[GitHub] ironfrown opened a new issue #8545: Incorrect results from R 3.4.2 in MNIST

[GitBox <gi...@apache.org>]

ironfrown opened a new issue #8545: Incorrect results from R 3.4.2 in MNIST
URL: https://github.com/apache/incubator-mxnet/issues/8545
 
 
   I have installed mxnet 0.12.0 on Ubuntu 16.04.03 LTS. I have tested with the included MNIST example. It works well with Python and returns correct results. However, in R 3.4.2 (both CRAN and MRAN) it produces either NAs or random (i.e. it is not learning) results, e.g. 
   
   `[10] Train-accuracy=0.112346420940171, [10] Validation-accuracy=0.113581730769231. `
   
   Any suggestions?
   
   ## sessionInfo()
   
   ```
   R version 3.4.2 (2017-09-28)
   Platform: x86_64-pc-linux-gnu (64-bit)
   Running under: Ubuntu 16.04.3 LTS
   
   Matrix products: default
   BLAS: /usr/lib/openblas-base/libblas.so.3
   LAPACK: /usr/lib/libopenblasp-r0.2.18.so
   
   locale:
    [1] LC_CTYPE=en_AU.UTF-8       LC_NUMERIC=C               LC_TIME=en_AU.UTF-8       
    [4] LC_COLLATE=en_AU.UTF-8     LC_MONETARY=en_AU.UTF-8    LC_MESSAGES=en_AU.UTF-8   
    [7] LC_PAPER=en_AU.UTF-8       LC_NAME=C                  LC_ADDRESS=C              
   [10] LC_TELEPHONE=C             LC_MEASUREMENT=en_AU.UTF-8 LC_IDENTIFICATION=C       
   
   attached base packages:
   [1] stats     graphics  grDevices utils     datasets  methods   base     
   
   other attached packages:
   [1] mxnet_0.12.0   argparse_1.0.7 proto_1.0.0   
   
   loaded via a namespace (and not attached):
    [1] Rcpp_0.12.13       compiler_3.4.2     RColorBrewer_1.1-2 influenceR_0.1.0   plyr_1.8.4        
    [6] bindr_0.1          viridis_0.4.0      tools_3.4.2        digest_0.6.12      jsonlite_1.5      
   [11] viridisLite_0.2.0  gtable_0.2.0       tibble_1.3.4       rgexf_0.15.3       pkgconfig_2.0.1   
   [16] rlang_0.1.2        igraph_1.1.2       rstudioapi_0.7     yaml_2.1.14        findpython_1.0.2  
   [21] bindrcpp_0.2       gridExtra_2.3      downloader_0.4     DiagrammeR_0.9.2   dplyr_0.7.4       
   [26] stringr_1.2.0      htmlwidgets_0.9    hms_0.3            grid_3.4.2         getopt_1.20.0     
   [31] glue_1.2.0         R6_2.2.2           Rook_1.1-1         XML_3.98-1.9       ggplot2_2.2.1     
   [36] tidyr_0.7.2        purrr_0.2.4        readr_1.1.1        magrittr_1.5       codetools_0.2-15  
   [41] scales_0.5.0       htmltools_0.3.6    assertthat_0.2.0   colorspace_1.3-2   brew_1.0-6        
   [46] stringi_1.1.5      visNetwork_2.0.1   lazyeval_0.2.1     munsell_0.4.3      rjson_0.2.15  
   
   ```
   
   ## Build info (Required if built from source)
   
   It was a plain vanilla installation from the Install web site, for R with GPUs, with:
   
   `make -j $(nproc) USE_OPENCV=1 USE_BLAS=openblas USE_CUDA=1 USE_CUDA_PATH=/usr/local/cuda USE_CUDNN=1`
   
   MXNet commit hash:
   (Paste the output of `git rev-parse HEAD` here.)
   `4f2af2d2e5216ab3a1faadcc117709b6836029dc`
   
   Build config:
   (Paste the content of config.mk, or the build command.)
   
   ## Error Message:
   No errors
   
   ## Minimum reproducible example
   train_mnist.R (from example folder)
   
   ## What have you tried to solve it?
   
   1. Run with R CRAN
   2. Run with R MRAN
   
   Same results. Note that mxnet fails to install on R 3.3.3.

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services