Skip to content

Commit

Permalink
Revert "tidyup vignettes"
Browse files Browse the repository at this point in the history 
This reverts commit dcbdee6.
  • Loading branch information
@florale
florale committed Jun 20, 2023
1 parent dcbdee6 commit 8fd06f1
Show file tree
Hide file tree
Showing 7 changed files with 229 additions and 153 deletions.
88 changes: 79 additions & 9 deletions vignettes/comp-outcome.Rmd
View file
Original file line number Diff line number Diff line change
Expand Up @@ -67,8 +67,13 @@ Our `brms` model can be then fitted using the `brmcoda()` function.
```r
mv <- brmcoda(compilr = cilr,
formula = mvbind(ilr1, ilr2, ilr3, ilr4) ~ STRESS + (1 | ID),
cores = 8, seed = 123)
#> Error in unserialize(socklist[[n]]): error reading from connection
cores = 8, seed = 123, backend = "cmdstanr")
#> Warning: In the future, 'rescor' will be set to FALSE by default for all models. It is thus
#> recommended to explicitely set 'rescor' via 'set_rescor' instead of using the default.
#> Warning: CmdStan's precompiled header (PCH) files may need to be rebuilt.
#> If your model failed to compile please run rebuild_cmdstan().
#> If the issue persists please open a bug report.
#> Error: An error occured during compilation! See the message above for more information.
```

Here is a `summary()` of the model.
Expand All @@ -77,7 +82,57 @@ We can see that stress significantly predicted `ilr1` and `ilr2`.

```r
summary(mv$Model)
#> Error in eval(expr, envir, enclos): object 'mv' not found
#> Family: MV(gaussian, gaussian, gaussian, gaussian)
#> Links: mu = identity; sigma = identity
#> mu = identity; sigma = identity
#> mu = identity; sigma = identity
#> mu = identity; sigma = identity
#> Formula: ilr1 ~ STRESS + (1 | ID)
#> ilr2 ~ STRESS + (1 | ID)
#> ilr3 ~ STRESS + (1 | ID)
#> ilr4 ~ STRESS + (1 | ID)
#> Data: tmp (Number of observations: 3540)
#> Draws: 8 chains, each with iter = 6000; warmup = 1000; thin = 1;
#> total post-warmup draws = 40000
#>
#> Group-Level Effects:
#> ~ID (Number of levels: 266)
#> Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
#> sd(ilr1_Intercept) 0.33 0.02 0.30 0.37 1.00 13574 21989
#> sd(ilr2_Intercept) 0.30 0.02 0.28 0.34 1.00 11200 19827
#> sd(ilr3_Intercept) 0.39 0.02 0.35 0.43 1.00 17013 24936
#> sd(ilr4_Intercept) 0.30 0.02 0.27 0.33 1.00 16596 25482
#>
#> Population-Level Effects:
#> Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
#> ilr1_Intercept -0.43 0.02 -0.48 -0.39 1.00 11069 19841
#> ilr2_Intercept 1.47 0.02 1.42 1.51 1.00 9583 18405
#> ilr3_Intercept -0.87 0.03 -0.93 -0.81 1.00 17975 27392
#> ilr4_Intercept 0.65 0.02 0.60 0.69 1.00 18563 26992
#> ilr1_STRESS -0.01 0.00 -0.02 -0.00 1.00 64970 35109
#> ilr2_STRESS 0.01 0.00 0.00 0.01 1.00 56689 36287
#> ilr3_STRESS 0.00 0.01 -0.01 0.01 1.00 67676 33111
#> ilr4_STRESS 0.01 0.00 -0.00 0.01 1.00 62003 35093
#>
#> Family Specific Parameters:
#> Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
#> sigma_ilr1 0.44 0.01 0.43 0.45 1.00 57828 33519
#> sigma_ilr2 0.38 0.00 0.37 0.39 1.00 59436 34195
#> sigma_ilr3 0.70 0.01 0.68 0.71 1.00 52722 34899
#> sigma_ilr4 0.53 0.01 0.51 0.54 1.00 52690 33648
#>
#> Residual Correlations:
#> Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
#> rescor(ilr1,ilr2) -0.54 0.01 -0.57 -0.52 1.00 57259 33739
#> rescor(ilr1,ilr3) -0.18 0.02 -0.21 -0.14 1.00 56781 34166
#> rescor(ilr2,ilr3) -0.05 0.02 -0.09 -0.02 1.00 56093 32472
#> rescor(ilr1,ilr4) 0.11 0.02 0.07 0.14 1.00 56993 31275
#> rescor(ilr2,ilr4) -0.05 0.02 -0.08 -0.01 1.00 54469 33148
#> rescor(ilr3,ilr4) 0.56 0.01 0.54 0.58 1.00 53173 34399
#>
#> Draws were sampled using sample(hmc). For each parameter, Bulk_ESS
#> and Tail_ESS are effective sample size measures, and Rhat is the potential
#> scale reduction factor on split chains (at convergence, Rhat = 1).
```

# Bayes Factor for compositional multilevel modelling
Expand All @@ -104,29 +159,44 @@ Let's examine whether stress predicts the overall sleep-wake composition.
mv0 <- brmcoda(compilr = cilr,
formula = mvbind(ilr1, ilr2, ilr3, ilr4) ~ 1 + (1 | ID),
iter = 6000, chains = 8, cores = 8, seed = 123, warmup = 1000,
save_pars = save_pars(all = TRUE))
#> Error in unserialize(socklist[[n]]): error reading from connection
backend = "cmdstanr", save_pars = save_pars(all = TRUE))
#> Warning: In the future, 'rescor' will be set to FALSE by default for all models. It is thus
#> recommended to explicitely set 'rescor' via 'set_rescor' instead of using the default.
#> Warning: CmdStan's precompiled header (PCH) files may need to be rebuilt.
#> If your model failed to compile please run rebuild_cmdstan().
#> If the issue persists please open a bug report.
#> Error: An error occured during compilation! See the message above for more information.
# full model
mv <- brmcoda(compilr = cilr,
formula = mvbind(ilr1, ilr2, ilr3, ilr4) ~ STRESS + (1 | ID),
iter = 6000, chains = 8, cores = 8, seed = 123, warmup = 1000,
save_pars = save_pars(all = TRUE))
#> Error in unserialize(socklist[[n]]): error reading from connection
backend = "cmdstanr", save_pars = save_pars(all = TRUE))
#> Warning: In the future, 'rescor' will be set to FALSE by default for all models. It is thus recommended to explicitely set 'rescor' via 'set_rescor' instead of using the default.

#> Warning: CmdStan's precompiled header (PCH) files may need to be rebuilt.
#> If your model failed to compile please run rebuild_cmdstan().
#> If the issue persists please open a bug report.
#> Error: An error occured during compilation! See the message above for more information.
```

We can now compare these models with the `bayesfactor_models()` function


```r
comparison <- bayesfactor_models(mv$Model, denominator = mv0$Model)
#> Error in eval(expr, envir, enclos): object 'mv' not found
```



```r
comparison
#> Error in eval(expr, envir, enclos): object 'comparison' not found
#> Bayes Factors for Model Comparison
#>
#> Model BF
#> [1] 2.90e-05
#>
#> * Against Denominator: [2]
#> * Bayes Factor Type: marginal likelihoods (bridgesampling)
```

With a $BF$ < 1, our data favours the intercept only model, showing that there is
Expand Down
10 changes: 5 additions & 5 deletions vignettes/comp-outcome.Rmd.orig
View file
Original file line number Diff line number Diff line change
Expand Up @@ -60,10 +60,10 @@ head(cilr$TotalILR)
A model with multilevel compositional outcomes is multivariate, as it has multiple ILR coordinate outcomes,each of which is predicted by a set of predictors.
Our `brms` model can be then fitted using the `brmcoda()` function.

```{r brmcoda, results = "hide", message=FALSE, warning=FALSE}
```{r brmcoda, results = "hide", message=FALSE}
mv <- brmcoda(compilr = cilr,
formula = mvbind(ilr1, ilr2, ilr3, ilr4) ~ STRESS + (1 | ID),
cores = 8, seed = 123)
cores = 8, seed = 123, backend = "cmdstanr")
```

Here is a `summary()` of the model.
Expand Down Expand Up @@ -91,17 +91,17 @@ Let's examine whether stress predicts the overall sleep-wake composition.
*Note*: To use Bayes factors, `brmsfit` models must be fitted with an additional non-default argument
`save_pars = save_pars(all = TRUE)`.

```{r bfmodel, results = "hide", message=FALSE, warning=FALSE}
```{r bfmodel, results = "hide", message=FALSE}
# intercept only
mv0 <- brmcoda(compilr = cilr,
formula = mvbind(ilr1, ilr2, ilr3, ilr4) ~ 1 + (1 | ID),
iter = 6000, chains = 8, cores = 8, seed = 123, warmup = 1000,
save_pars = save_pars(all = TRUE))
backend = "cmdstanr", save_pars = save_pars(all = TRUE))
# full model
mv <- brmcoda(compilr = cilr,
formula = mvbind(ilr1, ilr2, ilr3, ilr4) ~ STRESS + (1 | ID),
iter = 6000, chains = 8, cores = 8, seed = 123, warmup = 1000,
save_pars = save_pars(all = TRUE))
backend = "cmdstanr", save_pars = save_pars(all = TRUE))
```

We can now compare these models with the `bayesfactor_models()` function
Expand Down
Loading

0 comments on commit 8fd06f1

Please sign in to comment.