if batch = TRUE in training, fgraph of umap_transform is transposed if n_epoch is set to 0

[PedroMilanezAlmeida]

on my system, this is a reproducible example:

iris_train <- iris[1:100,1:4]
iris_test <- iris[101:150,1:4]

iris_train_umap <- uwot::umap(iris_train, 
                              ret_model = TRUE,
                              batch = TRUE)

dim(uwot::umap_transform(iris_test, 
                         iris_train_umap,
                         n_epochs = NULL,
                         ret_extra = "fgraph")$fgraph)
[1] 100  50

dim(uwot::umap_transform(iris_test, 
                         iris_train_umap,
                         n_epochs = 0,
                         ret_extra = "fgraph")$fgraph)
[1]  50 100

R version 4.3.2 (2023-10-31)
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: CentOS Linux 7 (Core)

Matrix products: default
BLAS/LAPACK: FlexiBLAS OPENBLAS;  LAPACK version 3.11.0

locale:
 [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C               LC_TIME=en_US.UTF-8        LC_COLLATE=en_US.UTF-8     LC_MONETARY=en_US.UTF-8   
 [6] LC_MESSAGES=en_US.UTF-8    LC_PAPER=en_US.UTF-8       LC_NAME=C                  LC_ADDRESS=C               LC_TELEPHONE=C            
[11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C       

time zone: America/New_York
tzcode source: system (glibc)

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

loaded via a namespace (and not attached):
 [1] microbenchmark_1.4.10 compiler_4.3.2        RcppAnnoy_0.0.21      Matrix_1.6-4          tools_4.3.2           rstudioapi_0.15.0    
 [7] Rcpp_1.0.11           codetools_0.2-19      grid_4.3.2            snedata_0.0.0.9000    lattice_0.22-5        uwot_0.1.16

[PedroMilanezAlmeida]

fgraph of umap_transform also transposed if batch = TRUE and n_epochs = NULL

iris_train_umap <- uwot::umap(iris_train, 
                              ret_model = TRUE,
                              batch = FALSE)

iris_test_umap <- uwot::umap_transform(iris_test, 
                                       iris_train_umap,
                                       n_epochs = NULL,
                                       ret_extra = "fgraph")
dim(iris_test_umap$fgraph)
[1]  50 100

iris_train_umap <- uwot::umap(iris_train, 
                              ret_model = TRUE,
                              batch = TRUE)

iris_test_umap <- uwot::umap_transform(iris_test, 
                                       iris_train_umap,
                                       n_epochs = NULL,
                                       ret_extra = "fgraph")
dim(iris_test_umap$fgraph)
[1] 100  50

[jlmelville]

Thank you for the report! This was due to trying to avoid doing unnecessary processing when the UMAP optimization won't occur (but some of it actually is necessary either way). Should be fixed in the dev version now.

[jlmelville]

Will be fixed in the next CRAN release

[PedroMilanezAlmeida]

amazing, thank you!

[PedroMilanezAlmeida]

fgraph of umap_transform also transposed if batch = TRUE and n_epochs = NULL

iris_train_umap <- uwot::umap(iris_train, 
                              ret_model = TRUE,
                              batch = FALSE)

iris_test_umap <- uwot::umap_transform(iris_test, 
                                       iris_train_umap,
                                       n_epochs = NULL,
                                       ret_extra = "fgraph")
dim(iris_test_umap$fgraph)
[1]  50 100

iris_train_umap <- uwot::umap(iris_train, 
                              ret_model = TRUE,
                              batch = TRUE)

iris_test_umap <- uwot::umap_transform(iris_test, 
                                       iris_train_umap,
                                       n_epochs = NULL,
                                       ret_extra = "fgraph")
dim(iris_test_umap$fgraph)
[1] 100  50

The code above gives the transposed fgraph in my system again (the only difference in that batch is FALSE in the first uwot::umap call and TRUE in the second).

R version 4.4.1 (2024-06-14)
Platform: x86_64-pc-linux-gnu
Running under: CentOS Linux 7 (Core)

Matrix products: default
BLAS/LAPACK: FlexiBLAS OPENBLAS;  LAPACK version 3.11.0

locale:
 [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C               LC_TIME=en_US.UTF-8        LC_COLLATE=en_US.UTF-8    
 [5] LC_MONETARY=en_US.UTF-8    LC_MESSAGES=en_US.UTF-8    LC_PAPER=en_US.UTF-8       LC_NAME=C                 
 [9] LC_ADDRESS=C               LC_TELEPHONE=C             LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C       

time zone: America/New_York
tzcode source: system (glibc)

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
[1] uwot_0.2.2   Matrix_1.7-0

loaded via a namespace (and not attached):
[1] compiler_4.4.1    RcppAnnoy_0.0.22  tools_4.4.1       rstudioapi_0.16.0 Rcpp_1.0.12       codetools_0.2-20 
[7] grid_4.4.1        lattice_0.22-6

[jlmelville]

Yeah I messed this one up and results were inconsistent between batch = TRUE and batch = FALSE.

@PedroMilanezAlmeida I think I should have been clearer about my understanding of what the correct dimensions for fgraph is. Where there were 100 training points and 50 test points like in your code I think the most sensible way to arrange the dimensions for fgraph should be (50, 100) and not (100, 50). Do you agree? If so, I got this completely the wrong way round with the original "fix".

[jlmelville]

@PedroMilanezAlmeida I've pushed a change (and added more tests) that should ensure that the output dim for fgraph would always be (50, 100). Let me know if that's contrary to your expectations, otherwise I will eventually close this. Not sure when the next release of uwot will be.

[PedroMilanezAlmeida]

Ugh sorry should have gotten back to you asap. I am not partial to which dimension (rows or cols) the mapped obs “should” be. Just consistency would be enough 👍 thanks again for the prompt response and action!

…

Sent from my phone

On Sat, Sep 7, 2024 at 12:00 James Melville ***@***.***> wrote: @PedroMilanezAlmeida <https://github.com/PedroMilanezAlmeida> I've pushed a change (and added more tests) that should ensure that the output dim for fgraph would always be (50, 100). Let me know if that's contrary to your expectations, otherwise I will eventually close this. Not sure when the next release of uwot will be. — Reply to this email directly, view it on GitHub <#118 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ADGZB7JZB6SCMUEJPZ62KP3ZVMPILAVCNFSM6AAAAABERAJJN2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGMZVGY4DOOJTHA> . You are receiving this because you were mentioned.Message ID: ***@***.***>