(433) Eros and (25143) Itokawa surface properties from reflectance spectra

Home

All issues

Volume 675 (July 2023)

A&A, 675 (2023) A50

Full HTML

Open Access

Table B.1

Hyperparameters, search ranges for the hyperparameters, and other properties of the searched models.

	Full res.		Eros res.		Itokawa res.
	Class	Comp	Class	Comp	Class	Comp

Property	Value						Range
Network type	conv.	conv.	conv.	conv.	conv.	conv.	conv., dense
Number of hidden layers	1	2	1	2	1	2	1–3
Nodes in the input 1.	401	401	78	78	64	64	fixed
Nodes/filters in hid. 1.	24	24 and 8	24	24 and 8	24	24 and 8	4–32
Nodes in the output 1.	16	10	16	10	16	10	fixed
Conv. kernel size	5	5 and 5	5	5 and 5	5	5 and 5	3–7
Hid. l. activation	ELU	ReLU	ELU	ReLU	ELU	ReLU	ReLU, tanh, sigmoid, ELU
Out. l. activation	softmax	sigmoid	softmax	sigmoid	softmax	sigmoid	sigmoid, softmax
Dropout rate (in.-hid.)	0.0	0.0	0.1	0.0	0.005	0.0	0.0–0.3
Dropout rate (hid.-hid.)	N/A	0.3	N/A	0.3	N/A	0.3	0.0–0.5
Dropout rate (hid.-out.)	0.3	0.4	0.3	0.4	0.3	0.4	0.0–0.5
L₁ trade-off parameter	0.1	0.005	0.01	0.005	0.01	0.005	0.00001–1.0
L₂ trade-off parameter	0.1	0.00001	0.01	0.00001	0.01	0.00001	0.00001–1.0
Training algorithm	Adam	Adam	Adam	Adam	Adam	Adam	Adam, SGD
Learning rate	0.0032	0.0005	0.001	0.0005	0.0013	0.0005	0.0001–1.0
Batch size	144	8	128	8	56	8	1–128
Batch norm. before activation	False	False	False	False	False	False	True, False
Max. number of epochs	1500	2000	1500	2000	1500	2000	fixed

Notes. Class and Comp stand for classification and composition models, respectively. N/A stands for ‘not applicable’.

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.