SocialBehavior

Social Bebehavior Modeling of 2-mice data

• SocialBehaviorptc

  PyTorch implementation of state-space models, build on the ssm package.

  Currently, we have

  • ARHMM with Normal emission notebook demo

  • ARHMM with Sigmoid-Normal emission notebook demo

    • Sigmoid-Noral distribution:

      • parameters: $\mu, \sigma$, lower & upper bounds $l,u$.

      • generative process:

        $z \sim N(0,1)​$

        $x = (u-l) * \sigma (\mu + \sigma z) + l ​$

      • In training the real data, another parameter $\alpha$ can be added: $x = (u-l) * \sigma (\alpha(\mu + \sigma z)) + l$ to tune the initialization.

      • See how the parameters affect the distribution here SigmoidNormal distribution demo

• check how the models work on mice data project notebooks

  • ARHMM with Gaussian observation notebook1, 2 (different learing rates)

  • ARHMM with Sigmoid-Normal observation notebook1, 2