infill_loss.InfillLossCollection Class Reference

Inheritance diagram for infill_loss.InfillLossCollection:

Public Member Functions
def	__init__ (self, weight=None, size_average=False, ignore_index=-100)
def	forward (self, predict, true, input)

Public Attributes
	ignore_index
	reduce
	size_average

Detailed Description

Definition at line 111 of file infill_loss.py.

Constructor & Destructor Documentation

def infill_loss.InfillLossCollection.__init__	(		self,
			weight = None,
			size_average = False,
			ignore_index = -100
	)

Definition at line 112 of file infill_loss.py.

    def __init__(self, weight=None, size_average=False, ignore_index=-100):
        super(InfillLossCollection, self).__init__(weight, size_average)
        self.ignore_index = ignore_index
        self.reduce = False
        self.size_average = size_average

Member Function Documentation

def infill_loss.InfillLossCollection.forward	(		self,
			predict,
			true,
			input
	)

Definition at line 118 of file infill_loss.py.

    def forward(self, predict, true, input):
        _assert_no_grad(true)

        # Want three losses: non-dead, dead w/o charge, dead w/ charge
        L1loss=torch.nn.L1Loss(self.size_average)
        nondeadweight = 1.0
        deadnochargeweight = 500.0
        deadlowchargeweight = 100.0
        deadhighchargeweight = 100.0
        deadhighestchargeweight = 100.0

        # Identify dead channels
        goodch = (input != 0)
        for idx, col in enumerate(goodch[0, 0, :, :].T):
            if col.sum():
                goodch[0, 0, :, idx] = 1
        deadch = (~goodch).float()
        goodch = goodch.float()

        # Compute non-dead loss
        predictgood = goodch * predict
        adcgood = goodch * true
        totnondead = goodch.sum().float()
        if (totnondead == 0):
            totnondead = 1.0
        nondeadloss = (L1loss(predictgood, adcgood)*nondeadweight)/totnondead

        # Compute dead with highest true charge loss
        deadchhighestcharge = deadch * (true.abs() > 70).float()
        predictdeadhighestcharge = predict*deadchhighestcharge
        adcdeadhighestcharge = true*deadchhighestcharge
        totdeadhighestcharge = deadchhighestcharge.sum().float()
        if (totdeadhighestcharge == 0):
            totdeadhighestcharge = 1.0
        deadhighestchargeloss = (L1loss(predictdeadhighestcharge,adcdeadhighestcharge)*deadhighestchargeweight)/totdeadhighestcharge

        # Compute dead with high true charge loss
        deadchhighcharge = deadch * (true.abs() > 40).float()*(true.abs() < 70).float()
        predictdeadhighcharge = predict*deadchhighcharge
        adcdeadhighcharge = true*deadchhighcharge
        totdeadhighcharge = deadchhighcharge.sum().float()
        if (totdeadhighcharge == 0):
            totdeadhighcharge = 1.0
        deadhighchargeloss = (L1loss(predictdeadhighcharge,adcdeadhighcharge)*deadhighchargeweight)/totdeadhighcharge

        # Compute dead with low true charge loss
        deadchlowcharge = deadch * (true.abs() > 10).float() *(true.abs() < 40).float()
        deadchnocharge = deadch * (true.abs() < 10).float()
        predictdeadlowcharge = predict*deadchlowcharge
        adcdeadlowcharge = true*deadchlowcharge
        totdeadlowcharge = deadchlowcharge.sum().float()
        if (totdeadlowcharge == 0):
            totdeadlowcharge = 1.0
        deadlowchargeloss = (L1loss(predictdeadlowcharge,adcdeadlowcharge)*deadlowchargeweight)/totdeadlowcharge

        # Compute dead with no (< 10) true charge loss
        predictdeadnocharge = predict*deadchnocharge
        adcdeadnocharge = true*deadchnocharge
        totdeadnocharge = deadchnocharge.sum().float()
        if (totdeadnocharge == 0):
            totdeadnocharge = 1.0
        deadnochargeloss = (L1loss(predictdeadnocharge,adcdeadnocharge)*deadnochargeweight)/totdeadnocharge

        totloss = nondeadloss + deadnochargeloss + deadlowchargeloss +deadhighchargeloss+deadhighestchargeloss
        return nondeadloss, deadnochargeloss, deadlowchargeloss, deadhighchargeloss,deadhighestchargeloss, totloss

Member Data Documentation

infill_loss.InfillLossCollection.ignore_index

Definition at line 114 of file infill_loss.py.

infill_loss.InfillLossCollection.reduce

Definition at line 115 of file infill_loss.py.

infill_loss.InfillLossCollection.size_average

Definition at line 116 of file infill_loss.py.

---

The documentation for this class was generated from the following file:

• dunereco/dunereco/InfillChannels/scripts/infill_loss.py

---

Generated by   1.8.11