Skip to content
/ dagsim Public

A framework and specification language for simulating data based on graphical models

License

Notifications You must be signed in to change notification settings

uio-bmi/dagsim

Repository files navigation

DagSim

Binder

DagSim is a Python-based framework and specification language for simulating data based on a Directed Acyclic Graph ( DAG) structure, without any constraints on variable types or functional relations. A succinct YAML format for defining the structure of the simulation model promotes transparency, while separate user-provided functions for generating each variable based on its parents ensure the modularization of the simulation code.

Link to the paper: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0284443

Installation

DagSim can be easily installed using pip.

Installing DagSim using pip

To install the DagSim package using pip, run:

pip install dagsim

Quickstart

To check that DagSim is installed properly, run the following command in the console/terminal:

dagsim-quickstart

Installing graphviz

If you use pip, you need to install graphviz on the system level in order to use the drawing functionality in DagSim. Please follow the instrcutions here on how to install graphviz depending on the operating system.

Simple example

Python code

Suppose we are interested in simulating two variables, X and Y, where X follows a standard Gaussian distribution, and Y is the square of X.

For each node we need a function to simulate the node's values:

  • For X, we can use the numpy.random.normal function
  • For Y, we can use either numpy.power or define our own function. We will use the second to illustrate how one can use user-define functions.
# needed imports
import dagsim.base as ds
import numpy as np

Here, we define our own square function:

def square(arg):
    return arg * arg

Then, we define the nodes in our graph/model by giving each node a name, the function to use in order to evaluate its value, and the arguments of the function, if any:

X = ds.Node(name="X", function=np.random.normal)
Y = ds.Node(name="Y", function=square, kwargs={"arg": X})

After that, we define the graph itself by giving it a name (optional) and a list of all the nodes to be included:

graph = ds.Graph(name="demo_graph", list_nodes=[X, Y])

If you wish, you can draw the graph by calling the draw method, as follows:

graph.draw()

Finally, we simulate data from this graph by calling the simulate method, and giving it the number of samples you want to simulate, and a name for the csv_file (optional) where the data should be saved.

data = graph.simulate(num_samples=10, csv_name="demo_data")

Here, data would be a dictionary with keys being the names of the nodes in the graph, and the corresponding values being the simulated values for each node returned as a Python list.

For more detailed instructions, check this page, and for other simple examples, please refer to the tutorials folder.

YAML Specification

dagsim also allows the specification of a simulation using a YAML file. You can run dagsim on a YAML file by running:

dagsim path/to/yaml/file [-v|--verbose] [-d|--draw] [-o output/path|--output_path=output/path]

For a tutorial on using a YAMl file for simulation, check this page.