LOSS-FUNCTION 1 -- Log-likelihood loss

[pfarisel]

Hi, it appears to me that in the Loss you compute
a*x +(1-a)*x = x
Thus there is no differences between censored and non-censured instances.
Anyway, this does not implement the loss indicated in the paper.

[Anivader]

Hi @pfarisel ,
DeepHit loss is defined as L = a*L1 + (1-a)*L2. Not sure how you are replacing the terms with "x"

[pfarisel]

Hi,
what it is not cleat to me in def loss_Log_Likelihood(self) is the fact that tmp1 is equal to tmp2
then you put zero some element in tmp1 end the opposite element in tmp2 the you add them
Obtaining back the original tmp1 (or tmp2).
In my opinion, since tmp1 = tmp2
tmp1 + 1.0*tmp2 == I_1 * log(tmp1) + (1. - I_1) * log(tmp1) = log(tmp1)

What am I missing?

Then the code:
I_1 = tf.sign(self.k)
tmp1 = tf.reduce_sum(tf.reduce_sum(self.fc_mask1 * self.out, reduction_indices=2), reduction_indices=1, keep_dims=True)
tmp1 = I_1 * log(tmp1)
#for censored: log \sum P(T>t|x)
tmp2 = tf.reduce_sum(tf.reduce_sum(self.fc_mask1 * self.out, reduction_indices=2), reduction_indices=1, keep_dims=True) # this equal to tmp1 before I_1
tmp2 = (1. - I_1) * log(tmp2)
self.LOSS_1 = - tf.reduce_mean(tmp1 + 1.0*tmp2) # = I_1 * log(tmp1) + (1. - I_1) * log(tmp1)

[chl8856]

Hi all,
Sorry for checking this late. There was a typo after the version change.

self.fc_mask1 should be changed to self.fc_mask2 as the following:

tmp2 = tf.reduce_sum(tf.reduce_sum(self.fc_mask2 * self.out, reduction_indices=2), reduction_indices=1, keep_dims=True) # this equal to tmp1 before I_1

Sorry for the confusion and late response.