Mining Finite-State Automata with Annotations - Mining Software Specifications: Methodologies and Applications - page 49

TABLE 2.5: Automatically Identify Data Clusters and Rewriting Strategies
#

Events Attributes Cl. Actual

Values

Symb.
val.

Rewritten trace

1 hasNext hasNext

2 next return.availSeats A 4 0 next A0 B0

return.flightNo B KL1022 0

3 getAvailableSeats return A 4 1 getAvailableSeats A1

4 add flight.availSeats A 4 1 add A1 B1

ight.ightNo B KL1022 1

5 hasNext hasNext

6 next return.availSeats A 7 0 next A0 B0

return.flightNo B KL1028 0

7 getAvailableSeats return A 7 1 getAvailableSeats A1

8 add flight.availSeats A 7 1 add A1 B1

ight.ightNo B KL1028 1

9 hasNext hasNext

10 next return.availSeats A 4 4 next A4 B0

return.flightNo B KL1058 0

11 getAvailableSeats return A 4 1 getAvailableSeats A1

12 add flight.availSeats A 4 1 add A1 B1

ight.ightNo B KL1058 1

13 hasNext hasNext

Legend
Cl.

data cluster

Symb. val. symbolic value

Table 2.5 shows some of the events recorded during an execution of the
running example. KLFA automatically identifies two data clusters: one that
groups attributes that represent the number of available seats (indicated as
A in column Cl.) and one that groups the flight numbers (indicated as B in
column Cl.). After the application of the three rewriting strategies to these
two data clusters, KLFA is able to identify the rewriting strategy that better
rewrites the concrete values. The value of the flight number changes at every
iteration, thus the relative to instantiation rewriting strategy works better
than the other strategies. The number of available seats tends to be the same
for different flights, thus the relative to access rewriting strategy better adapts
to this cluster.
Column Rewritten trace shows how KLFA rewrites event names and at-
tributes. Each rewritten event consists of the event name followed by the
symbolic values of the parameters separated by underscores. To ease the un-
derstanding of the data-flow relations among attributes, KLFA specifies the
name of the data cluster before the symbolic value of each attribute.
In some cases, data clusters can include attribute values whose distribution
does not match the requirements of any of our strategies, producing in poor
results. To avoid poor results due to inadequate strategies, KLFA applies a
rewriting strategy for a data-flow cluster only if at least 50% of the attribute
values can be rewritten by using at most 10 symbols (we empirically derived