//- UNCLASSIFIED
//- FATS dashboard

extends site

append site.help
	:markdown
		(U) Additional FATS info

append site.parms
	- dock = "left"
	- view = (query.option=="min") ? "Nada" : "FATS"
	- parms = {min:"min", less:"less", full:"full", brief:"brief"}
	- dims = "1800,1000"

append site.body

	block fats.parms
		- zone = "FATSometer"
		- full = query.option == "full"
		- bighelp = false

	case query.option

		when "min"

			#hold.Zones(path="/lookups.db?Ref=setpoint&enabled=1",cols="Name,Path",nomenu)

			#border.FATS(
				title="#{zone} || #{client}",crush,
				head="Zones,Prefs(Locales,Themes,Options,Edit),Files(View,Uploads,$uploads,Stores,$stores),LikeUs,Help",
				LikeUs="/likeus.db",
				View="/files.view",
				Edit="/edit.view?name=#{source}")

				:markdown
					Set your **QoS-Cost comfort zone** here.  Visit [detector developers](fats.view) if you want
					to fine-tune your detectors and your policies.  Please
					[likeus](likeus.view) to improve your QoS profile.  Other useful views:
					[administrators](dmin.view), [security](files.view), [machine learner developers](engines.view),
					[project scientist](project.view), [proof editors](files.view).  

					Dont forget to check-out the [geoNode home](home.view) for other exciting **geoNode** products.

				if bighelp
					#folder.More
						#fit.H
							:markdown
								p $$ J_\alpha(x) = \sum\limits_{m=0}^\infty \frac{(-1)^m}{m! \, \Gamma(m + \alpha + 1)}{\left({\frac{x}{2}}\right)}^{2 m + \alpha} $$

						#cesium.C(path="tips.db",dims="400,400")

						#treefan.T(path="/files.db&pivots=L0,L1,L2,L3,L4,L5",dims="400,400")

						#post.P(path="/uploads.test1040.pdf")

				#grid.Search(
					path="/CATALOG.db",center,crush,
					head="Search,Execute,Print,Refresh",
					page=100,dims="#{dims}",
					search="Content",
					cols="Ref,Name,Link.h,Dated.d,Searched")

		when "brief"
			#brief_fats.Briefing(dims="#{dims}")

		default
			#hold.Execute(path="/states.db?Class=Detector",cols="Name,Name")
			#hold.Action(path="/states.db?Class=Action",cols="Name,Name")

			#folder.FATS

				if full
					#grid.News(
						path="news.db",
						sorts="Stay,Starts",page=30,dims="#{dims}",calc,head="Search,Print,Refresh",
						cols="Message.h")

						:markdown
							The rules governing the periodic roll-off and update of these news items can 
							be adjusted [here](home.view?option=newsedit).

					#fit.Conops
					
						:markdown
							When FATS happens over an area with known ground truth (e.g. when a BE is encounted in
							the NTM channel), FATS updates the ROC (false-alarm, hit-rate) score of all detectors
							in the same focus area.  By retaining detector ROC curves, FATS can intelligently boosting
							the hypothesis from all detectors, thus creating a stronger detector from a pool of
							weaker detectors.
							
							Define your Areas of Interest (by country, BE, sensor, etc) in the [AOI Filter](fats.view?goto=AOI Filter),
							then let the tipper deliver image chips over this AOI to [your detectors](fats.view?goto=Detectors).
							[Detection events](fats.view?goto=Detections) can be consulted at any time.  The automatically populated
							[training and detection jobs](fats.view?goto=Jobs) can be managed by your job queue strategies/priorites.
							Try [liking us](likeus.view) to improve your priority in the job queues.
							
							Before you can attach a detector to an AOI, the detector will need to be trained.  If your detector (positive
							feature over negative background) does not yet exists, clone or revise an existing detector, tune its
							parameters, attach it to an AOI, then Execute (train) it to put it in the job queue.
							
							How are your detectors doing today?  Inpsect their [ROC performance](fats.view?goto=ROC).
							
							Find an under-performing detector?  Begone with thee!  Simply go to [your detector](fats.view?goto=Detectors)
							and disconnect it from its AOI(s).  Or perhaps your feeling more kindly and would like to give
							your detector a second chance?  Then, leave it connected, adjust more aggressive training parameters,
							the Execute (train) it.  Most machine provide some form of "transfer leaning" and will thus continue
							training your detector from its current state.
							
							Want to add a better machine learner?  Simply copy-and-paste your Python, Matlab-like, or 
							JS code (or bind your C code) into one of FATS's [language agnostic engines](engines.view).
							Soon you will be able to use these engines in [workflows](nodeflow.view) as well.
							
							Consult the [FAQs]() for additional Heilmier related thoughts.

					#fit.Status.and.POCs
						:markdown
							Please note that FATS is non-operational and thus enjoys no funding of any kind.  NGA/IIG thus
							reserves the exclusive right to terminate FATS at and time without notice.
							
							Project status can be found [here](http://jira.dev.ic.gov/jira/browse/GRD-1527) and 
							[here](http://jira.dev.ic.gov/jira/browse/GRD-1700) for the NTM and commerical systems, 
							respectively.
							
							For additional information, please contact:  
							*	[Brian James](mailto://brian.d.james@nga.ic.gov) 811-1285  or [low-side](mailto://brian.d.james@nga.mil) 571-721-3589  
							*	[Duncan McCharthy](mailto://duncan.l.mccharthy@nga.ic.gov) 578-6240  
							*	[David Castor](mailto://david.m.castor@nga.ic.gov) TBD  
							*	[Scott VanDusen](mailto://scott.w.vandusen@nga.ic.gov) TBD  
							*	[Will King](mailto://william.t.king@nga.ic.gov) 811-1384  

							NGA/IIG is developing a FATS prototype under a JIRA projects GRD-1527,1700 with intent to deliver
							a machine learning service assessable with the [GIAT R&D infrastructure](http://intellipedia.intelink.ic.gov/wiki/GIAT):
							GEOHUB, C2S, ILE, and AWS are just four [ISPs](requirements.view?goto=ISP) under
							evaluation.  There are presently **NO** active plans to transition this technology
							to an operational environment.

					#fit.Disclaimers
						:markdown
							Please note that FATS is non-operational and thus enjoys no funding of any kind.  NGA/IIG thus
							reserves the exclusive right to terminate FATS at and time without notice.
					
							NGA does not, under any circumstance, warrant, underwrite, endorse, promote or sell
							any technologies used herein.  Use this service at your own risk.  Changes to 
							[FATS's build-to requirements](requirements.view) must be coordinated through
							NGA IIG's [agile project management system](http://jira.dev.ic.gov/jira).  See
							[FATS's TTA/EULA](requirements.view?goto=TTA) for additional technology 
							transfer and end-user agreement information.  FATS welcomes your [issues, bugs, and 
							ideas](issues.view).
							
					#fit.Architecture
						:markdown
							FATS currently endores the OpenCV (low-power), Theano (high-power), and Caffe (high-power)
							machine learning frameworks spanning low-high power compute clouds (w/o and w GPU support).

							FATS's classified (unclassified) systems ingests IDOPs tipped by HYDRA (Enhanced View),
							then orthorectifed and chipped by OMAR (Enhanced View), then circulates jpgs to all 
							detectors sharing that area of interest.  Detectors -- those automatically built
							by the machine learners -- return to FATS a JSON-formatted detection event list.
							
							FATS integrates mature chipping (e.g. OMAR, JPIP streaming, Digital Globe Enhanced View), 
							query (e.g. Hydra, DGEV) into mature [hyperthreaded workflows](exsite.view?options=workflows)
							running language agnostic (Python, Matlab-like, JS, or C) engines.

							While FATS has currently consumed the OpenCV machine learning framework -- low-power
							learners built by the Russians for future Chinese millitary sensors -- FATS has reserved
							the Theano LeNet and Caffe engines for its high-power (GPU CUDA assisted) 
							learners.  FATS also has interest in the Torch7 and OverFear learners.
							
							FATS thus seeks to support a wide range of neural network machine learners including
							decision, SVM and softmax capped convolutional, fully convolutional, attractor, and 
							self organizing neural networks to provide detection and symantic segmentation
							of NTM-commerical images.
							
							Data mining and workflow management are key concepts utilized by FATS.  Architected 
							around web-sockets, a JS-everywhere philosophy, and client-defined skins, FATS 
							suppports the OpenIT and Content Management goals of many organizations.  Consult
							the [Briefing](\fats.view?goto=Briefing) for detailed architecture drawings.

					#fit.FAQs

					#fit.Configure
						#form.Workflow.Parameters(
							path="/config.db",
							dims="#{dims}",crush,
							head="Insert,Update,Delete,Select,Execute,Refresh,Restart",Restart="/start.sys",
							cols="Hawks(SetPoint,MaxFaults.i,Hawks.c),Regulators(QoS0.n,QoS1.n,QoS2.n,QoS3.n,QoS4.n),Chip Cache(MaxAge.n),Services(WMS,WFS)")

						#grid.Forecasting.Models(
							path="/models.db",
							dims="#{dims}",crush,
							head="Insert,Update,Delete,Select,Execute,Refresh,Restart",Restart="/start.sys",
							cols="enabled.c,name,srcs.x,rots.x,sizes.x,flips.x,levels.x")

			#folder.Detectors
			
				:markdown
					NGA does not, under any circumstance, warrant, underwrite, endorse, promote or sell
					any technologies used herein.  Use this service at your own risk.  Changes to 
					[FATS's build-to requirements](requirements.view) must be coordinated through
					NGA IIG's [agile project management system](http://jira.dev.ic.gov/jira).  See
					[FATS's TTA/EULA)(requirements.view?goto=TTA) for additional technology 
					transfer and end-user agreement information.  FATS welcomes your [issues, bugs, and 
					ideas](issues.view).
					
				#grid.GMS(
					path="/detectors.db", 
					dims="#{dims}",page="#{page}",
					cols="Enabled.c,Name,Execute.p,MaxDirty.n,Quality(Evaled.d,Dirty.n,Stars.n,Client),Proofs(Labels.h,Overhead.c),Sampling(Samples.i,ImageBg.i,ImageDev.i,Projections(xRotate.n,yRotate.n,zRotate.i)),Training(ROC(Data(MaxPos.i,MaxNeg.i),HAAR(Width.i,Height.i),MaxStages.i,MinTPR.n,MaxFPR.n),Tweak(MaxDepth.i,MaxWeak.i,TrimRate.n)),Detection(Channel,Pixels.i,Pack.i,Feature.n,Hits.i,SizeRange.n,SizeStep.n)")

					:markdown
						This is the Godel Mathematical Society (HAAR neural network) learning machine.

				if full
					#grid.CNN(
						path="/detectors.db", 
						dims="#{dims}",page="#{page}",
						cols="Name")

						:markdown
							Reserved for Convolutional Neural Network learning machine.

					#grid.FCN(
						path="/detectors.db", 
						dims="#{dims}",page="#{page}",
						cols="Name)")

						:markdown
							Reserved for Fully Connected neural Network learning machine.

					#grid.ANN(
						path="/detectors.db", 
						dims="#{dims}",page="#{page}",
						cols="Name)")

						:markdown
							Reserved for Attractor Neural Network learning machine.

					#grid.GET(
						path="/detectors.db", 
						dims="#{dims}",page="#{page}",
						cols="Name)")

						:markdown
							Reserved for adhoc Good Edge to Track learning machine.

					#grid.SOM(
						path="/detectors.db", 
						dims="#{dims}",page="#{page}",
						cols="Name)")

						:markdown
							Reserved for Self Organizing Map learning machine.

					#grid.HMM(
						path="/detectors.db", 
						dims="#{dims}",page="#{page}",
						cols="Name)")

						:markdown
							Reserved for Hidden Markov Method learning machine.

					#grid.ACE(
						path="/detectors.db", 
						dims="#{dims}",page="#{page}",
						cols="Name)")

						:markdown
							Reserved for an Adaptive Cosine Estimator learning machine.

			if full
				#folder.Jobs
			
					:markdown
						The job queue is automatically populated by [training, detection and user supervision 
						jobs](fats.view?goto=Jobs).  Feel free to Hawk over this queue to insure that
						your jobs complete in the strategy/priority you design.  Try [liking FATS](likeus.view) 
						to improve your job priority in these queues.
					
					#grid.Queue(
						path="/queues.db",
						dims="#{dims}",page="#{page}",refresh=60,
						cols="Flagged.c,Classif,Client,Class,Job,QoS.i,State.i,Work.n,Util(Util0.n,Util1.n,Util2.n,Util3.n),Arrived.d,Departed.d,RunTime.n,Notes.h")

					#grid.Hawks(
						path="hawks.db",
						links="Table.queues",
						dims="#{dims}",page=30,
						cols="Rule(Name,Enabled.c,Condition.h,Action.p,Period.n,Status(Status.h,Changed.i,Matched.i,Pulse.i,Faults.i)")

						:markdown
							Use these action-condition-table rules to hawk over your job queue with actions:

								stop=halt=kill to kill matched jobs and update its queuing history
								remove=destroy=delete to kill matched jobs and obliterate its queuing history
								log=notify=flag=tip to email owner a status of matched jobs
								improve=promote to increase priority of matched jobs
								reduce=demote to reduce priority of matached jobs
								start=run to create jobs from dirty records
								set expression to revise queuing history of matched jobs

					#accordion.Predictive.Look
						#grid.Sim.Events(
							path="/jobs.db",
							crush,
							cols="label,job,tDepart.n,tArrive.n,tWait.n,tDelay.n,tService.n,tStep.n,depth.i")
							
						#plot.QoS.Stats(
							path="jobstats.db&index=delay,pr&init=0,0&ref=[0,0],[100,1]&marker=dot",refresh="1",
							dims="#{dims}")

					#accordion.Detects(head="none")
				
					#pivot.Events(
						path="rocs.db",
						dims="#{dims}",page="#{page}",
						pivots="Name",head="Search,Print,Execute",
						cols="detector,channel")

						:markdown
							Define your Areas of Interest (by country, BE, sensor, etc) in the [AOI Filter](fats.view?goto=AOI Filter),
							then let the tipper deliver image chips over this AOI to [your detectors](fats.view?goto=Detectors).
							[Detection events](fats.view?goto=Detections) can be consulted at any time.
				
					#folder.Visuals(dock="top")
						#post.Tip.Sheet(path="tipsheet.view?src=tips.db&pivots=channel,chip,links,label,count,Stars",dims="#{dims}")

						#map.Tip.Map(
							path="/maps.db&fill=Type&index=Name&details=Info",
							dims="#{dims}",refresh="#{hold}",links="Browser")

						#force.Pivot1(
							path="/tips.db&pivots=channel,chip,label,count,Stars",
							dims="#{dims}",refresh="1",links="Events")
							
						#cluster.Pivot2(
							path="/tips.db&pivots=channel,chip,label,count,Stars",
							dims="#{dims}",refresh="1",links="Events")
							
						#treefan.Pivot3(
							path="/tips.db&pivots=channel,chip,label,count,Stars",
							dims="#{dims}",refresh="1",links="Events")

						#treemap.Pivot4(
							path="/tips.db&pivots=channel,chip,label,count,Stars",
							dims="#{dims}",refresh="1",links="Events")

						#plot.ROC(
							path="/rocs.db&index=FPR,TPR,model&_ref=[0,0],[1,1]&_min=0,0&max=1,1&marker=dot&legend=basic,hard,challenge&trace=1&_extra=[0,0]",
							dims="#{dims}",refresh="1",links="Events")

			#folder.AOIs
			
				:markdown
					Define your Areas of Interest (by country, BE, sensor, etc) in the [Tipper](fats.view?goto=AOIs),
					then let the tipper deliver image chips over this AOI to [your detectors](fats.view?goto=Detectors).
					[Detection events](fats.view?goto=Detections) can be consulted at any time.  AOIs can
					be connected to detectors (trained or not) at any time.  If you need multiple AOIs attached to a detector, 
					simply clone (select-insert) the detector and attach to the AOI.
					
				#post.Tipper(path="https://hydra.ilabs.ic.gov",dims="[750,500]")

				#grid.Override(
					path="tests.db",
					dims="#{dims}",page="#{page}",
					cols="Name,Channel,FilePath")

			if full
				#nodeflow.Flows(path="&view=Kiss",dims="#{dims}",refresh="1")


//- UNCLASSIFIED