This repository contains all the information needed to reproduce the results of the Udacity MLND Capstone Project: Classification of imaging mass spectrometry ion images provided by european project METASPACE.
Download and unpack the input data from AWS S3
wget https://s3-eu-west-1.amazonaws.com/intsco-datasets/udacity_mlnd_capstone_image_classes.tar.gz
tar -xvf udacity_mlnd_capstone_image_classes.tar.gz
All the needed data will be extracted into image_classes directory.
The code can be found in the train_cnn.ipynb Jupyter Notebook.
The notebook above has the following dependencies:
- Tensorflow>=1.4
- Keras>=2.06
- Pandas>=0.20
- Numpy>=1.13
- Matplotlib>=2
The code can be run on a CPU but using GPUs is highly recommeneded due to long network training time. We used a p2.xlarge AWS GPU instance with one NVIDIA Tesla K80. For easy start, one of the public Ubuntu images with most dependencies preinstalled can be used.