Issue with az.plot_ppc Showing a Single Straight Line in Non-linear Differential Equations

Questions v5
,
3

Thanks to @cluhmann for pointing me to the post! Following the advice there, I managed to partially resolve the issue—I changed my code like this:

    def analysis(self, trace):
        # result
        az.plot_trace(trace)
        
        # print
        print(az.summary(trace))
        az.plot_forest(trace, r_hat=True)
        az.plot_posterior(trace)

        with self.model:
            check = pm.sample_posterior_predictive(trace,extend_inferencedata=True)
        fig1, ax1 = plt.subplots()
        fig2, ax2 = plt.subplots()

        ax = az.plot_ppc(check, alpha=0.025, kind='kde', observed=True, ax=[ax1,ax2])
        for i, ax in enumerate(ax.flat):
            ax.set_xlim([-0.002,0.002])
        
        plt.show()

image
image1830×768 38.9 KB

now the observed data curve is no longer a straight line. However, I’m still facing some problems:

  1. Despite ‘fixing’(actually, it’s just amplification) the observed data curve, the other posterior predictive curves are still appearing as single straight lines. This seems unusual, and I’m not sure why only the observed data was corrected while the other two remain problematic.Was my data processing methods towards covid_new and influenza_new went wrong?

  2. Although the observed data curve is now showing properly in the az.plot_ppc output, it doesn’t match the curve I get when I plot the observed data directly using the same data processing code. The two curves it should be is like this:

    image
    image646×548 39.9 KB

If you need more and more detailed code, please feel free to let me know.

Thanks again for your help! I’ve been struggling with this issue for weeks :frowning: , any insights or suggestions on how to troubleshoot this would be greatly appreciated.