A SciUnit library for validation testing of neural network models.
A SciUnit library for validation testing of spiking networks.
pip install networkunit
The NetworkUnit module builds upon the formalized validation scheme of the SciUnit package,
which enables the validation of models against experimental data (or other models) via tests.
A test is matched to the model by capabilities and quantitatively evaluated by a score.
The following figure illustrates a typical test design within NetworkUnit.
The blue boxes indicate the components of the implementation of the validation test, i.e.,
classes, class instances, data sets, and parameters.
The relation between the boxes are indicated by annotated arrows.The basic functionality is
shown by green arrows. The difference in the test design for comparing against experimental
data (validation) and another simulation (substantiation) is indicated by yellow and
red arrows, respectively. The relevant functionality of some components for the
computation of test score is indicated by pseudo-code. The capability
ProducesProperty contains the function
calc_property(). The test
XYTest has a function
generate_prediction() which makes use of this capability, inherited by the model class,
to generate a model prediction. The initialized test instance
XYTest_paramZ makes use of its
judge() function to evaluate this model prediction and compute the score
XYTest can inherit from multiple abstract test classes (
which is for example used with the
M2MTest to add the functionality of evaluating multiple model classes.
To make the test executable it has to be linked to a ScoreType and all free parameters need to be set
Params dict) to ensure a reproducible result.
Showcase examples on how to use NetworkUnit can be found in this repository and interactive reveal.js slides are accessible via the launch-binder button at the top.
Overview of tests
|Class name||Parent class||Prediction measure|
|correlation_matrix_test||correlation_test||correlation coefficient matrix|
|generalized_correlation_matrix_test||correlation_matrix_test||matrix of derived cross-correlation measures|
|eigenvalue_test||correlation_test||eigenvalues of the correlation coefficient matrix|
|isi_variation_test||two_sample_test||inter-spike-intervals, their CV, or LV|
|graph_centrality_helperclass||sciunit.Test||graph centrality measures of given adjacency matrix|
Inheritance order in case of multiple inheritance for derived test classes:
class new_test(sciunit.TestM2M, graph_centrality_helperclass, <base_test_class>)
Overview of scores
|Class name||Test name||Comparison measure|
|students_t||Student’t test||sample mean|
|ks_distance||Kolmogorov-Smirnov test||sample distribution|
|kl_divergence||Kullback-Leibler divergence||sample entropy|
|mwu_statistic||Mann-Whitney U test||rank sum|
|levene_score||Levene’s test||sample variance|
|effect_size||Effect size||standardized mean|
|best_effect_size||Bayesian estimation effect size||standardized mean|
Overview of model classes
|Model name||Capability||Parent class||Purpose|
|loaded_data||-||sciunit.Model||loading simulated data|
|spiketrain_data||ProducesSpikeTrains||simulation_data||loading simulated spiking data|
|stochastic_activity||ProducesSpikeTrains||sciunit.Model||generating stochastic spiking data|
This open source software code was developed in part or in whole in the Human Brain Project, funded from the European Union’s Horizon 2020 Framework Programme for Research and Innovation under Specific Grant Agreements No. 720270 and No. 785907 (Human Brain Project SGA1 and SGA2).
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
|Filename, size||File type||Python version||Upload date||Hashes|
|Filename, size networkunit-0.1.1-py3-none-any.whl (2.6 MB)||File type Wheel||Python version py3||Upload date||Hashes View|
|Filename, size networkunit-0.1.1.tar.gz (2.5 MB)||File type Source||Python version None||Upload date||Hashes View|
Hashes for networkunit-0.1.1-py3-none-any.whl