Save/load a study

The objective of this example is to demonstrate how to save the structures created within a script session to disk in order to be able to load them in a future session.

There are several possible ways to achieve this:

  • with the standard pickle module

  • with openturns’s Study

import openturns as ot
import pickle


create objects to save

distribution = ot.Normal(4.0, 3.0)
function = ot.SymbolicFunction(["x1", "x2"], ["x1 + x2"])

With the pickle module

The objects are retrieved in the same order they are stored.

save objects

with open("study.pkl", "wb") as f:
    pickle.dump(distribution, f)
    pickle.dump(function, f)

load saved objects

with open("study.pkl", "rb") as f:
    loaded_distribution = pickle.load(f)
    loaded_function = pickle.load(f)
str(loaded_distribution), str(loaded_function)
('Normal(mu = 4, sigma = 3)', '[x1,x2]->[x1 + x2]')

With OpenTURNS’ Study

In order to be able to manipulate the objects contained in a Study, it is necessary to:

  • create the same empty structure in the new study,

  • fill this new empty structure with the content of the loaded structure, identified with its name or its id.

Each object is identified whether with:

  • its name: it is useful to give names to the objects we want to save. If no name has been given by the user, we can use the default name. The name of each saved object can be checked in the output XML file or with the python print command (applied to the Study object).

  • its id number: this id number is unique to each object. It distinguishes objects with identical type and name (like the default name “Unnamed”). This id number may be checked by printing the study after it has been loaded in the python interface (with the print command). It can differ from the id number indicated in the XML file the study was loaded from.

  • for HDF5 storage (see below): the id serves both as xml id and hdf5 dataset name. Id uniqueness forbids any misleading in reading/writing hdf5 datasets.

Create a Study Object

study = ot.Study()

Associate it to an XML file

fileName = "study.xml"

Alternatively, large amounts of data can be stored in binary HDF5 file. An XML file (study_h5.xml) serves as header for binary data, which are stored in the automatically created study_h5.h5 file.

study_h5 = ot.Study()
fileName_h5 = "study_h5.xml"

Add an object to the study; at this point it is not written to disk yet

study.add("distribution", distribution)
study.add("function", function)

Save the study; this writes into the file

Create a new study associated to the same file

study = ot.Study()

Load the file and all its objects


Check the content of the myStudy

print("Study = ", study)
Study =    1075 => FunctionImplementation
[x1,x2]->[x1 + x2]
  1076 => Normal
Normal(mu = 4, sigma = 3)
  'distribution' is aliased to 1076
  'function' is aliased to 1075

List names of stored objects


Check our ‘distribution’ labelled object was loaded


Load the objects; we must create a void object of the desired type (or parent type)

distribution2 = ot.Normal()
function2 = ot.Function()
study.fillObject("distribution", distribution2)
study.fillObject("function", function2)
str(distribution2), str(function2)
('Normal(mu = 4, sigma = 3)', '[x1,x2]->[x1 + x2]')