can't get keras3::layer_concatenate to work

[ChristelSwift]

I'm trying to create a biLSTM model where one feature has many categories so rather than using one-hot-encoding, I'd like to use a embedding layer. The other features are in OHE format. When i try to combine both inputs with keras3::layer_concatenate i get an error. It only works with keras::layer_concatenate.

here's a reprex:

suppressPackageStartupMessages(library(tidyverse))
library(keras3)
library(tensorflow)

# -------------------------------
# 1) Create tiny random data
# -------------------------------

set.seed(123)

n_samples <- 10
n_timesteps <- 5
n_stations_classes <- 12
embedding_dim <- 4
n_day_classes <- 7
n_hour_classes <- 24

context_dim <- n_day_classes + n_hour_classes  # 31 features

# Station IDs = integers (1..n_stations_classes)
X_station_train <- array(
  sample(1:n_stations_classes, n_samples * n_timesteps, replace = TRUE),
  dim = c(n_samples, n_timesteps)
)

X_station_val <- array(
  sample(1:n_stations_classes, n_samples * n_timesteps, replace = TRUE),
  dim = c(n_samples, n_timesteps)
)

# Context = random float32 matrix (one-hot-like but not required)
X_context_train <- array(
  runif(n_samples * n_timesteps * context_dim),
  dim = c(n_samples, n_timesteps, context_dim)
)

X_context_val <- array(
  runif(n_samples * n_timesteps * context_dim),
  dim = c(n_samples, n_timesteps, context_dim)
)

cat("station dtype:", typeof(X_station_train), " dim:", dim(X_station_train), "\n")
cat("context dtype:", typeof(X_context_train), " dim:", dim(X_context_train), "\n")


# -------------------------------
# 2) Define model (tests concatenate)
# -------------------------------

input_station <- layer_input(
  shape = c(n_timesteps),
  dtype = "int32",
  name = "station_id_input"
)

embedded_station <- input_station %>%
  layer_embedding(
    input_dim = n_stations_classes,
    output_dim = embedding_dim,
    mask_zero = TRUE,
    input_length = n_timesteps
  )

input_context <- layer_input(
  shape = c(n_timesteps, context_dim),
  dtype = "float32",
  name = "context_input"
)

# This is the operation we're testing:
combined_features <- layer_concatenate(
  list(embedded_station, input_context),
  name = "concatenate_features"
)

this gives:

Error in py_call_impl(callable, call_args$unnamed, call_args$named) : 
  KeyError: 0
Run `reticulate::py_last_error()` for details.

It only works if i specify keras::layer_concatenate...

Here's my session info:

> sessionInfo()
R version 4.5.0 (2025-04-11)
Platform: aarch64-apple-darwin20
Running under: macOS Sequoia 15.6.1

Matrix products: default
BLAS:   /System/Library/Frameworks/Accelerate.framework/Versions/A/Frameworks/vecLib.framework/Versions/A/libBLAS.dylib 
LAPACK: /Library/Frameworks/R.framework/Versions/4.5-arm64/Resources/lib/libRlapack.dylib;  LAPACK version 3.12.1

locale:
[1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8

time zone: Europe/London
tzcode source: internal

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

other attached packages:
 [1] caret_7.0-1       lattice_0.22-7    patchwork_1.3.2   glue_1.8.0        readxl_1.4.5      tensorflow_2.20.0 keras3_1.4.0      RJDBC_0.2-10      rJava_1.0-11     
[10] DBI_1.2.3         aws.s3_0.3.22     lubridate_1.9.4   forcats_1.0.0     stringr_1.5.2     dplyr_1.1.4       purrr_1.1.0       readr_2.1.5       tidyr_1.3.1      
[19] tibble_3.3.0      ggplot2_4.0.0     tidyverse_2.0.0  

loaded via a namespace (and not attached):
 [1] tidyselect_1.2.1     timeDate_4051.111    farver_2.1.2         S7_0.2.0             fastmap_1.2.0        pROC_1.19.0.1        digest_0.6.37        rpart_4.1.24        
 [9] timechange_0.3.0     lifecycle_1.0.4      survival_3.8-3       magrittr_2.0.4       compiler_4.5.0       rlang_1.1.6          tools_4.5.0          data.table_1.17.8   
[17] knitr_1.50           labeling_0.4.3       curl_7.0.0           here_1.0.2           reticulate_1.43.0    aws.signature_0.6.0  plyr_1.8.9           xml2_1.4.0          
[25] RColorBrewer_1.1-3   keras_2.16.0         withr_3.0.2          stats4_4.5.0         nnet_7.3-20          grid_4.5.0           e1071_1.7-16         future_1.67.0       
[33] globals_0.18.0       scales_1.4.0         iterators_1.0.14     MASS_7.3-65          zeallot_0.2.0        cli_3.6.5            generics_0.1.4       rstudioapi_0.17.1   
[41] future.apply_1.20.0  reshape2_1.4.4       httr_1.4.7           tzdb_0.5.0           tfruns_1.5.4         proxy_0.4-27         splines_4.5.0        parallel_4.5.0      
[49] cellranger_1.1.0     base64enc_0.1-3      vctrs_0.6.5          hardhat_1.4.2        Matrix_1.7-4         jsonlite_2.0.0       hms_1.1.3            listenv_0.9.1       
[57] foreach_1.5.2        gower_1.0.2          recipes_1.3.1        parallelly_1.45.1    codetools_0.2-20     stringi_1.8.7        gtable_0.3.6         pillar_1.11.1       
[65] ipred_0.9-15         lava_1.8.2           R6_2.6.1             rprojroot_2.1.1      evaluate_1.0.5       png_0.1-8            class_7.3-23         Rcpp_1.1.0          
[73] dotty_0.1.0          nlme_3.1-168         prodlim_2025.04.28   mgcv_1.9-3           whisker_0.4.1        xfun_0.53            ModelMetrics_1.2.2.2 pkgconfig_2.0.3    

[t-kalinowski]

Thanks for the report!

This is a bug in how layer_concatenate() processes a named argument like name = "foo".

Until this is fixed you can workaround by calling keras$layers$concatenate() like this:

combined_features <- keras$layers$concatenate(
  list(embedded_station, input_context),
  name = "concatenate_features"
)