This example is a write up of an answer I gave to a coin flip question on StackExchange. Essentially, the question was: given a series of coin flips, how many coin flips would it take (on average) to get the sequence \(Heads-Tails-Heads\).
It’s possible to model this scenario using absorbing Markov chains. Metrics within the package can then be used to answer the posed question, as well as explore other questions that may be of interest.
The complete code for this example is available at Github.
The first step is to create a \(P\) matrix representing the possible states. The key is to define \(HTH\) as our absorbing state. In other words, once the model reaches that state, it is complete.
# Coin flip probabilities <- 0.5 # Probability of heads p <- 1 - p # Probability of tails q <- matrix(c(0, 0, 0, 0, 0, 0, q, p, p_mat 0, 0, 0, 0, 0, 0, q, p, 0, 0, 0, 0, 0, 0, q, p, 0, 0, 0, 0, 0, 0, q, p, 0, 0, 0, 0, q, p, 0, 0, 0, 0, q, p, 0, 0, 0, 0, 0, 0, 0, 0, q, p, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1), 8, byrow = TRUE) rownames(p_mat) <- c("HHT", "HHH", "THT", "THH", "TTT", "TTH", "HTT", "HTH") colnames(p_mat) <- rownames(p_mat) # A samc object is the core of the package <- samc(p_mat) samc_obj #>  "Warning: Some checks for manually created P matrices are still missing:" #>  "1) Discontinuous data will not work with the cond_passage() function." #>  "2) Every disconnected region of the graph must have at least one non-zero absorption value."
First, the coin flip probabilities are defined using variables, which allows us to easily change the matrix to simulate a biased coin if we want.
The last row and column of the matrix represents \(HTH\), or the absorbing state. Notice that the last row makes it so that \(HTH\) can only transition to itself, and never to a different state. This construction is explained in the Overview background.
The remaining rows/columns represent the remaining possible combinations of 3 coin flips. Technically, this model assumes our starting point includes a sequence of 3 previous flips. However, if we decide to use \(TTT\) as our initial starting point, it will take at least 3 flips to get to the target sequence, which is no different than starting from an empty sequence of flips. Contrast this with the sequence of \(HHT\), which only requires one more \(H\) flip for the \(HTH\) sequence to occur.
Finally, notice that the rows sum to
1, which is required for the \(P\) matrix. This makes sense because each row represents the transition probabilities for a state. Each state has a 100% probability of doing something, whether it’s changing to a different transient state, an absorbing state, or back to the current state.
For the original question, it’s possible to collapse some of these states and simplify the matrix by only considering the last two coint flips instead of three. For example, \(HTT\) would also require a minimum of 3 flips to reach (like \(TTT\), which has the same last two flips), so it could also be used as a starting point to answer the original question. However, the current construction simplifies the interpretation a little and allows different questions to be asked about the model.
To answer the original question of how many coin flips it will take, on average, to see \(HTH\), all we need to do is calculate the expected time to absorption. With the samc package, that is accomplished using the
# Given the last 3 flips, how many more flips until we hit HTH (absorption)? survival(samc_obj) #>  6 8 6 8 10 8 10
The fifth element represents \(TTT\) and the the seventh element represents \(HTT\), or the starting points that do not have progress towards the result. They both have a value of
10, which indicates that it will take an average of
10 coin flips with a fair coin to see the sequence of \(HTH\).
The remaining elements give us insight into what would be expected if we have already made some coin flips with some progress towards the goal. For example, the first and third elements represent, \(HHT\) and \(THT\), which are only one flip away from \(HTH\). As a result, they have the lowest expected averages to completion, which should intuitively make sense. But they still are relatively high at
6 flips expected. Even though there is a 50% probability of completing the sequence in a single flip, a failure would put us back to the \(HTT\) state, which is an average of
10 steps away.
There are other aspects of the model that can be explored using the package. For example, the
visitation() metric calculates how many times each state is expected to be visited before absorption. In the context of the coin flip model, this means we can calculate how many times we expect a sequence to occur before hitting \(HTH\).
# Given a starting point (in this case, a sequence of 3 flips), how many times # would we expect the different combinations of 3 flips to occur before absorption? visitation(samc_obj, origin = "HHT") #>  1.5 0.5 0.5 0.5 1.0 1.0 1.0 sum(visitation(samc_obj, origin = "HHT")) # Compare to survival() result #>  6
In this case, we specify a starting sequence of \(HHT\), which is one flip away from \(HTH\) (absorption). As an example interpretation, with this starting point, we expect the sequence \(HHT\) to occur one and a half times, on average, before hitting the \(HTH\) sequence. What’s interesting mathematically is that the sum of these visitation values is the same as the expected remaining flips before hitting \(HTH\).
visitation() metric produces a matrix of all the possible combinations of states, and specifying an origin point simply gives us a single row of the matrix. It’s possible to get individual columns, as well as the entire matrix (which requires disabling a safety measure in place for larger problems).
# Instead of a start point, we can look at an end point and how often we expect # it to occur for each of the possible starting points visitation(samc_obj, dest = "THT") #>  0.5 0.5 1.5 0.5 1.0 1.0 1.0 # These results are just rows/cols of a larger matrix. We can get the entire matrix # of the start/end possibilities but first we have to disable some safety measures # in place because this package is designed to work with extremely large P matrices # (millions of rows/cols) where these types of results will consume too much RAM and # crash R $override <- TRUE samc_objvisitation(samc_obj) #> [,1] [,2] [,3] [,4] [,5] [,6] [,7] #> [1,] 1.5 0.5 0.5 0.5 1 1 1 #> [2,] 1.5 2.5 0.5 0.5 1 1 1 #> [3,] 0.5 0.5 1.5 0.5 1 1 1 #> [4,] 1.5 1.5 0.5 1.5 1 1 1 #> [5,] 1.0 1.0 1.0 1.0 3 2 1 #> [6,] 1.0 1.0 1.0 1.0 1 2 1 #> [7,] 1.0 1.0 1.0 1.0 2 2 2 rowSums(visitation(samc_obj)) # equivalent to survival() above #>  6 8 6 8 10 8 10
A third relevant metric is
dispersal(). Rather than the expected number of times we would expect a coin flip sequence to occur, it calculates the probability of a sequence occurring before hitting our target sequence of \(HTH\). The short-term variant can actually be used to specify a limit to the number of flips we will me make.
dispersal(samc_obj, dest = "TTT", time = 5) #>  0.28125 0.21875 0.28125 0.21875 NA 0.21875 0.59375
In this case, we’re trying to find the probability of \(TTT\) occurring within 5 flips. The results show the probability for every possible starting sequencing. For example, the seventh element, \(HTT\) is only one flip away and has the highest probability of