.torch_multi_head_attention

In order to calculate attentions with a query,
this function takes the dot product of query with the keys
and gets scores/weights for the values.
Each score/weight the relevance between the query and each key.
And you reweight the values with the scores/weights,
and take the summation of the reweighted values.
This implementation is based on the code made available by Vivien Garnot
https://github.com/VSainteuf/lightweight-temporal-attention-pytorch

internal

An end-to-end toolkit for land use and land cover classification
using big Earth observation data, based on machine learning methods
applied to satellite image data cubes, as described in Simoes et al (2021) <doi:10.3390/rs13132428>.
Builds regular data cubes from collections in AWS, Microsoft Planetary Computer,
Brazil Data Cube, and Digital Earth Africa using the STAC protocol <https://stacspec.org/>
and the 'gdalcubes' R package <doi:10.3390/data4030092>.
Supports visualization methods for images and time series and
smoothing filters for dealing with noisy time series.
Includes functions for quality assessment of training samples using self-organized maps
as presented by Santos et al (2021) <doi:10.1016/j.isprsjprs.2021.04.014>.
Provides machine learning methods including support vector machines,
random forests, extreme gradient boosting, multi-layer perceptrons,
temporal convolutional neural networks <doi:10.3390/rs11050523>,
residual networks <arxiv:1809.04356>, and temporal attention encoders
<arXiv:2007.00586>.
Performs efficient classification of big Earth observation data cubes and includes
functions for post-classification smoothing based on Bayesian inference, and
methods for uncertainty assessment. Enables best
practices for estimating area and assessing accuracy of land change as
recommended by Olofsson et al(2014) <doi:10.1016/j.rse.2014.02.015>.
Minimum recommended requirements: 16 GB RAM and 4 CPU dual-core.

Gilberto Camara

sits

Satellite Image Time Series Analysis for Earth Observation Data
Cubes

Rolf Simoes

Felipe Souza

Lorena Santos

Pedro Andrade

Karine Ferreira

Alber Sanchez

Gilberto Queiroz

.torch_multi_head_attention function

<dl><dt>n_heads</dt>
<dd>Number of attention heads.</dd>
<dt>d_k</dt>
<dd>Dimension of key tensor.</dd>
<dt>d_in</dt>
<dd>Dimension of input values.</dd></dl>

Arguments

Charlotte Pelletier, <a href="/link/charlotte.pelletier%40univ-ubs.fr?package=sits&version=1.1.0" data-mini-rdoc="sits::charlotte.pelletier@univ-ubs.fr">charlotte.pelletier@univ-ubs.fr</a>
Gilberto Camara, <a href="/link/gilberto.camara%40inpe.br?package=sits&version=1.1.0" data-mini-rdoc="sits::gilberto.camara@inpe.br">gilberto.camara@inpe.br</a>
Rolf Simoes, <a href="/link/rolf.simoes%40inpe.br?package=sits&version=1.1.0" data-mini-rdoc="sits::rolf.simoes@inpe.br">rolf.simoes@inpe.br</a>
Felipe Souza, <a href="/link/lipecaso%40gmail.com?package=sits&version=1.1.0" data-mini-rdoc="sits::lipecaso@gmail.com">lipecaso@gmail.com</a>

Author

Torch module for calculating multi-head attention — .torch_multi_head_attention

<dl>

<dt>n_heads</dt>
<dd>Number of attention heads.</dd>


<dt>d_k</dt>
<dd>Dimension of key tensor.</dd>


<dt>d_in</dt>
<dd>Dimension of input values.</dd>

</dl>

Charlotte Pelletier, <a href='mailto:charlotte.pelletier@univ-ubs.fr'>charlotte.pelletier@univ-ubs.fr</a>
Gilberto Camara, <a href='mailto:gilberto.camara@inpe.br'>gilberto.camara@inpe.br</a>
Rolf Simoes, <a href='mailto:rolf.simoes@inpe.br'>rolf.simoes@inpe.br</a>
Felipe Souza, <a href='mailto:lipecaso@gmail.com'>lipecaso@gmail.com</a>

.torch_multi_head_attention: Torch module for calculating multi-head attention

Description

Usage

Value

Arguments

Author

References