I'm looking to run a simple neural network on a Linux RHEL8. I copied the code from the cifar-10 example, but have run into a few snags.
Initially, the system was running the network fine but crashing after 1-10 epochs. I was able to install an NVIDIA GPU core on the computer, and that seems to have resolved the crashes. I'm still facing a few problems, however:
- It seems that I can't get the system to use more than 5% of my system's GPU core, no more than 61MiB of memory. With a still slow network (when it works), it would be nice to maximize this resource.
- The computer I'm running the code on does not have internet access. Where the code works on an identical computer with network access,
#code to load packages to run the model
library(reticulate)
library(keras)
install_keras(method = c("conda", conda = "auto", version = "default", tensorflow = "gpu")
comes up with an error:
EnvironmentLocationNotFound: Not a conda environment: /home/megan/anaconda3/envs/r-reticulate
Error: installation of 'python=3.6' into environment 'r-reticulate' failed [error code 1]
Evidence points to this problem being a lack of internet connection. Is there a way to locally install keras so that I can use it for a network?
- When #2 shows up as an error, I also receive the following warnings:
2020-02-13 I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
2020-02-13 I tensorflow/core/platform/profile_utils/cpu_utils.cc:94] CPU Frequency: 3312000000 Hz
2020-02-13 I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x55efd1b83680 executing computations on platform Host. Devices:
2020-02-13 : I tensorflow/compiler/xla/service/service.cc:175] StreamExecutor device (0): Host, Default Version
Would love advice on how to optimize this system. I have been looking for all sorts of solutions but am new to both machine learning and linux (not as new to R).