Skip to content

microns-ariadne/ariadne-synapse-detector-development

Repository files navigation

Instructions for Harvard lab install for classifier development

These are instructions for running the Ariadne pipeline on the R0 dataset to produce evaluation metrics for classifiers.

Installation

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 .

Specifying your classifier

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.

Running

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"))

Troubleshooting

During the pipeline, Luigi is running on your computer at port 8082. Go to http://localhost:8082 to see your progress.

About

Scripts for running the pipeline to score a synapse detector.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 2

  •  
  •