get_unique_name() in node_class.R might not be unique

[njtierney]

Perhaps this is not likely to happen, or for this to be an issue, but it seems that the rhex() function as defined isn't gauranteed to create a unique name if there are many many nodes (like 1 million).

This is used in node_class.R.

See example below.

n_rhex <- 1e6

# generate a random 8-digit hexadecimal string
rhex <- function() paste(as.raw(sample.int(256L, 4, TRUE) - 1L), collapse = "")

many_rhex <- replicate(n = n_rhex, expr = rhex(), simplify = "vector")

dplyr::n_distinct(many_rhex)
#> [1] 999874

dplyr::n_distinct(many_rhex) == n_rhex
#> [1] FALSE

^{Created on 2021-04-08 by the reprex package (v2.0.0)}

Perhaps digest or something like https://github.com/coolbutuseless/xxhashlite could be used to give nodes unique IDs

[njtierney]

Or https://github.com/reside-ic/ids or do something like what reprex did (https://github.com/tidyverse/reprex/blob/d2996e01f045b04cd537653a39deece1025dbf35/R/aaa.R), btu this might not be unique enough.

[njtierney]

this is currently being worked on here https://github.com/njtierney/greta/tree/unique-names

[njtierney]

• stopgap solution - create error to catch unique names of nodes - basically tell the user there is something wrong with duplicated node names and that rerunning the model should fix this.
• create unit test to test the mechanism that generates node names and that they are unique
• test if loading multiple greta objects from past environments results in mixed up / wrongly defined nodes/models

[njtierney]

There is an issue where an error appears:

Error in distrib_constructor(tf_parameter_list, dag = self) : could not find function "distrib_constructor"

Which means it is not finding

greta/R/dag_class.R

Line 410 in 112a968

distrib_constructor <- self$get_tf_object(distribution_node)

It makes me wonder if perhaps this is related to this issue. We have not been able to reliably develop a small reprex for this issue, so it might not be related to this one.