search

djimenezsanchez / NaroNet

Trained only with subject-level labels, NaroNet discovers phenotypes, neighborhoods, and areas with the highest influence when classifying subject types.
GNU Affero General Public License v3.0
15 stars 7 forks source link

Low classification accuracy on Endometrial POLE experiment reproduction #8

Open CarolRameder opened 2 months ago

CarolRameder commented 2 months ago

Hello!

I tried to reproduce the experiment described in Section 3.2. I used the same hyperparameter setup presented in Supplementary Table 3. The PCL component yielded 61.87% contrast accuracy compared to the 81.11% reported in the paper.

PCA_result

The only modification was to remove 5 images with missing labels from the dataset and use 331 instead of 336. The results can be seen in the table below. I repeated the NaroNet training with the same embeddings 5 times to analyze how the results vary.

results

Do you have any idea why this happens? Thanks in advance!

TechnoBear1000 commented 1 week ago

You can try danijimnzs@gmail.com