jobrunner.nw

#!/usr/bin/env nextflow
/* Run IDInteraction analysis jobs (i.e. classifying from ground truth
and arbitrary sources of tracking data) using Nextflow */
/* August 2017 */

/*
vim: syntax=groovy
-*- mode: groovy;-*-
*/


params.outfile="outfile.csv"
params.sources="openface,cppmt"
params.includeframefortracking=false
params.targetted=false
experimentparts=[1,2]
shuffleoptions = ["","no"]
trainingframes=[30,60,120,240,450,900,1800]
participantrange = 1..20 // Want to detect automatically (pass as environment var?)
participantcode = participantrange.collect{"P" + String.valueOf(it).padLeft(2,'0')}
participants=Channel.from(participantcode) 

process runClassifier {
	// Run the classifier for a participant/experimentpart combination
	input:
	val participant from participants
	each experimentpart from experimentparts
	each trainingframe from trainingframes
	each shuffle from shuffleoptions

	output:
	// We need each output filename to be unique. 
	// We do this by getting the directory name we run in in the 
	// shell script
	
	file "result*.csv" into classres

	exec:
	sources = ["cppmt":"/idinteraction/results/cppMT/part${experimentpart}/${participant}_front.csv",
	"pca":"/idinteraction/results/depth/${participant}.pca${experimentpart}",
	"mixture":"/idinteraction/results/depth/${participant}.mixture${experimentpart}",
	"openface":"/idinteraction/results/OpenFace/${participant}_front.openface",
	"mixtureside":"/idinteraction/results/depth_side/${participant}.mixture${experimentpart}",
	"pcaside": "/idinteraction/results/depth_side/${participant}.pca${experimentpart}"]

	// Derive the string we need to pass as --tracker parameters
	sourcelist = params.sources.tokenize(",")
	String sourcestring = ""
	for (String s : sourcelist){
		thissource = sources[s]
		if (!thissource){
			throw new IOException("No frame source " + thissource)
		}
		sourcestring += ("--tracker " + thissource + " ")
	}

	includeframefortracking = " "
	if (params.includeframefortracking){
		includeframefortracking = "--includeframefortracking"
	}

	targetted = " "
	shufflecode = shuffle
	if(params.targetted){
		targetted = "--targetted --windowsize 1000 --advancesize 500 --batchsize 50"
		shufflecode="targetted"	
	}

	groundtruth = "/idinteraction/results/Groundtruth/${participant}_attention.csv"

	startstring = "start" + "$experimentpart"
	endstring = "end" + "$experimentpart"

	// Only run for shuffled when doing targetted training
	when:
	!params.targetted || (params.targetted && shuffle=="")

	shell:
	// ${PWD##*/} gets the last part of the current directory
	"""
	abc-classify.py  --${shuffle}shuffle --externaltrainingframes ${trainingframe} ${sourcestring} --summaryfile result\${PWD##*/}.csv --startframe \$(cat /idinteraction/results/EventFrames/${participant}_attention.${startstring}) --endframe \$(cat /idinteraction/results/EventFrames/${participant}_attention.${endstring})   --extgt /idinteraction/results/Groundtruth/${participant}_attention.csv --useexternalgt  --maxmissing 600  --participantcode a:${experimentpart}::f:${trainingframe}::p:${participant}::s:${shufflecode}
	"""
}

process produceOutput {
	// Combine the results of each individual run

	publishDir "data/runresults", mode:"copy" , overwrite:"false"

	input:
	file classreslist from classres.collect()

	output:
	file '*.csv' 

	script:
	"""
	for resfile in ${classreslist}
	do
		cat \$resfile >> ${params.outfile} 
	done
	"""
}