A Markov transition kernel.

Construct a Transition from sampling, state transformation and logging functions.

NB: The three function arguments are used in the order in which they appear here:

1. a random sample w :: t is produced, using the current state x :: s as input
2. output z :: a and next state x' :: s are computed using w and x
3. a logging message is constructed, using z and x' as arguments.

Run a Transition for a number of steps, while logging each iteration.

Returns both the list of outputs and the final state.

Run a Transition for a number of steps, while logging each iteration.

Returns the list of outputs.

Run a Transition for a number of steps, while logging each iteration.

Returns the final state.

Perform one Transition and check output and updated state against the current state, producing an Either with the result of the comparison.

Can be useful for detecting early divergence or lack of convergence etc.

Render a logging message along with an annotation of its severity.

Add "Severity" information to a log message. This is often used to convey how significant a log message is.

Classes of severity for log messages. These have been chosen to match syslog severity levels

Handlers are mechanisms to interpret the meaning of logging as an action in the underlying monad. They are simply functions from log messages to m-actions.

withFDHandler creates a new Handler that will append a given file descriptor (or Handle, as it is known in the "base" library). Note that this Handler requires log messages to be of type Doc. This abstractly specifies a pretty-printing for log lines. The two arguments two withFDHandler determine how this pretty-printing should be realised when outputting log lines.

These Handlers asynchronously log messages to the given file descriptor, rather than blocking.

Options that be used to configure withBatchingHandler.

Defaults for BatchingOptions

defaultBatchingOptions = BatchingOptions {flushMaxDelay = 1000000
                                         ,flushMaxQueueSize = 100
                                         ,blockWhenFull = True}

Create a new batched handler. Batched handlers take batches of messages to log at once, which can be more performant than logging each individual message.

A batched handler flushes under three criteria:

1. The flush interval has elapsed and the queue is not empty.
2. The queue has become full and needs to be flushed.
3. The scope of withBatchedHandler is exited.

Batched handlers queue size and flush period can be configured via BatchingOptions.

A handle managing output to the Haskell program's standard output channel.

A handle managing output to the Haskell program's standard error channel.