2025

View page source

Release Notes for 2025

mm-dd

11-28

In the documentation, change ‘child rate’ to ‘child rate effect’ because these are not really rates, but rather random effects on rates. This is explained in the csv fit value_prior

11-15

Fix documentation for float_precision . To be more specific, the comments about the no effect rates were moved to the plot documentation.

10-17

The at_cascade example/no_ode_xam.py was passing a list for the dismod_at.plot_rate_fit rate_set argument. This made a test crash using a newer version of dismod_at and has been fixed.

10-15

The problem with the optimization perturbation was introduced by commit c802cee on 08-21 and was fixed on 10-11 (see below). The csv.fit default value for these parameters was set to 0 and as been corrected to be backward compatible to before the change on 08-21; see perturb_optimization_scale and perturb_optimization_start .

10-14

The version number for dismod_at and at_cascade was added at the beginning of each log table in the data base for each fit. One way to view this is in the log.csv file that is created when db2csv is true in the csv.predict option file.

10-11

The perturb_optimization_scale and perturb_optimization_scale options were not working. This was a bug in fit_one_job . This has been fixed. In addition these options were added to option_fit.csv in a way that is backward compatible.

09-26

Add the csv.predict option descendant_std_factor ; see the csv.predict_descend example.

09-23

Both descendent and descendant were spread throughout the documentation and at_cascade source code. The spelling descendent was changed to descendant to be consistent While this included moving job_descendent to job_descendant , it did not change the csv.fit of csv.predict API .

08-23

Some of the python sqlite3 connections were not being closed. The corresponding variable just dropped out of scope. Dropping out of scope freed the corresponding resources on Unix but not on Windows. This has been fixed (all the connections are closed).

08-22

Switch the pip build back end from setuptools to hatchling ( see pyproject.toml ) . This fixes the install of at_cascade on conda (and maybe other systems).

08-21

Complete the subprocess wish list item; i.e. use dismod_at.db2csv_command(database) instead of dismodat.py database db2csv .

08-01

Add number_sample_predict to the option_fit.csv file. This requires dismod_at-2025.7.31 or later.

07-09

Add the csv.diphtheria example.

07-08

Add the subprocess wish list item.

07-07

The value of the random effects used to create tru_predict.csv where not correct when there was a non-zero covariate multiplier. This has been fixed.

07-05

Add the censor_asymptotic sample_method to the option_all table and the option_fit.csv file .

06-05

  1. Add the following wish list items: Child Job Priors , predict_goal.csv .

  2. The name of the new option asymptotic_rcond_lower was miss spelled one place in the csv.fit.py source code. This made it impossible to use and has been fixed.

06-03

  1. Add the csv.fit asymptotic_rcond_lower option.

  2. Add the child_prior_dage and child_prior_dtime options to the csv interface; for example, see csv.dage_prior . This also added the shift_prior_dage and shift_prior_dtime options in the option all table.

  3. Remove white space from the header in the data_in.csv file in the no_data.py example. Also improve the discussion of the option_fit.csv file for this example.

  4. Discuss a Problem with the current method used to create child job priors.

05-18

Add a discussion child job shifted priors under Theory in the Create Database With Shifted Covariate References section

05-11

Add discussion of the Negative Log Likelihood and the how it relates to the sam_predict.csv file.

05-10

The special case where there is no data at the root (node, sex) pair used to cause at_cascade to abort. This has been changed so that the no data case can be used to better understand how the priors work; see csv.no_data . Note the this requires a special case when run_job_id is zero.

02-06

Add the max_fit_parent option to the option_all table and the option_fit.csv file .

01-15

Start the 2025 stable versions of at_cascade; see the new Versions table. A release for 2025 has the stable version 2025.0.number where number is the release number. Only bug fixes, not new features, are added between release numbers.

This change has broken old web links to the at_cascade documentation. An old link of the form

        https://at-cascade.readthedocs.io/html_file_name

has the following form in the latest documentation:

        https://at-cascade.readthedocs.io/latest/html_file_name

and the following for in the 2025 stable documentation:

        https://at-cascade.readthedocs.io/stable-2025/html_file_name

01-13

  1. Add child_prior_std_factor_mulcov to the option_fit csv file and shift_prior_std_factor_mulcov to the option_all table.

  2. The computation of the standard deviation for the child priors was using one degree of freedom. This is not correct because we know the true mean of the samples; i.e., the fit value (the number of degrees of freedom should be zero). This has been fixed and the child prior standard deviations that are a factor of \(\sqrt{(N-1)/N}\) smaller where N is the number of samples.

  3. There was a bug in fit_one_process . It sometimes returned on the main branch before all the other processes had finished. This has been fixed. It was found by repeated running of the example/relrisk.py test which uses a large value for max_number_cpu .

01-12

The test/csv/fit.py and test/csv/sample_fail.py tests were failing on some systems; see pull request 29 . The failures have has been fixed by not expecting the plot_data.pdf file to exist for (node, sex) pairs that have only one data point.