search

FlorentF9 / DeepTemporalClustering

:chart_with_upwards_trend: Keras implementation of the Deep Temporal Clustering (DTC) model
MIT License
219 stars 58 forks source link

About the loss value #25

Closed lushunn closed 2 years ago

lushunn commented 2 years ago

When I usepython DeepTemporalClustering.py --heatmap False --dist_metric cid --dataset CBF --pool_size 8to train the DTC, no matter what dataset used, loss value always suddenly increases on the 8th epoch, looks like weight of Lr and Lc also change with epoch, which really confuses me 746f18f1e6f49fd74b2c6c4de1dfcaf 7676fc6860a4d01398f565615df05a7 Looking forward to your reply,thank you!

sajinpgupta commented 2 years ago

After 8th epoch the heatmap loss weight increases from 0.1 to 0.9. This increases the heatmap loss weightage and overall loss value can be high. Hope this helps

sajinpgupta commented 2 years ago 
    if epoch > self.finetune_heatmap_at_epoch:
        print('On epoch end fine tune heatmap', self.finetune_heatmap_at_epoch)
        K.set_value(self.heatmap_loss_weight, self.final_heatmap_loss_weight)