SciUnit Logo

Open In Colab

3. Chapter 3. Testing with help from the SciUnit standard library

(or back to Chapter 2)

3.1. If you are using this file in Google Colab, this block of code can help you install sciunit from PyPI in Colab environment.

!pip install -q sciunit
import sciunit

3.2. In this chapter we will use the same toy model in Chapter 1 but write a more interesting test with additional features included in SciUnit.

from sciunit.models.examples import ConstModel # One of many dummy models included for illustration.  
const_model_37 = ConstModel(37, name="Constant Model 37")

Now let’s write a test that validates the observation and returns more informative score type.

from sciunit.capabilities import ProducesNumber
from sciunit.scores import ZScore # One of many SciUnit score types.  
from sciunit.errors import ObservationError # An exception class raised when a test is instantiated 
                                     # with an invalid observation.
    
class MeanTest(sciunit.Test):
    """Tests if the model predicts 
    the same number as the observation."""   
    
    required_capabilities = (ProducesNumber,) # The one capability required for a model to take this test.  
    score_type = ZScore # This test's 'judge' method will return a BooleanScore.  
    
    def validate_observation(self, observation):
        if type(observation) is not dict:
            raise ObservationError("Observation must be a python dictionary")
        if 'mean' not in observation:
            raise ObservationError("Observation must contain a 'mean' entry")
        
    def generate_prediction(self, model):
        return model.produce_number() # The model has this method if it inherits from the 'ProducesNumber' capability.
    
    def compute_score(self, observation, prediction):
        score = ZScore.compute(observation,prediction) # Compute and return a ZScore object.  
        score.description = ("A z-score corresponding to the normalized location of the observation "
                             "relative to the predicted distribution.")
        return score

We’ve done two new things here:

  • The optional validate_observation method checks the observation to make sure that it is the right type, that it has the right attributes, etc. This can be used to ensures that the observation is exactly as the other core test methods expect. If we don’t provide the right kind of observation:

-> mean_37_test = MeanTest(37, name='=37')
ObservationError: Observation must be a python dictionary

then we get an error. In contrast, this is what our test was looking for:

observation = {'mean':37.8, 'std':2.1}
mean_37_test = MeanTest(observation, name='=37')
  • Instead of returning a BooleanScore, encoding a True/False value, we return a ZScore encoding a more quantitative summary of the relationship between the observation and the prediction. When we execute the test:

score = mean_37_test.judge(const_model_37)

Then we get a more quantitative summary of the results:

score.summarize()
score.describe()
'A z-score corresponding to the normalized location of the observation relative to the predicted distribution.'