Browse Source

Skip some checks in doctests because the generated output is not deterministic

refactoring
Matthias Volk 4 years ago
parent
commit
4ac88e934f
No known key found for this signature in database GPG Key ID: 83A57678F739FCD3
  1. 8
      doc/source/doc/models/building_ctmcs.rst
  2. 6
      doc/source/doc/models/building_dtmcs.rst
  3. 24
      doc/source/doc/models/building_mas.rst
  4. 21
      doc/source/doc/models/building_mdps.rst

8
doc/source/doc/models/building_ctmcs.rst

@ -30,7 +30,7 @@ In this example, we build the transition matrix using a numpy array::
The following function call returns a sparse matrix with default row groups::
>>> transition_matrix = stormpy.build_sparse_matrix(transitions)
>>> print(transition_matrix)
>>> print(transition_matrix) # doctest: +SKIP
0 1 2 3
---- group 0/3 ----
0 ( 0 1.5 0 0 ) 0
@ -67,13 +67,13 @@ Building the Model
Now, we can collect all components, including the choice labeling and the exit rates.
To let the transition values be interpreted as rates we set `rate_transitions` to `True`::
components = stormpy.SparseModelComponents(transition_matrix=transition_matrix, state_labeling=state_labeling, rate_transitions=True)
components.exit_rates = exit_rates
>>> components = stormpy.SparseModelComponents(transition_matrix=transition_matrix, state_labeling=state_labeling, rate_transitions=True)
>>> components.exit_rates = exit_rates
And finally, we can build the model::
>>> ctmc = stormpy.storage.SparseCtmc(components)
>>> print(ctmc)
>>> print(ctmc) # doctest: +SKIP
--------------------------------------------------------------
Model type: CTMC (sparse)
States: 4

6
doc/source/doc/models/building_dtmcs.rst

@ -82,7 +82,7 @@ Labels can be asociated with states. As an example, we label the state 0 with "i
>>> state_labeling.add_label_to_state('init', 0)
>>> print(state_labeling.get_states('init'))
bit vector(1/13) [0]
bit vector(1/13) [0 ]
Next, we set the associations between the remaining labels and states.::
@ -96,7 +96,7 @@ Next, we set the associations between the remaining labels and states.::
To set the same label for multiple states, we can use a BitVector representation for the set of states::
>>> state_labeling.set_states('done', stormpy.BitVector(13, [7, 8, 9, 10, 11, 12]))
>>> print(state_labeling)
>>> print(state_labeling) # doctest: +SKIP
9 labels
* one -> 1 item(s)
* four -> 1 item(s)
@ -131,7 +131,7 @@ Next, we collect all components::
And finally, we can build the DTMC::
>>> dtmc = stormpy.storage.SparseDtmc(components)
>>> print(dtmc)
>>> print(dtmc) # doctest: +SKIP
--------------------------------------------------------------
Model type: DTMC (sparse)
States: 13

24
doc/source/doc/models/building_mas.rst

@ -39,7 +39,7 @@ In the beginning, we create a numpy array that will be used to build the transit
When building the matrix we define a custom row grouping by passing a list containing the starting row of each row group in ascending order::
>>> transition_matrix = stormpy.build_sparse_matrix(transitions, [0, 2, 3, 4, 5])
>>> print(transition_matrix)
>>> print(transition_matrix) # doctest: +SKIP
0 1 2 3 4
---- group 0/4 ----
@ -65,17 +65,33 @@ we build a BitVector that contains the respective Markovian states::
Building the Model
====================
.. testsetup::
# Not displayed in documentation but needed for doctests
>>> state_labeling = stormpy.storage.StateLabeling(5)
>>> state_labels = {'init', 'deadlock'}
>>> for label in state_labels:
... state_labeling.add_label(label)
>>> state_labeling.add_label_to_state('init', 0)
>>> choice_labeling = stormpy.storage.ChoiceLabeling(6)
>>> choice_labels = {'alpha', 'beta'}
>>> for label in choice_labels:
... choice_labeling.add_label(label)
>>> choice_labeling.add_label_to_choice('alpha', 0)
>>> choice_labeling.add_label_to_choice('beta', 1)
>>> exit_rates = [0.0, 10.0, 12.0, 1.0, 1.0]
Now, we can collect all components::
>>> components = stormpy.SparseModelComponents(transition_matrix=transition_matrix, state_labeling=state_labeling,
markovian_states=markovian_states)
>>> components = stormpy.SparseModelComponents(transition_matrix=transition_matrix, state_labeling=state_labeling, markovian_states=markovian_states)
>>> components.choice_labeling = choice_labeling
>>> components.exit_rates = exit_rates
Finally, we can build the model::
>>> ma = stormpy.storage.SparseMA(components)
>>> print(ma)
>>> print(ma) # doctest: +SKIP
--------------------------------------------------------------
Model type: Markov Automaton (sparse)
States: 5

21
doc/source/doc/models/building_mdps.rst

@ -82,7 +82,7 @@ Recall that those actions where defined in row one and two of the transition mat
>>> choice_labeling.add_label_to_choice('a', 0)
>>> choice_labeling.add_label_to_choice('b', 1)
>>> print(choice_labeling)
>>> print(choice_labeling) # doctest: +SKIP
Choice 2 labels
* a -> 1 item(s)
* b -> 1 item(s)
@ -99,6 +99,23 @@ In this reward model the length of the action rewards coincides with the number
Building the Model
====================
.. testsetup::
# Not displayed in documentation but needed for doctests
>>> state_labeling = stormpy.storage.StateLabeling(13)
>>> labels = {'init', 'one', 'two', 'three', 'four', 'five', 'six', 'done', 'deadlock'}
>>> for label in labels:
... state_labeling.add_label(label)
>>> state_labeling.add_label_to_state('init', 0)
>>> state_labeling.add_label_to_state('one', 7)
>>> state_labeling.add_label_to_state('two', 8)
>>> state_labeling.add_label_to_state('three', 9)
>>> state_labeling.add_label_to_state('four', 10)
>>> state_labeling.add_label_to_state('five', 11)
>>> state_labeling.add_label_to_state('six', 12)
>>> state_labeling.set_states('done', stormpy.BitVector(13, [7, 8, 9, 10, 11, 12]))
We collect the components::
>>> components = stormpy.SparseModelComponents(transition_matrix=transition_matrix, state_labeling=state_labeling, reward_models=reward_models, rate_transitions=False)
@ -107,7 +124,7 @@ We collect the components::
We build the model::
>>> mdp = stormpy.storage.SparseMdp(components)
>>> print(mdp)
>>> print(mdp) # doctest: +SKIP
Model type: MDP (sparse)
States: 13
Transitions: 22

Loading…
Cancel
Save