Possibility to manually specify no_random_effects in mhmkin

4 files changed, 43 insertions, 39 deletions

diff --git a/tests/testthat/illparms_hfits_synth.txt b/tests/testthat/illparms_hfits_synth.txt
index affd1318..7a69645b 100644
--- a/tests/testthat/illparms_hfits_synth.txt
+++ b/tests/testthat/illparms_hfits_synth.txt
@@ -1,8 +1,4 @@
            error
-degradation const                      
-       SFO                             
-       FOMC sd(log_alpha), sd(log_beta)
-           error
-degradation tc                                       
-       SFO  sd(parent_0)                             
-       FOMC sd(parent_0), sd(log_alpha), sd(log_beta)
+degradation const        tc                        
+       SFO  sd(parent_0) sd(parent_0)              
+       FOMC sd(log_beta) sd(parent_0), sd(log_beta)
diff --git a/tests/testthat/print_fits_synth_const.txt b/tests/testthat/print_fits_synth_const.txt
index b4bbe6ca..5d076d3d 100644
--- a/tests/testthat/print_fits_synth_const.txt
+++ b/tests/testthat/print_fits_synth_const.txt
@@ -4,8 +4,6 @@ Status of individual fits:
       dataset
 model  1  2  3  4  5  6 
   SFO  OK OK OK OK OK OK
-  FOMC C  OK OK OK OK C 
+  FOMC OK OK OK OK OK OK
 
-C: Optimisation did not converge:
-false convergence (8)
 OK: No warnings
diff --git a/tests/testthat/summary_hfit_sfo_tc.txt b/tests/testthat/summary_hfit_sfo_tc.txt
index 0a61f75f..0618c715 100644
--- a/tests/testthat/summary_hfit_sfo_tc.txt
+++ b/tests/testthat/summary_hfit_sfo_tc.txt
@@ -8,7 +8,7 @@ Equations:
 d_parent/dt = - k_parent * parent
 
 Data:
-104 observations of 1 variable(s) grouped in 6 datasets
+95 observations of 1 variable(s) grouped in 6 datasets
 
 Model predictions using solution type analytical 
 
@@ -19,7 +19,7 @@ Variance model: Two-component variance function
 
 Starting values for degradation parameters:
     parent_0 log_k_parent 
-         101           -3 
+          94           -2 
 
 Fixed degradation parameter values:
 None
@@ -27,7 +27,7 @@ None
 Starting values for random effects (square root of initial entries in omega):
              parent_0 log_k_parent
 parent_0            4          0.0
-log_k_parent        0          0.4
+log_k_parent        0          0.7
 
 Starting values for error model parameters:
 a.1 b.1 
@@ -37,15 +37,15 @@ Results:
 
 Likelihood computed by importance sampling
   AIC BIC logLik
-  524 523   -257
+  542 541   -266
 
 Optimised parameters:
-                  est. lower  upper
-parent_0        100.68 99.27 102.08
-log_k_parent     -3.38 -3.55  -3.21
-a.1               0.87  0.59   1.14
-b.1               0.05  0.04   0.06
-SD.log_k_parent   0.21  0.09   0.33
+                 est. lower upper
+parent_0        92.52 89.11  95.9
+log_k_parent    -1.66 -2.07  -1.3
+a.1              2.03  1.60   2.5
+b.1              0.09  0.07   0.1
+SD.log_k_parent  0.51  0.22   0.8
 
 Correlation: 
              pr_0
@@ -53,18 +53,18 @@ log_k_parent 0.1
 
 Random effects:
                 est. lower upper
-SD.log_k_parent  0.2  0.09   0.3
+SD.log_k_parent  0.5   0.2   0.8
 
 Variance model:
     est. lower upper
-a.1 0.87  0.59  1.14
-b.1 0.05  0.04  0.06
+a.1 2.03  1.60   2.5
+b.1 0.09  0.07   0.1
 
 Backtransformed parameters:
-          est. lower upper
-parent_0 1e+02 99.27 1e+02
-k_parent 3e-02  0.03 4e-02
+         est. lower upper
+parent_0 92.5  89.1  95.9
+k_parent  0.2   0.1   0.3
 
 Estimated disappearance times:
        DT50 DT90
-parent   20   68
+parent    4   12
diff --git a/tests/testthat/test_mhmkin.R b/tests/testthat/test_mhmkin.R
index e2339f28..da063326 100644
--- a/tests/testthat/test_mhmkin.R
+++ b/tests/testthat/test_mhmkin.R
@@ -3,8 +3,11 @@ context("Batch fitting and diagnosing hierarchical kinetic models")
 test_that("Multiple hierarchical kinetic models can be fitted and diagnosed", {
 
   skip_on_cran()
-  fits_synth_const <- suppressWarnings(
-    mmkin(c("SFO", "FOMC"), ds_sfo[1:6], cores = n_cores, quiet = TRUE))
+  fits_synth_const <- mmkin(c("SFO", "FOMC"), ds_fomc[1:6], cores = n_cores, quiet = TRUE)
+
+  expect_known_output(
+    print(fits_synth_const),
+    "print_fits_synth_const.txt")
 
   fits_synth_tc <- suppressWarnings(
     update(fits_synth_const, error_model = "tc"))
@@ -19,8 +22,8 @@ test_that("Multiple hierarchical kinetic models can be fitted and diagnosed", {
     print(illparms(hfits)),
     "illparms_hfits_synth.txt")
 
-  expect_equal(which.min(AIC(hfits)), 3)
-  expect_equal(which.min(BIC(hfits)), 3)
+  expect_equal(which.min(AIC(hfits)), 4)
+  expect_equal(which.min(BIC(hfits)), 4)
 
   hfit_sfo_tc <- update(hfits[["SFO", "tc"]],
     covariance.model = diag(c(0, 1)))
@@ -38,12 +41,19 @@ test_that("Multiple hierarchical kinetic models can be fitted and diagnosed", {
   expect_known_output(print(test_summary, digits = 1),
     "summary_hfit_sfo_tc.txt")
 
-  # It depends on the platform exactly which of the datasets fail to converge
-  # with FOMC, because they were generated to be SFO
-  skip_on_travis()
-
-  expect_known_output(
-    print(fits_synth_const),
-    "print_fits_synth_const.txt")
-
+  hfits_sfo_reduced <- update(hfits,
+    no_random_effect = illparms(hfits))
+  expect_equal(
+    as.character(illparms(hfits_sfo_reduced)),
+    rep("", 4))
+
+  # We can also manually set up an object specifying random effects to be
+  # excluded. Entries in the inital list have to be by column
+  no_ranef <- list("parent_0", "log_beta", "parent_0", c("parent_0", "log_beta"))
+  dim(no_ranef) <- c(2, 2)
+
+  hfits_sfo_reduced_2 <- update(hfits,
+    no_random_effect = no_ranef)
+  expect_equivalent(round(anova(hfits_sfo_reduced), 0),
+    round(anova(hfits_sfo_reduced_2), 0))
 })