### Install chkptstanr from GitHub

Source: https://github.com/venpopov/chkptstanr/blob/master/README.md

Install the development version of the package from GitHub to access the latest fixes.

```r
remotes::install_github("venpopov/chkptstanr")
```

--------------------------------

### Reset Sampling with reset = TRUE

Source: https://github.com/venpopov/chkptstanr/blob/master/README.md

To reset sampling and start from scratch, use `reset = TRUE`. This is permissible as long as key arguments like formula, data, or family have not changed. The `stop_after` argument can be modified.

```r
fit1 <- chkpt_brms(count ~ zAge + zBase * Trt + (1|patient),
                   data = epilepsy, 
                   family = poisson(),
                   iter_per_chkpt = 200,
                   path = 'checkpoints/epilepsy',
                   stop_after = 1600,
                   reset = TRUE)
```

--------------------------------

### Load required packages

Source: https://github.com/venpopov/chkptstanr/blob/master/README.md

Load the necessary libraries to begin using chkptstanr with brms and lme4.

```r
library(chkptstanr)
library(brms)
library(lme4)
```

--------------------------------

### make_brmsfit

Source: https://context7.com/venpopov/chkptstanr/llms.txt

Converts checkpoint output files into a standard brmsfit object.

```APIDOC
## make_brmsfit

### Description
Convert checkpoint output files into a standard brmsfit object. Primarily used internally but can be called directly if needed.

### Parameters
#### Request Body
- **formula** (formula) - Required - The model formula.
- **data** (data.frame) - Required - The dataset used for fitting.
- **family** (family) - Required - The response distribution family.
- **path** (string) - Required - The path to the checkpoint directory.

### Response
- **brmsfit** (object) - A standard brmsfit object.
```

--------------------------------

### Create brmsfit Object from Checkpoints

Source: https://context7.com/venpopov/chkptstanr/llms.txt

Convert checkpoint output files into a standard brmsfit object. This is typically done automatically by chkpt_brms but can be called directly.

```r
library(chkptstanr)
library(brms)

# After checkpointed sampling is complete, create brmsfit
# (typically done automatically by chkpt_brms)
fit <- make_brmsfit(
  formula = count ~ zAge + zBase * Trt + (1|patient),
  data = epilepsy,
  family = poisson(),
  path = "checkpoints/epilepsy"
)

# Result is a standard brmsfit object
class(fit)
#> [1] "brmsfit"

summary(fit)
```

--------------------------------

### Fit Custom Stan Model with Checkpointing

Source: https://context7.com/venpopov/chkptstanr/llms.txt

Fit a custom Stan model using checkpointing. Provide the Stan model code, data, and checkpointing parameters to chkpt_stan.

```r
library(chkptstanr)
library(posterior)
library(bayesplot)

# Define custom Stan model (Eight Schools example)
stan_code <- "
data {
  int<lower=0> n;
  array[n] real y;
  array[n] real<lower=0> sigma;
}
parameters {
  real mu;
  real<lower=0> tau;
  vector[n] eta;
}
transformed parameters {
  vector[n] theta;
  theta = mu + tau * eta;
}
model {
  target += normal_lpdf(eta | 0, 1);
  target += normal_lpdf(y | theta, sigma);
}
"

# Prepare data as a list
stan_data <- list(
  n = 8,
  y = c(28, 8, -3, 7, -1, 1, 18, 12),
  sigma = c(15, 10, 16, 11, 9, 11, 10, 18)
)

# Fit model with checkpointing
fit <- chkpt_stan(
  model_code = stan_code,
  data = stan_data,
  iter_warmup = 1000,
  iter_sampling = 1000,
  iter_per_chkpt = 250,
  parallel_chains = 4,
  seed = 1234,
  path = "checkpoints/eight_schools"
)

# Combine checkpoint draws into a single draws_array
draws <- combine_chkpt_draws(fit)

# Summarize with posterior package
posterior::summarise_draws(draws)

# Visualize traces with bayesplot
bayesplot::mcmc_trace(draws, pars = c("mu", "tau"))
```

--------------------------------

### Configure Checkpoint Structure

Source: https://context7.com/venpopov/chkptstanr/llms.txt

Calculate and display the checkpoint structure before fitting. Useful for planning sampling strategies. Access individual components like warmup_chkpts and sample_chkpts.

```r
library(chkptstanr)

# Calculate checkpoint structure
setup <- chkpt_setup(
  iter_warmup = 2000,
  iter_sampling = 5000,
  iter_per_chkpt = 100
)

print(setup)

# Access individual components
setup$warmup_chkpts   # 20
setup$sample_chkpts   # 50
setup$total_chkpts    # 70
setup$iter_per_chkpt  # 100
```

--------------------------------

### Fit brms Model with Custom Priors and Checkpointing

Source: https://context7.com/venpopov/chkptstanr/llms.txt

Fit a brms model with custom priors using checkpointing. Define priors using the brms prior syntax and pass them to chkpt_brms.

```r
library(chkptstanr)
library(brms)

# Define custom priors
custom_priors <- prior(constant(1), class = "b") +
  prior(constant(2), class = "b", coef = "zBase") +
  prior(constant(0.5), class = "sd")

# Fit model with custom priors
fit <- chkpt_brms(
  formula = count ~ zAge + zBase + (1|patient),
  data = epilepsy,
  family = poisson(),
  prior = custom_priors,
  iter_warmup = 1000,
  iter_sampling = 1000,
  iter_per_chkpt = 250,
  path = "checkpoints/epilepsy_priors"
)

# Verify priors were applied
prior_summary(fit)
```

--------------------------------

### Resume brms Model Fitting After Interruption

Source: https://context7.com/venpopov/chkptstanr/llms.txt

Demonstrates how to interrupt and resume brms model fitting with checkpointing. Rerunning the same chkpt_brms call will continue sampling from the last saved checkpoint.

```r
library(chkptstanr)
library(brms)

# Start sampling - can be interrupted at any time (Ctrl+C)
fit <- chkpt_brms(
  formula = count ~ zAge + zBase * Trt + (1|patient),
  data = epilepsy,
  family = poisson(),
  iter_per_chkpt = 200,
  path = "checkpoints/epilepsy"
)
# User presses Ctrl+C after checkpoint 6...
#> Chkpt: 6 / 10; Iteration: 1200 / 2000 (sample)
#> Sampling aborted. You can examine the results or continue sampling by rerunning the same code.

# Examine partial results (if past warmup)
summary(fit)
#> Draws: 4 chains, each with iter = 1200; warmup = 1000; thin = 1;
#>        total post-warmup draws = 800

# Resume sampling by running the same code
fit <- chkpt_brms(
  formula = count ~ zAge + zBase * Trt + (1|patient),
  data = epilepsy,
  family = poisson(),
  iter_per_chkpt = 200,
  path = "checkpoints/epilepsy"
)
#> Chkpt: 7 / 10; Iteration: 1400 / 2000 (sample)
#> Chkpt: 8 / 10; Iteration: 1600 / 2000 (sample)
#> Chkpt: 9 / 10; Iteration: 1800 / 2000 (sample)
#> Chkpt: 10 / 10; Iteration: 2000 / 2000 (sample)
#> Checkpointing complete
```

--------------------------------

### Fit brms Model with Checkpointing

Source: https://context7.com/venpopov/chkptstanr/llms.txt

Fit a brms model using checkpointing. Specify the formula, data, family, and checkpointing parameters like iter_per_chkpt and path.

```r
fit <- chkpt_brms(
  formula = count ~ zAge + zBase * Trt + (1|patient),
  data = epilepsy,
  family = poisson(),
  iter_per_chkpt = 200,
  path = "checkpoints/epilepsy"
)
```

--------------------------------

### Summarize and Visualize Draws

Source: https://context7.com/venpopov/chkptstanr/llms.txt

Use the posterior and bayesplot packages to analyze draws extracted from a checkpointed model.

```r
posterior::summarise_draws(draws)
posterior::mcse_mean(draws)
```

```r
bayesplot::mcmc_dens(draws)
bayesplot::mcmc_pairs(draws, pars = c("mu", "tau"))
```

--------------------------------

### Fit brms Models with Checkpointing

Source: https://context7.com/venpopov/chkptstanr/llms.txt

Fit a Poisson mixed model using brms with checkpointing. Specify iterations per checkpoint and the path to save checkpoints. The output can be examined using standard brms functions.

```r
library(chkptstanr)
library(brms)

# Fit a Poisson mixed model with checkpointing
fit <- chkpt_brms(
  formula = count ~ zAge + zBase * Trt + (1|patient),
  data = epilepsy,
  family = poisson(),
  iter_warmup = 1000,        # Warmup iterations per chain
  iter_sampling = 1000,       # Sampling iterations per chain
  iter_per_chkpt = 200,       # Iterations between checkpoints
  parallel_chains = 4,        # Number of parallel chains
  seed = 1234,
  path = "checkpoints/epilepsy_model"
)

# Output during fitting:
#> Initial Warmup (Typical Set)
#> Chkpt: 1 / 10; Iteration: 200 / 2000 (warmup)
#> Chkpt: 2 / 10; Iteration: 400 / 2000 (warmup)
#> ...
#> Chkpt: 10 / 10; Iteration: 2000 / 2000 (sample)
#> Checkpointing complete

# Examine results using standard brms functions
summary(fit)
#> Family: poisson
#>   Links: mu = log
#> Formula: count ~ zAge + zBase * Trt + (1 | patient)
#>    Data: data (Number of observations: 236)
#>   Draws: 4 chains, each with iter = 2000; warmup = 1000; thin = 1; 
#>          total post-warmup draws = 4000

# Posterior predictive check
pp_check(fit)

# Model comparison with LOO-CV
fit2 <- chkpt_brms(
  formula = count ~ zAge + zBase,
  data = epilepsy,
  family = poisson(),
  iter_per_chkpt = 200,
  path = "checkpoints/epilepsy_model2"
)
loo(fit, fit2)
```

--------------------------------

### Predetermine Stopping Point with stop_after

Source: https://github.com/venpopov/chkptstanr/blob/master/README.md

Use the `stop_after` argument to specify the total number of iterations after which the sampler should stop. This allows for predetermining the run's stopping point.

```r
fit1 <- chkpt_brms(count ~ zAge + zBase * Trt + (1|patient),
                   data = epilepsy, 
                   family = poisson(),
                   iter_per_chkpt = 200,
                   stop_after = 1400,
                   path = 'checkpoints/epilepsy')
```

--------------------------------

### Reset Checkpoints and Recompile

Source: https://github.com/venpopov/chkptstanr/blob/master/README.md

If essential arguments like the formula, data, or family are changed, an error will occur. In such cases, you must reset the checkpoints and recompile the model using `reset_checkpoints(path, recompile = TRUE)`.

```r
reset_checkpoints('checkpoints/epilepsy', recompile = TRUE)
```

--------------------------------

### Examine Partially Sampled Model

Source: https://github.com/venpopov/chkptstanr/blob/master/README.md

After manually aborting the sampling process, examine the summary of the partially fitted brmsfit object.

```r
summary(fit1)
```

--------------------------------

### Error on Changing Formula with reset = TRUE

Source: https://github.com/venpopov/chkptstanr/blob/master/README.md

Attempting to change the formula while `reset = TRUE` will result in an error, indicating that important arguments have been modified and a full reset is required.

```r
fit1 <- chkpt_brms(count ~ 1 + (1|patient),
                   data = epilepsy, 
                   family = poisson(),
                   iter_per_chkpt = 200,
                   path = 'checkpoints/epilepsy',
                   stop_after = 1600,
                   reset = TRUE)
```

--------------------------------

### Extract Complete Stan State for Resuming

Source: https://context7.com/venpopov/chkptstanr/llms.txt

Extract the complete sampler state, including inverse metric, step size, and initial values, for continuing sampling. This is used internally for checkpointing.

```r
library(chkptstanr)
library(cmdstanr)

# Fit a model
fit <- cmdstanr_example("schools_ncp")

# Extract state for resuming
state <- extract_stan_state(fit, phase = "sample")

# State contains:
# - inv_metric: Inverse mass matrix for each chain
# - step_size_adapt: Adapted step sizes
# - inits: Initial values for next iteration (last draw from each chain)

names(state)
#> [1] "inv_metric" "step_size_adapt" "inits"
```

--------------------------------

### Checkpoint brms Model Fitting

Source: https://context7.com/venpopov/chkptstanr/llms.txt

Fit a brms model with checkpointing enabled. This function accepts the same arguments as brm() and saves progress to a specified path.

```r
fit_brms <- chkpt_brms(
  formula = count ~ zAge + zBase,
  data = epilepsy,
  family = poisson(),
  iter_per_chkpt = 200,
  path = "checkpoints/brms_model"
)
draws_brms <- combine_chkpt_draws(fit_brms)
```

--------------------------------

### Extract HMC Sampler Information

Source: https://context7.com/venpopov/chkptstanr/llms.txt

Extract the inverse metric and step size adaptation from CmdStanMCMC objects. This is primarily for advanced diagnostics.

```r
library(chkptstanr)
library(cmdstanr)

# Fit a model with cmdstanr
fit <- cmdstanr_example("schools_ncp")

# Extract HMC information
hmc_info <- extract_hmc_info(fit)

# Access inverse metric for each chain
hmc_info$inv_metric
#> [[1]]
#> [1] 0.0234 0.0156 0.0289 ...
#> [[2]]
#> [1] 0.0198 0.0145 0.0312 ...

# Access adapted step sizes
hmc_info$step_size_adapt
#> [1] 0.892 0.756 0.834 0.901
```

--------------------------------

### extract_hmc_info

Source: https://context7.com/venpopov/chkptstanr/llms.txt

Extracts the inverse metric and step size adaptation from CmdStanMCMC objects for advanced diagnostics.

```APIDOC
## extract_hmc_info

### Description
Extract the inverse metric and step size adaptation from CmdStanMCMC objects. Primarily used internally but available for advanced diagnostics.

### Parameters
#### Request Body
- **fit** (CmdStanMCMC) - Required - The fitted model object from cmdstanr.

### Response
- **inv_metric** (list) - Inverse mass matrix for each chain.
- **step_size_adapt** (numeric) - Adapted step sizes.
```

--------------------------------

### Manually Abort brms Model Sampling

Source: https://github.com/venpopov/chkptstanr/blob/master/README.md

Run a brms model with checkpointing enabled and manually abort after a certain number of iterations. This saves intermediate samples.

```r
fit1 <- chkpt_brms(count ~ zAge + zBase * Trt + (1|patient),
                   data = epilepsy, 
                   family = poisson(),
                   iter_per_chkpt = 200,
                   path = 'checkpoints/epilepsy')
```

--------------------------------

### Extract Draws from CmdStanMCMC Object

Source: https://context7.com/venpopov/chkptstanr/llms.txt

Extract draws from a CmdStanMCMC object, with options to include or exclude warmup draws.

```r
library(chkptstanr)
library(cmdstanr)

# Fit a model
fit <- cmdstanr_example("schools_ncp")

# Extract only post-warmup draws
draws_sample <- extract_chkpt_draws(fit, phase = "sample")
dim(draws_sample)
#> [1] 1000    4   19  # iterations x chains x variables

# Extract including warmup draws
draws_warmup <- extract_chkpt_draws(fit, phase = "warmup")
dim(draws_warmup)
#> [1] 2000    4   19  # includes warmup iterations
```

--------------------------------

### Combine Checkpoint Draws

Source: https://context7.com/venpopov/chkptstanr/llms.txt

Combine posterior draws from all checkpoints into a single draws_array object. Compatible with posterior and bayesplot packages. Use after fitting with chkpt_stan.

```r
library(chkptstanr)
library(posterior)
library(bayesplot)

# After fitting with chkpt_stan
fit <- chkpt_stan(
  model_code = stan_code,
  data = stan_data,
  iter_warmup = 1000,
  iter_sampling = 1000,
  iter_per_chkpt = 250,
  path = "checkpoints/model"
)

# Combine draws from all checkpoints
draws <- combine_chkpt_draws(fit)
class(draws)

```

--------------------------------

### Reset or Clear Checkpoints

Source: https://context7.com/venpopov/chkptstanr/llms.txt

Reset checkpoint data to restart sampling from the beginning. Optionally keep the compiled model to avoid recompilation for faster restarts. Can also be used directly in chkpt_brms.

```r
library(chkptstanr)

# Reset checkpoints but keep compiled model (faster restart)
reset_checkpoints(
  path = "checkpoints/epilepsy",
  reset = TRUE,
  recompile = FALSE
)

# Completely reset including recompiling the model
reset_checkpoints(
  path = "checkpoints/epilepsy",
  reset = TRUE,
  recompile = TRUE
)

# Can also use reset argument directly in chkpt_brms
fit <- chkpt_brms(
  formula = count ~ zAge + zBase * Trt + (1|patient),
  data = epilepsy,
  family = poisson(),
  iter_per_chkpt = 200,
  path = "checkpoints/epilepsy",
  reset = TRUE  # Start fresh without recompiling
)
```

--------------------------------

### extract_chkpt_draws

Source: https://context7.com/venpopov/chkptstanr/llms.txt

Extracts draws from a CmdStanMCMC object, with options for including warmup draws.

```APIDOC
## extract_chkpt_draws

### Description
Extract draws from a CmdStanMCMC object, with options for including warmup draws.

### Parameters
#### Request Body
- **fit** (CmdStanMCMC) - Required - The fitted model object.
- **phase** (string) - Required - The phase to extract ("sample" or "warmup").

### Response
- **draws** (array) - An array of iterations x chains x variables.
```

--------------------------------

### Stop brms Model Fitting After Predetermined Iterations

Source: https://context7.com/venpopov/chkptstanr/llms.txt

Use the `stop_after` argument in `chkpt_brms` to automatically halt sampling after a specified number of iterations. This is useful for testing convergence or managing computational time.

```r
library(chkptstanr)
library(brms)

# Stop after 1400 iterations (7 checkpoints)
fit <- chkpt_brms(
  formula = count ~ zAge + zBase * Trt + (1|patient),
  data = epilepsy,
  family = poisson(),
  iter_per_chkpt = 200,
  stop_after = 1400,
  path = "checkpoints/epilepsy"
)
#> Sampling will stop after checkpoint 7
#> Initial Warmup (Typical Set)
#> Chkpt: 1 / 10; Iteration: 200 / 2000 (warmup)
#> ...
#> Chkpt: 7 / 10; Iteration: 1400 / 2000 (sample)
#> Sampling aborted. You can examine the results or continue sampling by rerunning the same code.

# Check convergence at this point
summary(fit)
```

--------------------------------

### extract_stan_state

Source: https://context7.com/venpopov/chkptstanr/llms.txt

Extracts the complete sampler state including inverse metric, step size, and initial values for continuing sampling.

```APIDOC
## extract_stan_state

### Description
Extract the complete sampler state including inverse metric, step size, and initial values for continuing sampling. Used internally for checkpointing.

### Parameters
#### Request Body
- **fit** (CmdStanMCMC) - Required - The fitted model object.
- **phase** (string) - Required - The sampling phase (e.g., "sample").

### Response
- **inv_metric** (list) - Inverse mass matrix for each chain.
- **step_size_adapt** (numeric) - Adapted step sizes.
- **inits** (list) - Initial values for next iteration.
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.