adding DSZI fitter

README.md

@@ -21,6 +21,7 @@ Detailed documentation and examples are available at [readthedocs](https://relia
 ## Key features
 - Fitting probability distributions to data including right censored data
 - Fitting Weibull mixture models and Weibull Competing risks models
+- Fitting Weibull Defective Subpopulation (DS), Weibull Zero Inflated (ZI) models, and Weibull Defective Subpopulation Zero Inflated (DSZI) models
 - Calculating the probability of failure for stress-strength interference between any combination of the supported distributions
 - Support for Exponential, Weibull, Gamma, Gumbel, Normal, Lognormal, Loglogistic, and Beta probability distributions
 - Mean residual life, quantiles, descriptive statistics summaries, random sampling from distributions

docs/Changelog.rst

@@ -18,7 +18,7 @@ Changelog
 
 -    Ability to specify "best" optimizer will result in multiple optimizers being tried and the best result being used. Optimizers tried are "L-BFGS-B", "TNC", "powell" and "nelder-mead". For more detail see the documentation on `Optimizers <https://reliability.readthedocs.io/en/latest/Optimizers.html>`_.
 -    DSZI_Model has been added to the Distributions module. This model allows for the CDF to start above 0 and finish below 1.
--    Fitters for DSZI models, including Fit_Weibull_DS and Fit_Weibull_ZI
+-    Fitters for DSZI models, including Fit_Weibull_DS, Fit_Weibull_ZI, Fit_Weibull_DSZI
 
 **API Changes**
 

docs/DSZI models.rst

@@ -109,6 +109,7 @@ The three Fitters available are Fit_Weibull_DS, Fit_Weibull_ZI, and Fit_Weibull_
 If your data contains zeros then only the Fit_Weibull_ZI and Fit_Weibull_DSZI fitters are appropriate. Using anything else will cause the zeros to be automatically removed and a warning to be printed.
 Fit_Weibull_ZI does not mandate that the failures contain zeros, but if failures does not contain zeros then ZI will be 0 and the alpha and beta parameters will be equivalent to the results from Fit_Weibull_2P.
 Fit_Weibull_DS does not mandate that right_censored data is provided, but if right_censored data is not provided then DS will be 1 and the alpha and beta parameters will be equivalent to the results from Fit_Weibull_2P.
+Fit_Weibull_DSZI does not mandate that failures contain zeros or that right_censored data is provided. If right_censored data is not provided then DS will be 1. If failures does not contain zeros then ZI will be 0. If failures does not contain zeros and no right censored data is provided then DS will be 1, ZI will be 0 and the alpha and beta parameters will be equivalent to the results from Fit_Weibull_2P.
 
 Example 3
 ---------
@@ -271,4 +272,53 @@ In addition the the visual proof provided by the probability plot (left) and the
 Example 6
 ---------
 
-This will be written soon and will show how to use Fit_Weibull_DSZI, which is a four parameter model including both the DS and ZI parameters.
+In this example we will create a DSZI model with DS=0.7 and ZI=0.2.
+Based on these parameters, we expect the random samples to be around 70% failures and of those failures 20% of the total samples (failures + right censored) should be zeros due to the zero inflated fraction.
+We draw the random samples from the model and then fit a Weibull_DSZI model to the data.
+The result is surprisingly accurate showing DS=0.700005 and ZI=0.22, with the alpha and beta parameters closely resembling the parameters of the input Weibull Distribution.
+The plot below shows the CDF on the Weibull probability plot (top) and on linear axes (bottom) which each provide a different perspective of how the distribution models the failure points.
+
+.. code:: python
+    
+    from reliability.Distributions import DSZI_Model, Weibull_Distribution
+    from reliability.Probability_plotting import plot_points
+    import matplotlib.pyplot as plt
+    from reliability.Fitters import Fit_Weibull_DSZI
+    
+    model = DSZI_Model(distribution=Weibull_Distribution(alpha=1200,beta=3),DS=0.7,ZI=0.2)
+    failures, right_censored = model.random_samples(100,seed=5,right_censored_time=3000)
+    
+    plt.subplot(211)
+    fit = Fit_Weibull_DSZI(failures=failures,right_censored=right_censored,show_probability_plot=True)
+    model.CDF(label='true model')
+    plt.legend()
+    
+    plt.subplot(212)
+    fit.distribution.CDF(label='fitted Weibull_DSZI')
+    model.CDF(label='true model')
+    plot_points(failures=failures,right_censored=right_censored)
+    plt.tight_layout()
+    plt.show()
+
+    '''
+    Results from Fit_Weibull_DSZI (95% CI):
+    Analysis method: Maximum Likelihood Estimation (MLE)
+    Optimizer: TNC
+    Failures / Right censored: 70/30 (30% right censored) 
+    
+    Parameter  Point Estimate  Standard Error  Lower CI  Upper CI
+        Alpha         1170.12         68.0933   1043.99   1311.49
+         Beta         2.60255        0.299069   2.07771   3.25997
+           DS        0.700005        0.045826  0.603391  0.781602
+           ZI            0.22       0.0414247  0.149465  0.311627 
+    
+    Goodness of fit    Value
+     Log-likelihood -463.613
+               AICc  935.647
+                BIC  945.646
+                 AD  166.025 
+    '''
+
+.. image:: images/DSZI_example6.png
+
+The DSZI model is a model of my own making. It combines the well established DS and ZI models together for the first time to enable heavily right censored data to be modelled using a DS distribution while also allowing for zero inflation of the failures.

docs/Development roadmap.rst

@@ -11,13 +11,9 @@ This roadmap is regularly changing and you may see some things remain on here fo
 If you have a suggested feature or you find a bug, please raise an `Issue <https://github.com/MatthewReid854/reliability/issues>`_ on Github or email me (alpha.reliability@gmail.com) and I will endeavour to either add it rapidly (for simple tasks and bug fixes) or add it to the roadmap.
 The current release schedule is approximately every 6 to 8 weeks.
 
-**Planned for version 0.6.0 (around August 2021)**
+**Planned for version 0.7.0 (around October 2021)**
 
--    Fit_Weibull_DSZI
 -    Correct the formatting in the API docs for every function - still need to do ALT_Fitters, Convert_data, Datasets, PoF, and Utils
-
-**Planned for version 0.7.0 (around December 2021)**
-
 -    Within all fitters, use the FNRN format to give speed improvements in the same way as Fit_Weibull_2P_grouped works internally. This will subsequently result in the deprecation of Fit_Weibull_2P_grouped once its advantage is integrated in Fit_Weibull_2P. Need to confirm this method does not introduce too much cumulative error due to floating point precision limitations.
 -    Improvement to the online documentation for how some of these methods work, including the addition of more formulas, algorithms, and better referencing.
 -    Make tests for everything that doesn't have a test yet.