Updating scripts and tests

Different_Tests/Info_Trainings.txt

@@ -0,0 +1,19 @@
+- Test_1:
+	- Normal (vanilla) transformer trained on GALACTIC data. At this point we did not have the confusion matrix plots. At this point, we had a Linear embedding for the data before entering into the transformer.
+- Test_1_Convs:
+	- This was a try to train a simple convolutional model on the data, but we did not have the confusion matrices plots yet. We can retrain it whenever is necessary.
+- Test_2:
+	- Same test as 'Test_1' but we added the confusion matrices.
+- Test_3:
+	- Same test as 'Test_2' but we changed the Linear embedding to a Conv embedding. The performance was a little better, but there was not a difference such as the one we observe in the reference paper.
+- Test_3_with_errors:
+	- Same test as 'Test_3' but we consider 2 inputs of 6 channels each one independently for the brightness and for the uncertainty respectively.
+- Test_ZEROES:
+	- Sample test where we interpolated using zeroes instead of the Gaussian process.
+- Test_4_multiply_errors:
+	- Test with a transformer using as input the brightness multiplied by the inverse of the square root of the uncertainty obtained from the Gaussian interpolation process, i.e., ( 1/sqrt(error) ) * brightness. Therefore, there will be only one input of 6 channels.
+- Test_4:
+	- Same test as 'Test_3' but considering the same weight for all labels.
+
+All these tests have been done for the GALACTIC data. Furthermore, we have considered, regarding the loss function, a weighting that goes in an inverse manner with the frequency of each class thereby giving more 'importance' to those classes with less representation within the dataset.
+