights if not specified otherwise.Examples
# import libraries import samplefit as sf import statsmodels.api as sm # get data boston = sm.datasets.get_rdataset("Boston", "MASS") Y = boston.data['crim'] # per capita crime rate X = boston.data['lstat'] # % lower status population X = sm.add_constant(X) # specify model model = sm.OLS(endog=Y, exog=X) # specify sample sample = sf.SFR(linear_model=model) # sample fit with defaults sample_fit = sample.fit() # sample fit with bootstrapping sample_fit = sample.fit(n_boot=1000) # get summary of sample fit sample_fit.summary() # get confidence intervals ci_low, ci_up = sample_fit.conf_int() # get predictions (in-sample) preds = sample_fit.predict()Expand source code
def fit(self, weights=None, n_boot=None): """ Sample fit based on the reliability scores via the SFR algorithm. Parameters ---------- weights : array-like of shape (n_obs, 1) or NoneType An array of weights for weighted regression. If None, squared reliability scores will be used as weights as a default. Note, that if bootstrapping is used for inference, the estimation of user-supplied weights is not reflected. Default is None. n_boot : int or NoneType Number of bootstrap replications for inference. If None specified, asymptotic approximation is used for inference instead. For valid inference, bootstrapping is recommended. Note that bootstrapping requires longer computation time. Default is None. Returns ------- Results of class SFRFitResults. Following methods are available: .summary(), .conf_int() and .predict(). Notes ----- [`.fit()`](#samplefit.Reliability.SFR.fit) estimates the reliability scores via the SFR algorithm in the first step and estimates weighted regression in the second step, with the squared reliability scores as weights if not specified otherwise. Examples -------- ```py # import libraries import samplefit as sf import statsmodels.api as sm # get data boston = sm.datasets.get_rdataset("Boston", "MASS") Y = boston.data['crim'] # per capita crime rate X = boston.data['lstat'] # % lower status population X = sm.add_constant(X) # specify model model = sm.OLS(endog=Y, exog=X) # specify sample sample = sf.SFR(linear_model=model) # sample fit with defaults sample_fit = sample.fit() # sample fit with bootstrapping sample_fit = sample.fit(n_boot=1000) # get summary of sample fit sample_fit.summary() # get confidence intervals ci_low, ci_up = sample_fit.conf_int() # get predictions (in-sample) preds = sample_fit.predict() ``` """ return super().fit( weights=weights, n_boot=n_boot ) def score(self)-
Estimation of reliability scores via the SFR algorithm.
Parameters
None.
Returns
Resultsofclass SFRScoreResults. Following methods are available:
.plot().
Notes
.score()estimates the reliability scores via the SFR algorithm. Each observation is scored for the reliability with 0 being the most unreliable observation and with 1 being the most reliable observation within a sample.Examples
# import libraries import samplefit as sf import statsmodels.api as sm # get data boston = sm.datasets.get_rdataset("Boston", "MASS") Y = boston.data['crim'] # per capita crime rate X = boston.data['lstat'] # % lower status population X = sm.add_constant(X) # specify model model = sm.OLS(endog=Y, exog=X) # specify sample sample = sf.SFR(linear_model=model) # score reliability sample_scores = sample.score() # extract reliability scores scores = sample_scores.scores # plot reliability scores sample_scores.plot()Expand source code
def score(self): """ Estimation of reliability scores via the SFR algorithm. Parameters ---------- None. Returns ------- Results of class SFRScoreResults. Following methods are available: .plot(). Notes ----- [`.score()`](#samplefit.Reliability.SFR.score) estimates the reliability scores via the SFR algorithm. Each observation is scored for the reliability with 0 being the most unreliable observation and with 1 being the most reliable observation within a sample. Examples -------- ```py # import libraries import samplefit as sf import statsmodels.api as sm # get data boston = sm.datasets.get_rdataset("Boston", "MASS") Y = boston.data['crim'] # per capita crime rate X = boston.data['lstat'] # % lower status population X = sm.add_constant(X) # specify model model = sm.OLS(endog=Y, exog=X) # specify sample sample = sf.SFR(linear_model=model) # score reliability sample_scores = sample.score() # extract reliability scores scores = sample_scores.scores # plot reliability scores sample_scores.plot() ``` """ return super().score()
class SFRAnnealResults (sample=None, params=None, params_boot=None, stand_err=None, drop_idx=None)-
Annealing results class labeled
SFRAnnealResults. Initializes output of SFR.anneal().Expand source code
class SFRAnnealResults(BaseSFRAnnealResults): """ Annealing results class labeled `SFRAnnealResults()`. Initializes output of SFR.anneal(). """ # define init function def __init__(self, sample=None, params=None, params_boot=None, stand_err=None, drop_idx=None ): # access inherited methods super().__init__( sample=sample, params=params, params_boot=params_boot, stand_err=stand_err, drop_idx=drop_idx ) def plot(self, yname=None, xname=None, title=None, alpha=0.05, percentile=False, color=None, path=None, figsize=None, ylim=None, xlabel=None, dpi=None, fname=None): """ Plot the Annealing based on the reliability scores from the SFR algorithm. Parameters ---------- yname : str or NoneType Name of the y axis. Default is 'Effect'. xname : list, tuple, str or NoneType Name or list of names of the exog variables for which parameter an annealing plot should be constructed. Must be one of the exog variable names. If not supplied annealing plots for all parameters are constructed. Default are the supplied exog names. title : str or NoneType Title for the annealing plot. Default is 'SFR: Annealing'. alpha : float or NoneType Confidence level alpha. Default is 0.05. percentile : bool Percentile method for confidence intervals based on bootstrapping. If bootstrapping has not been used for annealing, it is ignored. Default is False. color : str or NoneType Color used for the confidence interval. Must be one of the matplotlib supported colors. Default is grey. path : str or NoneType Valid path to save the plot. If None, plot is not saved. Default is None. figsize : tuple or NoneType Tuple of x and y axis size for matplotlib figsize argument. Default is (10,5). ylim : tuple, list or NoneType Tuple of upper and lower limits of y axis. Default is automatic. xlabel : str or NoneType Label for the x axis for the exog variable. Default is 'xname'. dpi : float, int or NoneType The resolution for matplotlib scatter plot. Default is 100. fname : str or NoneType Valid figure name to save the plot. If None, generic name is used. Default is None. Returns ------- Dictionary of matplotlib figures and axes. Prints annealing plots. Notes ----- [`.plot()`](#samplefit.Reliability.SFRAnnealResults.plot) produces an annealing plot for assessment of sample fit sensitivity, together with parameters and confidence intervals. Examples -------- ```py # import libraries import samplefit as sf import statsmodels.api as sm # get data boston = sm.datasets.get_rdataset("Boston", "MASS") Y = boston.data['crim'] # per capita crime rate X = boston.data['lstat'] # % lower status population X = sm.add_constant(X) # specify model model = sm.OLS(endog=Y, exog=X) # specify sample sample = sf.SFR(linear_model=model) # sample annealing sample_annealing = sample.anneal() # default annealing plot sample_annealing.plot() # custom annealing sample_annealing.plot(title='My Title') ``` """ return super().plot( yname=yname, xname=xname, title=title, alpha=alpha, percentile=percentile, color=color, path=path, figsize=figsize, ylim=ylim, xlabel=xlabel, dpi=dpi, fname=fname ) def conf_int(self, alpha=0.05, percentile=False): """ Confidence intervals based on the annealing via the SFR algorithm. Parameters ---------- alpha : float or NoneType Confidence level alpha. Default is 0.05. percentile : bool Percentile method for confidence intervals based on bootstrapping. If bootstrapping has not been used for annealing, it is ignored. Default is False. Returns ------- Tuple of arrays of confidence bounds, lower and upper. Notes ----- [`.conf_int()`](#samplefit.Reliability.SFRAnnealResults.conf_int) constructs confidence intervals for estimated paramaters. If annealed without bootstrapping, asymptotic approximations are used. If annealed with bootstrapping, the standard deviation of bootstrapped parameters is used for standard error approximation. If percentile=True, the percentile method is used instead of standard deviation. Examples -------- ```py # import libraries import samplefit as sf import statsmodels.api as sm # get data boston = sm.datasets.get_rdataset("Boston", "MASS") Y = boston.data['crim'] # per capita crime rate X = boston.data['lstat'] # % lower status population X = sm.add_constant(X) # specify model model = sm.OLS(endog=Y, exog=X) # specify sample sample = sf.SFR(linear_model=model) # sample annealing sample_annealing = sample.anneal() # compute confidence intervals with default settings ci_low, ci_up = sample_annealing.conf_int() # compute confidence intervals with custom alpha ci_low, ci_up = sample_fit.conf_int(alpha=0.1) ``` """ return super().conf_int( alpha=alpha, percentile=percentile )Ancestors
- samplefit._BaseResultsReliability.BaseSFRAnnealResults
Methods
def conf_int(self, alpha=0.05, percentile=False)-
Confidence intervals based on the annealing via the SFR algorithm.
Parameters
alpha:floatorNoneType- Confidence level alpha. Default is 0.05.
percentile:bool- Percentile method for confidence intervals based on bootstrapping. If bootstrapping has not been used for annealing, it is ignored. Default is False.
Returns
Tuple of arrays of confidence bounds, lower and upper.
Notes
.conf_int()constructs confidence intervals for estimated paramaters. If annealed without bootstrapping, asymptotic approximations are used. If annealed with bootstrapping, the standard deviation of bootstrapped parameters is used for standard error approximation. If percentile=True, the percentile method is used instead of standard deviation.Examples
# import libraries import samplefit as sf import statsmodels.api as sm # get data boston = sm.datasets.get_rdataset("Boston", "MASS") Y = boston.data['crim'] # per capita crime rate X = boston.data['lstat'] # % lower status population X = sm.add_constant(X) # specify model model = sm.OLS(endog=Y, exog=X) # specify sample sample = sf.SFR(linear_model=model) # sample annealing sample_annealing = sample.anneal() # compute confidence intervals with default settings ci_low, ci_up = sample_annealing.conf_int() # compute confidence intervals with custom alpha ci_low, ci_up = sample_fit.conf_int(alpha=0.1)Expand source code
def conf_int(self, alpha=0.05, percentile=False): """ Confidence intervals based on the annealing via the SFR algorithm. Parameters ---------- alpha : float or NoneType Confidence level alpha. Default is 0.05. percentile : bool Percentile method for confidence intervals based on bootstrapping. If bootstrapping has not been used for annealing, it is ignored. Default is False. Returns ------- Tuple of arrays of confidence bounds, lower and upper. Notes ----- [`.conf_int()`](#samplefit.Reliability.SFRAnnealResults.conf_int) constructs confidence intervals for estimated paramaters. If annealed without bootstrapping, asymptotic approximations are used. If annealed with bootstrapping, the standard deviation of bootstrapped parameters is used for standard error approximation. If percentile=True, the percentile method is used instead of standard deviation. Examples -------- ```py # import libraries import samplefit as sf import statsmodels.api as sm # get data boston = sm.datasets.get_rdataset("Boston", "MASS") Y = boston.data['crim'] # per capita crime rate X = boston.data['lstat'] # % lower status population X = sm.add_constant(X) # specify model model = sm.OLS(endog=Y, exog=X) # specify sample sample = sf.SFR(linear_model=model) # sample annealing sample_annealing = sample.anneal() # compute confidence intervals with default settings ci_low, ci_up = sample_annealing.conf_int() # compute confidence intervals with custom alpha ci_low, ci_up = sample_fit.conf_int(alpha=0.1) ``` """ return super().conf_int( alpha=alpha, percentile=percentile ) def plot(self, yname=None, xname=None, title=None, alpha=0.05, percentile=False, color=None, path=None, figsize=None, ylim=None, xlabel=None, dpi=None, fname=None)-
Plot the Annealing based on the reliability scores from the SFR algorithm.
Parameters
yname:strorNoneType- Name of the y axis. Default is 'Effect'.
xname:list, tuple, strorNoneType- Name or list of names of the exog variables for which parameter an annealing plot should be constructed. Must be one of the exog variable names. If not supplied annealing plots for all parameters are constructed. Default are the supplied exog names.
title:strorNoneType- Title for the annealing plot. Default is 'SFR: Annealing'.
alpha:floatorNoneType- Confidence level alpha. Default is 0.05.
percentile:bool- Percentile method for confidence intervals based on bootstrapping. If bootstrapping has not been used for annealing, it is ignored. Default is False.
color:strorNoneType- Color used for the confidence interval. Must be one of the matplotlib supported colors. Default is grey.
path:strorNoneType- Valid path to save the plot. If None, plot is not saved. Default is None.
figsize:tupleorNoneType- Tuple of x and y axis size for matplotlib figsize argument. Default is (10,5).
ylim:tuple, listorNoneType- Tuple of upper and lower limits of y axis. Default is automatic.
xlabel:strorNoneType- Label for the x axis for the exog variable. Default is 'xname'.
dpi:float, intorNoneType- The resolution for matplotlib scatter plot. Default is 100.
fname:strorNoneType- Valid figure name to save the plot. If None, generic name is used. Default is None.
Returns
Dictionary of matplotlib figures and axes. Prints annealing plots.
Notes
.plot()produces an annealing plot for assessment of sample fit sensitivity, together with parameters and confidence intervals.Examples
# import libraries import samplefit as sf import statsmodels.api as sm # get data boston = sm.datasets.get_rdataset("Boston", "MASS") Y = boston.data['crim'] # per capita crime rate X = boston.data['lstat'] # % lower status population X = sm.add_constant(X) # specify model model = sm.OLS(endog=Y, exog=X) # specify sample sample = sf.SFR(linear_model=model) # sample annealing sample_annealing = sample.anneal() # default annealing plot sample_annealing.plot() # custom annealing sample_annealing.plot(title='My Title')Expand source code
def plot(self, yname=None, xname=None, title=None, alpha=0.05, percentile=False, color=None, path=None, figsize=None, ylim=None, xlabel=None, dpi=None, fname=None): """ Plot the Annealing based on the reliability scores from the SFR algorithm. Parameters ---------- yname : str or NoneType Name of the y axis. Default is 'Effect'. xname : list, tuple, str or NoneType Name or list of names of the exog variables for which parameter an annealing plot should be constructed. Must be one of the exog variable names. If not supplied annealing plots for all parameters are constructed. Default are the supplied exog names. title : str or NoneType Title for the annealing plot. Default is 'SFR: Annealing'. alpha : float or NoneType Confidence level alpha. Default is 0.05. percentile : bool Percentile method for confidence intervals based on bootstrapping. If bootstrapping has not been used for annealing, it is ignored. Default is False. color : str or NoneType Color used for the confidence interval. Must be one of the matplotlib supported colors. Default is grey. path : str or NoneType Valid path to save the plot. If None, plot is not saved. Default is None. figsize : tuple or NoneType Tuple of x and y axis size for matplotlib figsize argument. Default is (10,5). ylim : tuple, list or NoneType Tuple of upper and lower limits of y axis. Default is automatic. xlabel : str or NoneType Label for the x axis for the exog variable. Default is 'xname'. dpi : float, int or NoneType The resolution for matplotlib scatter plot. Default is 100. fname : str or NoneType Valid figure name to save the plot. If None, generic name is used. Default is None. Returns ------- Dictionary of matplotlib figures and axes. Prints annealing plots. Notes ----- [`.plot()`](#samplefit.Reliability.SFRAnnealResults.plot) produces an annealing plot for assessment of sample fit sensitivity, together with parameters and confidence intervals. Examples -------- ```py # import libraries import samplefit as sf import statsmodels.api as sm # get data boston = sm.datasets.get_rdataset("Boston", "MASS") Y = boston.data['crim'] # per capita crime rate X = boston.data['lstat'] # % lower status population X = sm.add_constant(X) # specify model model = sm.OLS(endog=Y, exog=X) # specify sample sample = sf.SFR(linear_model=model) # sample annealing sample_annealing = sample.anneal() # default annealing plot sample_annealing.plot() # custom annealing sample_annealing.plot(title='My Title') ``` """ return super().plot( yname=yname, xname=xname, title=title, alpha=alpha, percentile=percentile, color=color, path=path, figsize=figsize, ylim=ylim, xlabel=xlabel, dpi=dpi, fname=fname )
class SFRFitResults (sample=None, params=None, params_boot=None, stand_err=None, fittedvalues=None)-
Fit Results class labeled
SFRFitResults. Initializes output of SFR.fit().Expand source code
class SFRFitResults(BaseSFRFitResults): """ Fit Results class labeled `SFRFitResults()`. Initializes output of SFR.fit(). """ # define init function def __init__(self, sample=None, params=None, params_boot=None, stand_err=None, fittedvalues=None ): # access inherited methods super().__init__( sample=sample, params=params, params_boot=params_boot, stand_err=stand_err, fittedvalues=fittedvalues ) def predict(self, params=None, exog=None): """ Predict outcomes based on the sample fit via the SFR algorithm. Parameters ---------- params : array-like or NoneType Array of parameters to predict with. If None supplied, the estimated parameters from the sample fit will be used. Default is None. exog : array-like or NoneType Matrix of features/covariates for which the outcomes should be predicted (out-of-sample). Column dimensions must be identical to the training data supplied to the statsmodels model class. If None supplied, in-sample predictions (fitted values) are returned. Default is None. Returns ------- Array of predictions. Notes ----- [`.predict()`](#samplefit.Reliability.SFRFitResults.predict) constructs predictions for outcome variable based on the estimated parameters. Predictions are based on the parameters of weighted fit. If no new values for exogeneous variables are supplied, fitted values are returned. Examples -------- ```py # import libraries import samplefit as sf import statsmodels.api as sm # get data boston = sm.datasets.get_rdataset("Boston", "MASS") Y = boston.data['crim'] # per capita crime rate X = boston.data['lstat'] # % lower status population X = sm.add_constant(X) # specify model model = sm.OLS(endog=Y, exog=X) # specify sample sample = sf.SFR(linear_model=model) # sample fit sample_fit = sample.fit() # predict in-sample preds = sample_fit.predict() # predict out-of-sample preds = sample_fit.predict(exog=X[0, :]) ``` """ return super().predict( params=params, exog=exog ) def conf_int(self, alpha=0.05, percentile=False): """ Confidence intervals based on the sample fit via the SFR algorithm. Parameters ---------- alpha : float or NoneType Confidence level alpha. Default is 0.05. percentile : bool Percentile method for confidence intervals based on bootstrapping. If bootstrapping has not been used for fitting, it is ignored. Default is False. Returns ------- Tuple of arrays of confidence bounds, lower and upper. Notes ----- [`.conf_int()`](#samplefit.Reliability.SFRFitResults.conf_int) constructs confidence intervals for estimated paramaters. If fitted without bootstrapping, asymptotic approximations are used. If fitted with bootstrapping, the standard deviation of bootstrapped parameters is used for standard error approximation. If percentile=True, the percentile method is used instead of standard deviation. Examples -------- ```py # import libraries import samplefit as sf import statsmodels.api as sm # get data boston = sm.datasets.get_rdataset("Boston", "MASS") Y = boston.data['crim'] # per capita crime rate X = boston.data['lstat'] # % lower status population X = sm.add_constant(X) # specify model model = sm.OLS(endog=Y, exog=X) # specify sample sample = sf.SFR(linear_model=model) # sample fit sample_fit = sample.fit() # compute confidence intervals with default settings ci_low, ci_up = sample_fit.conf_int() # compute confidence intervals with custom alpha ci_low, ci_up = sample_fit.conf_int(alpha=0.1) ``` """ return super().conf_int( alpha=alpha, percentile=percentile ) def summary(self, yname=None, xname=None, title=None, alpha=0.05, percentile=False, get_table=False, verbose=True): """ Summary of the sample fit via the SFR algorithm. Parameters ---------- yname : str or NoneType Name of the endog variable. Default is 'y'. xname : list, tuple or NoneType List of name of the exog variables. Must have the same dimension as exog columns. Default are the supplied exog names. title : str or NoneType Title for the summary table. Default is 'SFR: Fitting'. alpha : float or NoneType Confidence level alpha. Default is 0.05. percentile : bool Percentile method for confidence intervals based on bootstrapping. If bootstrapping has not been used for fitting, it is ignored. Default is False. get_table : bool If a summary table should be returned or not. If True, a pandas DataFrame with estimation results is returned. Default is False. verbose : bool If a summary table should be printed to console or not. Default is True. Returns ------- None. Prints summary table. Notes ----- [`.summary()`](#samplefit.Reliability.SFRFitResults.summary) produces a summary table including information on sample fit as well as model fit together with parameters, standard errors, t-values, p-values and confidence intervals. Examples -------- ```py # import libraries import samplefit as sf import statsmodels.api as sm # get data boston = sm.datasets.get_rdataset("Boston", "MASS") Y = boston.data['crim'] # per capita crime rate X = boston.data['lstat'] # % lower status population X = sm.add_constant(X) # specify model model = sm.OLS(endog=Y, exog=X) # specify sample sample = sf.SFR(linear_model=model) # sample fit sample_fit = sample.fit() # default summary sample_fit.summary() # custom summary title sample_fit.summary(title='My Title') ``` """ return super().summary( yname=yname, xname=xname, title=title, alpha=alpha, percentile=percentile, get_table=get_table, verbose=verbose )Ancestors
- samplefit._BaseResultsReliability.BaseSFRFitResults
Methods
def conf_int(self, alpha=0.05, percentile=False)-
Confidence intervals based on the sample fit via the SFR algorithm.
Parameters
alpha:floatorNoneType- Confidence level alpha. Default is 0.05.
percentile:bool- Percentile method for confidence intervals based on bootstrapping. If bootstrapping has not been used for fitting, it is ignored. Default is False.
Returns
Tuple of arrays of confidence bounds, lower and upper.
Notes
.conf_int()constructs confidence intervals for estimated paramaters. If fitted without bootstrapping, asymptotic approximations are used. If fitted with bootstrapping, the standard deviation of bootstrapped parameters is used for standard error approximation. If percentile=True, the percentile method is used instead of standard deviation.Examples
# import libraries import samplefit as sf import statsmodels.api as sm # get data boston = sm.datasets.get_rdataset("Boston", "MASS") Y = boston.data['crim'] # per capita crime rate X = boston.data['lstat'] # % lower status population X = sm.add_constant(X) # specify model model = sm.OLS(endog=Y, exog=X) # specify sample sample = sf.SFR(linear_model=model) # sample fit sample_fit = sample.fit() # compute confidence intervals with default settings ci_low, ci_up = sample_fit.conf_int() # compute confidence intervals with custom alpha ci_low, ci_up = sample_fit.conf_int(alpha=0.1)Expand source code
def conf_int(self, alpha=0.05, percentile=False): """ Confidence intervals based on the sample fit via the SFR algorithm. Parameters ---------- alpha : float or NoneType Confidence level alpha. Default is 0.05. percentile : bool Percentile method for confidence intervals based on bootstrapping. If bootstrapping has not been used for fitting, it is ignored. Default is False. Returns ------- Tuple of arrays of confidence bounds, lower and upper. Notes ----- [`.conf_int()`](#samplefit.Reliability.SFRFitResults.conf_int) constructs confidence intervals for estimated paramaters. If fitted without bootstrapping, asymptotic approximations are used. If fitted with bootstrapping, the standard deviation of bootstrapped parameters is used for standard error approximation. If percentile=True, the percentile method is used instead of standard deviation. Examples -------- ```py # import libraries import samplefit as sf import statsmodels.api as sm # get data boston = sm.datasets.get_rdataset("Boston", "MASS") Y = boston.data['crim'] # per capita crime rate X = boston.data['lstat'] # % lower status population X = sm.add_constant(X) # specify model model = sm.OLS(endog=Y, exog=X) # specify sample sample = sf.SFR(linear_model=model) # sample fit sample_fit = sample.fit() # compute confidence intervals with default settings ci_low, ci_up = sample_fit.conf_int() # compute confidence intervals with custom alpha ci_low, ci_up = sample_fit.conf_int(alpha=0.1) ``` """ return super().conf_int( alpha=alpha, percentile=percentile ) def predict(self, params=None, exog=None)-
Predict outcomes based on the sample fit via the SFR algorithm.
Parameters
params:array-likeorNoneType- Array of parameters to predict with. If None supplied, the estimated parameters from the sample fit will be used. Default is None.
exog:array-likeorNoneType- Matrix of features/covariates for which the outcomes should be predicted (out-of-sample). Column dimensions must be identical to the training data supplied to the statsmodels model class. If None supplied, in-sample predictions (fitted values) are returned. Default is None.
Returns
Array of predictions.
Notes
.predict()constructs predictions for outcome variable based on the estimated parameters. Predictions are based on the parameters of weighted fit. If no new values for exogeneous variables are supplied, fitted values are returned.Examples
# import libraries import samplefit as sf import statsmodels.api as sm # get data boston = sm.datasets.get_rdataset("Boston", "MASS") Y = boston.data['crim'] # per capita crime rate X = boston.data['lstat'] # % lower status population X = sm.add_constant(X) # specify model model = sm.OLS(endog=Y, exog=X) # specify sample sample = sf.SFR(linear_model=model) # sample fit sample_fit = sample.fit() # predict in-sample preds = sample_fit.predict() # predict out-of-sample preds = sample_fit.predict(exog=X[0, :])Expand source code
def predict(self, params=None, exog=None): """ Predict outcomes based on the sample fit via the SFR algorithm. Parameters ---------- params : array-like or NoneType Array of parameters to predict with. If None supplied, the estimated parameters from the sample fit will be used. Default is None. exog : array-like or NoneType Matrix of features/covariates for which the outcomes should be predicted (out-of-sample). Column dimensions must be identical to the training data supplied to the statsmodels model class. If None supplied, in-sample predictions (fitted values) are returned. Default is None. Returns ------- Array of predictions. Notes ----- [`.predict()`](#samplefit.Reliability.SFRFitResults.predict) constructs predictions for outcome variable based on the estimated parameters. Predictions are based on the parameters of weighted fit. If no new values for exogeneous variables are supplied, fitted values are returned. Examples -------- ```py # import libraries import samplefit as sf import statsmodels.api as sm # get data boston = sm.datasets.get_rdataset("Boston", "MASS") Y = boston.data['crim'] # per capita crime rate X = boston.data['lstat'] # % lower status population X = sm.add_constant(X) # specify model model = sm.OLS(endog=Y, exog=X) # specify sample sample = sf.SFR(linear_model=model) # sample fit sample_fit = sample.fit() # predict in-sample preds = sample_fit.predict() # predict out-of-sample preds = sample_fit.predict(exog=X[0, :]) ``` """ return super().predict( params=params, exog=exog ) def summary(self, yname=None, xname=None, title=None, alpha=0.05, percentile=False, get_table=False, verbose=True)-
Summary of the sample fit via the SFR algorithm.
Parameters
yname:strorNoneType- Name of the endog variable. Default is 'y'.
xname:list, tupleorNoneType- List of name of the exog variables. Must have the same dimension as exog columns. Default are the supplied exog names.
title:strorNoneType- Title for the summary table. Default is 'SFR: Fitting'.
alpha:floatorNoneType- Confidence level alpha. Default is 0.05.
percentile:bool- Percentile method for confidence intervals based on bootstrapping. If bootstrapping has not been used for fitting, it is ignored. Default is False.
get_table:bool- If a summary table should be returned or not. If True, a pandas DataFrame with estimation results is returned. Default is False.
verbose:bool- If a summary table should be printed to console or not. Default is True.
Returns
None. Prints summary table.
Notes
.summary()produces a summary table including information on sample fit as well as model fit together with parameters, standard errors, t-values, p-values and confidence intervals.Examples
# import libraries import samplefit as sf import statsmodels.api as sm # get data boston = sm.datasets.get_rdataset("Boston", "MASS") Y = boston.data['crim'] # per capita crime rate X = boston.data['lstat'] # % lower status population X = sm.add_constant(X) # specify model model = sm.OLS(endog=Y, exog=X) # specify sample sample = sf.SFR(linear_model=model) # sample fit sample_fit = sample.fit() # default summary sample_fit.summary() # custom summary title sample_fit.summary(title='My Title')Expand source code
def summary(self, yname=None, xname=None, title=None, alpha=0.05, percentile=False, get_table=False, verbose=True): """ Summary of the sample fit via the SFR algorithm. Parameters ---------- yname : str or NoneType Name of the endog variable. Default is 'y'. xname : list, tuple or NoneType List of name of the exog variables. Must have the same dimension as exog columns. Default are the supplied exog names. title : str or NoneType Title for the summary table. Default is 'SFR: Fitting'. alpha : float or NoneType Confidence level alpha. Default is 0.05. percentile : bool Percentile method for confidence intervals based on bootstrapping. If bootstrapping has not been used for fitting, it is ignored. Default is False. get_table : bool If a summary table should be returned or not. If True, a pandas DataFrame with estimation results is returned. Default is False. verbose : bool If a summary table should be printed to console or not. Default is True. Returns ------- None. Prints summary table. Notes ----- [`.summary()`](#samplefit.Reliability.SFRFitResults.summary) produces a summary table including information on sample fit as well as model fit together with parameters, standard errors, t-values, p-values and confidence intervals. Examples -------- ```py # import libraries import samplefit as sf import statsmodels.api as sm # get data boston = sm.datasets.get_rdataset("Boston", "MASS") Y = boston.data['crim'] # per capita crime rate X = boston.data['lstat'] # % lower status population X = sm.add_constant(X) # specify model model = sm.OLS(endog=Y, exog=X) # specify sample sample = sf.SFR(linear_model=model) # sample fit sample_fit = sample.fit() # default summary sample_fit.summary() # custom summary title sample_fit.summary(title='My Title') ``` """ return super().summary( yname=yname, xname=xname, title=title, alpha=alpha, percentile=percentile, get_table=get_table, verbose=verbose )
class SFRScoreResults (sample=None)-
Scoring results class labeled
SFRScoreResults. Initializes output of SFR.score().Expand source code
class SFRScoreResults(BaseSFRScoreResults): """ Scoring results class labeled `SFRScoreResults()`. Initializes output of SFR.score(). """ # define init function def __init__(self, sample=None ): # access inherited methods super().__init__( sample=sample ) def plot(self, yname=None, xname=None, title=None, cmap=None, path=None, figsize=None, s=None, ylim=None, xlim=None, xlabel=None, dpi=None, fname=None, jitter=False): """ Plot the reliability scores based on the SFR algorithm. Parameters ---------- yname : str or NoneType Name of the endog variable. Default is 'y'. xname : list, tuple, str or NoneType Name or list of names of the exog variables for which parameter an scoring plot should be constructed. Must be one of the exog variable names. If not supplied scoring plots for all parameters are constructed. Default are the supplied exog names. title : str or NoneType Title for the scoring plot. Default is 'SFR: Scoring'. cmap : str or NoneType Color map used for the reliability score. Must be one of the matplotlib supported color maps. Default is 'RdYlGn'. path : str or NoneType Valid path to save the plot. If None, plot is not saved. Default is None. figsize : tuple or NoneType Tuple of x and y axis size for matplotlib figsize argument. Default is (10,5). s : float, int or NoneType The marker size in points**2 as for in matplotlib scatter plot. Default is automatic. ylim : tuple, list or NoneType Tuple of upper and lower limits of y axis. Default is automatic. xlim : tuple, list or NoneType Tuple of upper and lower limits of x axis. Default is automatic. xlabel : str or NoneType Label for the x axis for the exog variable. Default is 'xname'. dpi : float, int or NoneType The resolution for matplotlib scatter plot. Default is 100. fname : str or NoneType Valid figure name to save the plot. If None, generic name is used. Default is None. jitter : bool Logical, if scatterplot should be jittered for categorical features. Note, that this involves random perturbation of the values of features along X axis, fixing seed is thus necessary for reproducibility. Default is False. Returns ------- Dictionary of matplotlib figures and axes. Prints scoring plots. Notes ----- [`.plot()`](#samplefit.Reliability.SFRScoreResults.plot) produces a scoring plot for assessment of sample fit reliability. Examples -------- ```py # import libraries import samplefit as sf import statsmodels.api as sm # get data boston = sm.datasets.get_rdataset("Boston", "MASS") Y = boston.data['crim'] # per capita crime rate X = boston.data['lstat'] # % lower status population X = sm.add_constant(X) # specify model model = sm.OLS(endog=Y, exog=X) # specify sample sample = sf.SFR(linear_model=model) # sample reliability sample_scores = sample.score() # default scoring plot sample_scores.plot() # custom scoring sample_scores.plot(title='My Title') ``` """ return super().plot( yname=yname, xname=xname, title=title, cmap=cmap, path=path, figsize=figsize, s=s, ylim=ylim, xlim=xlim, xlabel=xlabel, dpi=dpi, fname=fname, jitter=jitter )Ancestors
- samplefit._BaseResultsReliability.BaseSFRScoreResults
Methods
def plot(self, yname=None, xname=None, title=None, cmap=None, path=None, figsize=None, s=None, ylim=None, xlim=None, xlabel=None, dpi=None, fname=None, jitter=False)-
Plot the reliability scores based on the SFR algorithm.
Parameters
yname:strorNoneType- Name of the endog variable. Default is 'y'.
xname:list, tuple, strorNoneType- Name or list of names of the exog variables for which parameter an scoring plot should be constructed. Must be one of the exog variable names. If not supplied scoring plots for all parameters are constructed. Default are the supplied exog names.
title:strorNoneType- Title for the scoring plot. Default is 'SFR: Scoring'.
cmap:strorNoneType- Color map used for the reliability score. Must be one of the matplotlib supported color maps. Default is 'RdYlGn'.
path:strorNoneType- Valid path to save the plot. If None, plot is not saved. Default is None.
figsize:tupleorNoneType- Tuple of x and y axis size for matplotlib figsize argument. Default is (10,5).
s:float, intorNoneType- The marker size in points**2 as for in matplotlib scatter plot. Default is automatic.
ylim:tuple, listorNoneType- Tuple of upper and lower limits of y axis. Default is automatic.
xlim:tuple, listorNoneType- Tuple of upper and lower limits of x axis. Default is automatic.
xlabel:strorNoneType- Label for the x axis for the exog variable. Default is 'xname'.
dpi:float, intorNoneType- The resolution for matplotlib scatter plot. Default is 100.
fname:strorNoneType- Valid figure name to save the plot. If None, generic name is used. Default is None.
jitter:bool- Logical, if scatterplot should be jittered for categorical features. Note, that this involves random perturbation of the values of features along X axis, fixing seed is thus necessary for reproducibility. Default is False.
Returns
Dictionary of matplotlib figures and axes. Prints scoring plots.
Notes
.plot()produces a scoring plot for assessment of sample fit reliability.Examples
# import libraries import samplefit as sf import statsmodels.api as sm # get data boston = sm.datasets.get_rdataset("Boston", "MASS") Y = boston.data['crim'] # per capita crime rate X = boston.data['lstat'] # % lower status population X = sm.add_constant(X) # specify model model = sm.OLS(endog=Y, exog=X) # specify sample sample = sf.SFR(linear_model=model) # sample reliability sample_scores = sample.score() # default scoring plot sample_scores.plot() # custom scoring sample_scores.plot(title='My Title')Expand source code
def plot(self, yname=None, xname=None, title=None, cmap=None, path=None, figsize=None, s=None, ylim=None, xlim=None, xlabel=None, dpi=None, fname=None, jitter=False): """ Plot the reliability scores based on the SFR algorithm. Parameters ---------- yname : str or NoneType Name of the endog variable. Default is 'y'. xname : list, tuple, str or NoneType Name or list of names of the exog variables for which parameter an scoring plot should be constructed. Must be one of the exog variable names. If not supplied scoring plots for all parameters are constructed. Default are the supplied exog names. title : str or NoneType Title for the scoring plot. Default is 'SFR: Scoring'. cmap : str or NoneType Color map used for the reliability score. Must be one of the matplotlib supported color maps. Default is 'RdYlGn'. path : str or NoneType Valid path to save the plot. If None, plot is not saved. Default is None. figsize : tuple or NoneType Tuple of x and y axis size for matplotlib figsize argument. Default is (10,5). s : float, int or NoneType The marker size in points**2 as for in matplotlib scatter plot. Default is automatic. ylim : tuple, list or NoneType Tuple of upper and lower limits of y axis. Default is automatic. xlim : tuple, list or NoneType Tuple of upper and lower limits of x axis. Default is automatic. xlabel : str or NoneType Label for the x axis for the exog variable. Default is 'xname'. dpi : float, int or NoneType The resolution for matplotlib scatter plot. Default is 100. fname : str or NoneType Valid figure name to save the plot. If None, generic name is used. Default is None. jitter : bool Logical, if scatterplot should be jittered for categorical features. Note, that this involves random perturbation of the values of features along X axis, fixing seed is thus necessary for reproducibility. Default is False. Returns ------- Dictionary of matplotlib figures and axes. Prints scoring plots. Notes ----- [`.plot()`](#samplefit.Reliability.SFRScoreResults.plot) produces a scoring plot for assessment of sample fit reliability. Examples -------- ```py # import libraries import samplefit as sf import statsmodels.api as sm # get data boston = sm.datasets.get_rdataset("Boston", "MASS") Y = boston.data['crim'] # per capita crime rate X = boston.data['lstat'] # % lower status population X = sm.add_constant(X) # specify model model = sm.OLS(endog=Y, exog=X) # specify sample sample = sf.SFR(linear_model=model) # sample reliability sample_scores = sample.score() # default scoring plot sample_scores.plot() # custom scoring sample_scores.plot(title='My Title') ``` """ return super().plot( yname=yname, xname=xname, title=title, cmap=cmap, path=path, figsize=figsize, s=s, ylim=ylim, xlim=xlim, xlabel=xlabel, dpi=dpi, fname=fname, jitter=jitter )