These are instructions for running the Ariadne pipeline on the R0 dataset to produce evaluation metrics for classifiers.
Installation is for Linux only, tested on Ubuntu 16.04. You must have Anaconda installed. You must also have read access to the Github repos:
https://github.com/VCG/microns_skeletonization
https://github.com/microns-ariadne/pipeline_engine
These are the steps:
$ git clone git@github.com:microns-ariadne/pipeline_engine
$ cd pipeline_engine
$ conda env create -f conda-install.yaml
$ source activate ariadne_microns_pipeline
$ pip install --process-dependency-links .
$ pip install keras==1.2.1
$ pip install .
$ cd ..
$ git clone https://github.com/rhoana/butterfly
$ cd butterfly
$ pip install .
The script will pickle your classifier, which should consist of a Keras model .json file and an HDF5 weights file. You have to rename "example_classifier.yaml" to "classifier.yaml" (in this directory) and then edit classifier.yaml with the details regarding your classifier. The classifier.yaml file should be self-documenting.
You should have coxfs01 mounted on your system as /n/coxfs01 (if not, you'll have to hand-edit lab-rh-config.yaml to change paths).
Set the following environment variables:
MICRONS_TMP_DIR - a directory for intermediate results and logfiles MICRONS_ROOT_DIR - a directory for the report.json file
Do the following from this directory:
$ source activate ariadne_microns_pipeline
$ ./run_synapse.sh
An example pickle file is /n/coxfs01/leek/classifiers/2017-05-04/2017-05-04_synapse.pkl Pickling a classifier is TODO. For now, start IPython and type
from ariadne_microns_pipeline.classifiers.keras_classifier import KerasClassifier
from ariadne_microns_pipeline.algorithms.normalize import NormalizeMethod
help(KerasClassifier)
k = KerasClassifier(<path-to-model-file>, <path-to-weights-file>...)
import cPickle
cPickle.dump(k, open(<path-to-pickle-file>, "w"))
During the pipeline, Luigi is running on your computer at port 8082. Go to http://localhost:8082 to see your progress.