| 1 |
<h2>DESCRIPTION</h2> |
| 2 |
|
| 3 |
<p><em>i.svm.predict</em> predicts values with a Support Vector Machine (SVM) |
| 4 |
and stores them in a raster file. Predictions are based on a signature file |
| 5 |
generated with <a href="i.svm.train.html">i.svm.train</a>. |
| 6 |
</p> |
| 7 |
|
| 8 |
<p>Internally the module performs input value rescaling of each of imagery |
| 9 |
group rasters by minimum and maximum range determined during training.</p> |
| 10 |
|
| 11 |
<h2>NOTES</h2> |
| 12 |
|
| 13 |
<p><em>i.svm.train</em> internally is using the LIBSVM. For introduction |
| 14 |
into value prediction or estimation with LIBSVM, see |
| 15 |
<a href="https://www.csie.ntu.edu.tw/~cjlin/papers/guide/guide.pdf">a |
| 16 |
Practical Guide to Support Vector Classification</a> by |
| 17 |
Chih-Wei Hsu, Chih-Chung Chang, and Chih-Jen Lin.</p> |
| 18 |
|
| 19 |
<p>It is strongly suggested to have semantic labels set for each raster |
| 20 |
map in the training data (feature value) and in value prediction imagery groups. |
| 21 |
Use <a href="r.support.html">r.support</a> to set semantic labels.</p> |
| 22 |
|
| 23 |
<h2>PERFORMANCE</h2> |
| 24 |
|
| 25 |
<p>Value prediction is done cell by cell and thus memory consumption |
| 26 |
should be constant.</p> |
| 27 |
|
| 28 |
<p>The <em>cache</em> parameter determines the maximum memory allocated |
| 29 |
for kernel caching to enhance computational speed. It's important to |
| 30 |
note that the actual module's memory consumption may vary from this |
| 31 |
setting, as it solely impacts LIBSVM's internal caching. The cache is |
| 32 |
utilized on an as-needed basis, so it's unlikely to reach the specified value.</p> |
| 33 |
|
| 34 |
<h2>EXAMPLE</h2> |
| 35 |
|
| 36 |
<p>This is the second part of classification process. See |
| 37 |
<a href="i.svm.train.html">i.svm.train</a> for the first part.</p> |
| 38 |
|
| 39 |
<p>Predict land use classes form a LANDSAT scene from |
| 40 |
October of 2002 with a SVM trained on a 1996 land |
| 41 |
use map <em>landuse96_28m</em>.</p> |
| 42 |
<div class="code"><pre> |
| 43 |
i.svm.predict group=lsat7_2002 subgroup=res_30m \ |
| 44 |
signaturefile=landuse96_rnd_points output=pred_landuse_2002 |
| 45 |
</pre></div> |
| 46 |
|
| 47 |
<h2>SEE ALSO</h2> |
| 48 |
|
| 49 |
<em> |
| 50 |
Train SVM: <a href="i.svm.train.html">i.svm.train</a><br> |
| 51 |
Set semantic labels: <a href="r.support.html">r.support</a><br> |
| 52 |
Other classification modules: <a href="i.maxlik.html">i.maxlik</a>, |
| 53 |
<a href="i.smap.html">i.smap</a> |
| 54 |
</em><br> |
| 55 |
LIBSVM home page: <a href="https://www.csie.ntu.edu.tw/~cjlin/libsvm/">LIBSVM - A |
| 56 |
Library for Support Vector Machines</a> |
| 57 |
|
| 58 |
<h2>REFERENCES</h2> |
| 59 |
|
| 60 |
<p>Please cite both - LIBSVM and i.svm.</p> |
| 61 |
<ul> |
| 62 |
<li> |
| 63 |
For i.svm.* modules:<br> |
| 64 |
Nartiss, M., & Melniks, R. (2023). Improving pixel-based classification of GRASS |
| 65 |
GIS with support vector machine. Transactions in GIS, 00, 1–16. |
| 66 |
https://doi.org/10.1111/tgis.13102 |
| 67 |
</li> |
| 68 |
<li> |
| 69 |
For LIBSVM:<br> |
| 70 |
Chang, C.-C., & Lin, C.-J. (2011). LIBSVM : a library for support vector machines. |
| 71 |
ACM Transactions on Intelligent Systems and Technology, 2:27:1--27:27. |
| 72 |
</li> |
| 73 |
</ul> |
| 74 |
|
| 75 |
<h2>AUTHOR</h2> |
| 76 |
|
| 77 |
Maris Nartiss, University of Latvia. |