17 validation framework for model

[BZ-BowenZhang]

Update a notebook for validating employment data of SPC with the BRES dataset as discussed in the issue #27 and validating the home-work places flow with census data as discussed in the issue #17

[sgreenbury]

Thanks for this @BZ-BowenZhang, adding some comments including from our discussion last week:

• The metrics and plots for validating the flows look great and look readily applicable to the full population runs for AcBM. It will be interesting to see how they compare when run with the generated activity chains with the AcBM pipeline.
• The comparison with the SPC workplace assignment is very helpful for also validating the SPC approach, thanks for including this analysis. Specifically of interest is looking at the number of jobs available in SPC from business registry (total is 671,189) compared to the business registry dataset used here (total: 1,025,985). I've pushed code to the notebook loading in the business registry data derived here for SPC and we can see that all the jobs are being assigned but this number is many fewer than expected given: 1) the alternative business registry data loaded in the notebook (1,025,985) and 2) the number with expected employment with a SIC code (1,324,680). It would be helpful to understand what is driving this difference.
• Related to the above, I was wondering whether we can see the discrepancy in the flow plots with absolute numbers. Are we able to derive the expected number of people with workplaces from the commuting flows? Since only 671,189 out of an expected number of over 1,000,000 will be commuting, the absolute number of flows I would expect to aggregate to ~300,000 less compared to the commuting flows and we might be able to see this as an overall negative value across the flow matrix for SPC commuting flows compared to census flows?