In these experiments, the advantages and disadvantages of eliminating high-frequency as well as many low-frequency components are evaluated. The parameters are estimated by a support-vector machine (RBF kernel) and RBF neural network on SDSS samples. Their performances are assessed by MAE. WT(i, 0) and WT(i, 1) represent the coefficients of a wavelet transform with i-level decomposition in the approximation and high-frequency sub-bands. {Ti}, {Li} and {Fi} denote the extracted features for log Teff, log g and [Fe/H], respectively. The number after the colon represents the total number of features utilized.
| log Teff . | log g . | [Fe/H] . | ||||||
|---|---|---|---|---|---|---|---|---|
| Features . | SVRG . | RBFNN . | Features . | SVR . | RBFNN . | Features . | SVRG . | RBFNN . |
| {Ti}:17 | 0.0062 | 0.0065 | {Li}:24 | 0.2035 | 0.2159 | {Fi}:25 | 0.1512 | 0.1547 |
| WT(4,0):239 | 0.0055 | 0.0062 | WT(4,0):239 | 0.1909 | 0.2267 | WT(4,0):239 | 0.1311 | 0.1486 |
| WT(4,1)+{Ti}:256 | 0.0165 | 0.0083 | WT(4,1)+{Li}:263 | 0.2368 | 0.2449 | WT(4,1)+{Fi}:264 | 0.1862 | 0.1770 |
| Full:3823 | 0.0460 | 0.0131 | Full:3823 | 0.3726 | 0.2366 | Full:3823 | 0.4118 | 0.1769 |
| log Teff . | log g . | [Fe/H] . | ||||||
|---|---|---|---|---|---|---|---|---|
| Features . | SVRG . | RBFNN . | Features . | SVR . | RBFNN . | Features . | SVRG . | RBFNN . |
| {Ti}:17 | 0.0062 | 0.0065 | {Li}:24 | 0.2035 | 0.2159 | {Fi}:25 | 0.1512 | 0.1547 |
| WT(4,0):239 | 0.0055 | 0.0062 | WT(4,0):239 | 0.1909 | 0.2267 | WT(4,0):239 | 0.1311 | 0.1486 |
| WT(4,1)+{Ti}:256 | 0.0165 | 0.0083 | WT(4,1)+{Li}:263 | 0.2368 | 0.2449 | WT(4,1)+{Fi}:264 | 0.1862 | 0.1770 |
| Full:3823 | 0.0460 | 0.0131 | Full:3823 | 0.3726 | 0.2366 | Full:3823 | 0.4118 | 0.1769 |
In these experiments, the advantages and disadvantages of eliminating high-frequency as well as many low-frequency components are evaluated. The parameters are estimated by a support-vector machine (RBF kernel) and RBF neural network on SDSS samples. Their performances are assessed by MAE. WT(i, 0) and WT(i, 1) represent the coefficients of a wavelet transform with i-level decomposition in the approximation and high-frequency sub-bands. {Ti}, {Li} and {Fi} denote the extracted features for log Teff, log g and [Fe/H], respectively. The number after the colon represents the total number of features utilized.
| log Teff . | log g . | [Fe/H] . | ||||||
|---|---|---|---|---|---|---|---|---|
| Features . | SVRG . | RBFNN . | Features . | SVR . | RBFNN . | Features . | SVRG . | RBFNN . |
| {Ti}:17 | 0.0062 | 0.0065 | {Li}:24 | 0.2035 | 0.2159 | {Fi}:25 | 0.1512 | 0.1547 |
| WT(4,0):239 | 0.0055 | 0.0062 | WT(4,0):239 | 0.1909 | 0.2267 | WT(4,0):239 | 0.1311 | 0.1486 |
| WT(4,1)+{Ti}:256 | 0.0165 | 0.0083 | WT(4,1)+{Li}:263 | 0.2368 | 0.2449 | WT(4,1)+{Fi}:264 | 0.1862 | 0.1770 |
| Full:3823 | 0.0460 | 0.0131 | Full:3823 | 0.3726 | 0.2366 | Full:3823 | 0.4118 | 0.1769 |
| log Teff . | log g . | [Fe/H] . | ||||||
|---|---|---|---|---|---|---|---|---|
| Features . | SVRG . | RBFNN . | Features . | SVR . | RBFNN . | Features . | SVRG . | RBFNN . |
| {Ti}:17 | 0.0062 | 0.0065 | {Li}:24 | 0.2035 | 0.2159 | {Fi}:25 | 0.1512 | 0.1547 |
| WT(4,0):239 | 0.0055 | 0.0062 | WT(4,0):239 | 0.1909 | 0.2267 | WT(4,0):239 | 0.1311 | 0.1486 |
| WT(4,1)+{Ti}:256 | 0.0165 | 0.0083 | WT(4,1)+{Li}:263 | 0.2368 | 0.2449 | WT(4,1)+{Fi}:264 | 0.1862 | 0.1770 |
| Full:3823 | 0.0460 | 0.0131 | Full:3823 | 0.3726 | 0.2366 | Full:3823 | 0.4118 | 0.1769 |
This PDF is available to Subscribers Only
View Article Abstract & Purchase OptionsFor full access to this pdf, sign in to an existing account, or purchase an annual subscription.
