Model trained from blocks of size 08x08

Note

This model is trained using the data of size 08x08. It is used for the predictions of blocks with size 08x08.

Statistics

1	global step	133000
2	batch size	128
3	samples each class	3075
4	number of classes	32
5	training time	16h 45m 48s
6	epoch	173

Epoch calculating

>>> 133000*128/3075/32.0
173.0081300813008

Model Performance on Validating Dataset

Evaluation batch size 100, number of batches 192.

Using validating dataset, the details are documented below:

0.1	Name of dataset	val_08x08.csv
0.2	Size of dataset	4.8 MB
0.3	Samples	600*32
0.4	Usage of dataset	validation
1	top 5 accuracy	0.650
2	top 6 accuracy	0.711
3	top 7 accuracy	0.759
4	top 8 accuracy	0.823
5	top 9 accuracy	0.846
6	top 10 accuracy	0.846
7	top 11 accuracy	0.867
8	top 12 accuracy	0.884
9	top 16 accuracy	0.929
10	top 17 accuracy	0.936
11	top 18 accuracy	0.944
12	top 19 accuracy	0.950
13	top 20 accuracy	0.955
14	top 28 accuracy	0.980

Model Performance on Testing Dataset

Evaluation batch size 100, number of batches 192.

Using testing dataset, the details are documented below:

0.1	Name of dataset	test_08x08.csv
0.2	Size of dataset	4.7 MB
0.3	Samples	600*32
0.4	Usage of dataset	test
1	top 5 accuracy	0.651
2	top 6 accuracy	0.710
3	top 7 accuracy	0.756
4	top 8 accuracy	0.791
5	top 9 accuracy	0.820
6	top 10 accuracy	0.843
7	top 11 accuracy	0.863
8	top 12 accuracy	0.879
9	top 16 accuracy	0.922
10	top 17 accuracy	0.931
11	top 18 accuracy	0.938
12	top 19 accuracy	0.944
13	top 20 accuracy	0.951
14	top 28 accuracy	0.988