search

Zero-We / BCL

A bayesian collaborative learning framework for whole-slide image classification
GNU General Public License v3.0
24 stars 4 forks source link

What is the optimal value of k on TCGA_lung? #4

Open Rainydu184 opened 11 months ago

Rainydu184 commented 11 months ago

After many attempts on TCGA_lung, the best results I have obtained so far are: 0.9001 (round 0), 0.9203 (round 1), and 0.9189 (round 2). It's strange that the metrics decreased in round 2.

As you mentioned in your text, different values of k have an impact on the results. Currently, I am using the default value of k (10). Did you use a better value in your results as mentioned in your text?

Zero-We commented 11 months ago

Hi Rainydu, I set a larger k(50) for TCGA data, considering they have larger tumor regions and stronger heterogeneity. And it is normal that the performance fluctuations slightly during the iteration process.

Rainydu184 commented 11 months ago

Thank you!!!