The flexibility that can be seen below can be exploited on these contexts:

Running some experiments from config files in yaml/json
Automatize them and use the json representation to send them as messages
Attach metadata to the experiment (use data stores)



In [2]:

    
from fito import as_operation, SpecField, PrimitiveField, Operation
from time import sleep

class DatabaseConnection(Operation):
    host = PrimitiveField(pos=0)
    
    def __repr__(self): return "connection(db://{})".format(self.host)

@as_operation(database=SpecField, experiment_config=SpecField)
def run_experiment(database, experiment_config):
    sleep(0.25)

@as_operation()
def experiment_1(alpha=0.5, beta=10):
    return alpha + beta / 2

It's __repr__ is descent



In [4]:

    
run = run_experiment(DatabaseConnection('localhost'), experiment_1(alpha=10))
print run









    



run_experiment(database=connecti..., experiment_config...)

Json serializable:



In [5]:

    
print run.json.dumps()









    



{
  "experiment_config": {
    "alpha": 10, 
    "beta": 10, 
    "type": "experiment_1"
  }, 
  "type": "run_experiment", 
  "database": {
    "host": "localhost", 
    "type": "DatabaseConnection"
  }
}

My beloved yaml-serializable



In [6]:

    
print run.yaml.dumps()









    



database:
  host: localhost
  type: DatabaseConnection
experiment_config:
  alpha: 10
  beta: 10
  type: experiment_1
type: run_experiment

And of course, parseable :)



In [7]:

    
print Operation.from_yaml(run.yaml.dumps()).yaml.dumps()









    



database:
  host: localhost
  type: DatabaseConnection
experiment_config:
  alpha: 10
  beta: 10
  type: experiment_1
type: run_experiment