On Sat, 26 Oct 2019, Arne Babenhauserheide wrote:
Hi Jan,
Jan Wedekind <j...@wedesoft.de> writes:
On Wed, 23 Oct 2019, Arne Babenhauserheide wrote:
Is aiscm already at a stage where I could experiment with using it for
handwriting recognition even though I don’t have prior experience with
setting up machine learning pipelines?
I have only tested the Tensorflow C bindings with a CPU. However
there is also a GPU version of the library available [1].
I put an MNIST (handwritten digit recognition) example on the
website [1]. I am not sure how to do general handwriting recognition.
That sounds great! Maybe I can use it to create a first prototype.
Though I’m slightly afraid of having to write 20k examples of steno
words :-)
Best of luck. Maybe you can find training data on the Internet.
The Tensorflow C bindings lack some higher level stuff which is
implemented in Python (e.g. dropout regularization, saving
checkpoints).
Is there any chance of getting a guix.scm file for the dependencies?
Sorry, but I am not familiar with Guix. However you can look at the
Dockerfile on how to set things up.