|
|
import tensorflow.keras as keras |
|
|
|
|
|
class LossHistory(keras.callbacks.Callback): |
|
|
""" |
|
|
Callbacks to store train, validation loss at the the end of every batch and the end of every epoch. |
|
|
You can also track the counts loss and profile loss seperatley using the callbacks provided. |
|
|
""" |
|
|
|
|
|
def __init__(self,model_output_path_logs_name,to_track): |
|
|
self.model_output_path_logs_name=model_output_path_logs_name |
|
|
self.to_track=to_track |
|
|
self.outf=open(self.model_output_path_logs_name,'w') |
|
|
self.outf.write('Epoch\tBatch\t'+'\t'.join(self.to_track)+'\n') |
|
|
keras.callbacks.Callback.__init__(self) |
|
|
|
|
|
def on_train_begin(self, logs={}): |
|
|
self.losses ={} |
|
|
|
|
|
def on_epoch_begin(self,epoch, logs={}): |
|
|
self.losses[epoch]={} |
|
|
for trackable in self.to_track: |
|
|
self.losses[epoch][trackable]=[] |
|
|
self.cur_epoch=epoch |
|
|
|
|
|
def on_batch_end(self, batch, logs={}): |
|
|
for trackable in self.to_track: |
|
|
self.losses[self.cur_epoch][trackable].append(logs.get(trackable)) |
|
|
|
|
|
def on_epoch_end(self,epoch,logs={}): |
|
|
marker=self.to_track[0] |
|
|
num_batches=len(self.losses[self.cur_epoch][marker]) |
|
|
for i in range(num_batches): |
|
|
self.outf.write(str(epoch)+'\t'+str(i)) |
|
|
for trackable in self.to_track: |
|
|
self.outf.write('\t'+str(self.losses[epoch][trackable][i])) |
|
|
self.outf.write('\n') |
|
|
|
|
|
|
|
|
def on_train_end(self,logs={}): |
|
|
self.outf.close() |
|
|
|
