final model runs - Githubissues

sdtaylor / phenology_dataset_study

1 stars 1 forks source link

final model runs #24

Closed sdtaylor closed 5 years ago

sdtaylor commented 6 years ago

Will be going through modelling 1 more time to deal with #22, #17, and #23

[x] implement hold out data for #22

Without grouping sites together

[x] process data from scratch
[x] run models on hipergator
[x] do predictions
[x] make graphs

With grouping sites together

[x] process data from scratch
[x] run models on hipergator
[x] do predictions
[x] make graphs * minimum observation cutoff for Harvard with this is set to 20 instead of 30 * total species comparisons drops to 20 from 38 because of tighter restrictions

sdtaylor commented 6 years ago

The graphs between the two grouping strategies are essentially identical, hence interpretation would be the same. I'll use the results with un-grouped data. The results grouped together can go in a supplement if someone asks. See #23

sdtaylor commented 6 years ago

Note the final run using non-grouping took 35.5 hours with 844 cpus, roughly ~3.5 years of CPU time

sdtaylor commented 6 years ago

Adding spatial models #29

[x] MSB
[x] M1

Redo just these two models without grouping sites

[x] process data from scratch
[x] run models on hipergator
[x] do predictions
[x] make graphs

sdtaylor commented 6 years ago

One more time to try a shorter threshold in the NPN data. ie only keep observations with a prior "no" at most 15 days before. 15 days is the threshold used by Crimmins et al. 2017

Not grouping sites together

[x] process data from scratch
[x] run models on hipergator
[x] do predictions
[x] make graphs

sdtaylor commented 5 years ago

done and done