Make loss bias use module level variable

pull/19/head
Matthew Scholefield 2018-07-05 13:30:51 -05:00
parent 3fc1fe4789
commit 252abbeb15
1 changed files with 4 additions and 4 deletions

View File

@ -13,6 +13,8 @@
# limitations under the License.
from typing import *
LOSS_BIAS = 0.9 # [0..1] where 1 is inf bias
def weighted_log_loss(yt, yp) -> Any:
"""
@ -21,23 +23,21 @@ def weighted_log_loss(yt, yp) -> Any:
yp: Prediction
"""
from keras import backend as K
weight = 0.7 # [0..1] where 1 is inf bias
pos_loss = -(0 + yt) * K.log(0 + yp + K.epsilon())
neg_loss = -(1 - yt) * K.log(1 - yp + K.epsilon())
return weight * K.mean(neg_loss) + (1. - weight) * K.mean(pos_loss)
return LOSS_BIAS * K.mean(neg_loss) + (1. - LOSS_BIAS) * K.mean(pos_loss)
def weighted_mse_loss(yt, yp) -> Any:
from keras import backend as K
weight = 0.9 # [0..1] where 1 is inf bias
total = K.sum(K.ones_like(yt))
neg_loss = total * K.sum(K.square(yp * (1 - yt))) / K.sum(1 - yt)
pos_loss = total * K.sum(K.square(1. - (yp * yt))) / K.sum(yt)
return weight * neg_loss + (1. - weight) * pos_loss
return LOSS_BIAS * neg_loss + (1. - LOSS_BIAS) * pos_loss
def false_pos(yt, yp) -> Any: