I create a new version of Expected Improvement for my problem.
When I run it I obtain the next configuration to analyse in the model.
Then an error occur in this line :
Algorithm.update_model(next_config)
where my update_model function is defined like this:
def update_model(self, cell_config):
self.model = self.model.condition_on_observations(cell_config.get_tilt(), cell_config.get_reward())
mll = ExactMarginalLogLikelihood(self.model.likelihood, self.model)
fit_gpytorch_model(mll)
return self.model
The error is this one:
RuntimeError: Fantasy observations can only be added after making predictions with a model so that all test independent caches exist. Call the model on some data first!
The error message seems to be raised by a method, which is specific to the repository you are using, so I would recommend to ask the authors or create an issue in the repository directly. 
That being said, based on the error message it seems you would have to call the model with data before calling the code snippet, which is raising this error.