Error: The h5py Python package is required to use pre-built Keras models

[holgerbrandl]

To reproduce the issue start a docker container

docker pull rocker/tidyverse
docker run -it rocker/tidyverse /bin/bash --login

within docker prep python and install keras package in R

## prep python installation (pip, virtualenv are needed for install_keras)
wget https://bootstrap.pypa.io/get-pip.py
python3 get-pip.py

pip install virtualenv
pip install h5py

Rscript - <<"EOF"
install.packages("keras")
library(keras)
install_keras()

conv_base <- application_vgg16(weights = "imagenet", include_top = FALSE, input_shape = c(150, 150, 3))
summary(conv_base)
EOF

Which finally reports

Installation complete.

Error: The h5py Python package is required to use pre-built Keras models
Execution halted

So I've checked if the h5py package is correctly installed:

# python3
Python 3.5.3 (default, Jan 19 2017, 14:11:04) 
[GCC 6.3.0 20170118] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import h5py
/usr/local/lib/python3.5/dist-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.
  from ._conv import register_converters as _register_converters

Which looks OKish. What am I doing wrong?

The issue seems related to https://stackoverflow.com/questions/46819212/error-in-r-the-h5py-python-package-is-required-to-save-and-load-models but none of the suggested workarounds seem to solve the problem

[obarisk]

It's failed in have_h5py() function.
reticulate requires libpython3.x depending on your python version.

[stenpiren]

i am having the same issue!

[ras44]

I am having the same issue and don't see a resolution in this thread- @skeydan any reason this was closed? Could you point me to the solution if its in this thread? I've tried to implement everything above but still have the same problem. Thank you!

[ras44]

@skeydan this worked for me:

sudo ln -s /usr/lib64/libpython2.7.so.1.0 /usr/lib64/libpython2.7.so

Solution description:

@obarisk mentioned:

reticulate requires libpython3.x depending on your python version

I created an R test script with reticulate to make sure that worked:

# reticulate test script
library(reticulate)
os <- import("os")
os$listdir(".")

If failed, initially with:

# Python shared library '/usr/lib64/libpython2.7.so' not found, Python bindings not loaded.

This indicates the libraries are missing from /usr/lib64/.... So I created a symbolic link as above to make the link to the python 2.7 libs. FYI this was on CentOS 6/7.

[skeydan]

Thanks for posting your solution!
(This was closed as part of a batch close of inactive issues, in order get an overview of what's still active.)

[paulexcoff]

I had this exact issue from trying to run Keras without installing tensorflow. So if someone else has this issue, make sure you've installed tensorflow

`
install.packages("tensorflow")

library(tensorflow)

install_tensorflow()
`

[luostrowski]

I'm having the same problem..

I have tensorflow installed but I get the following error when I try to run the load_model_hdf5 function:

Error in load_model_hdf5(model_file): The h5py Python package is required to save and load models

[ras44]

hi @luostrowski a couple thoughts that might help:

• are you able to import h5py in python (in the virtualenv if you're using one)?
• also, what version of python are you using? if it's 2.7 you may have to follow: Error: The h5py Python package is required to use pre-built Keras models #361 (comment) if its 3.x

[t-kalinowski]

@luostrowski

• Can you please open a new issue? (This one is quite old, much has changed since this was first opened)
• How did you install keras? keras::install_keras() should also install h5py.

[luostrowski]

Hi @ras44 and @t-kalinowski,

Thank you both for the replies!

I figured it out the problem, somehow the code was importing Python in a virtual environment different than the one I had h5py installed. Once I adjusted this, the script worked.