Problem with multiple step transition matrix

464 views
Skip to first unread message

Math

unread,
Oct 22, 2012, 2:52:18 PM10/22/12
to dinami...@googlegroups.com
Hello,

I'm trying to do the H0 example "pattern of change" with my own data (EGO 64 1.8.9).

First, I have two images (1984 and 2011) and they have some null values and 2 of their 11 categories are not de same. I made the transition matrix for 27 years and it worked for the simple but not for the multiple step matrix. I heard something about the ergodic matrix but I don't really get how to make my images good for i.

So after that I tried to add my simple matrix to the "calculate transition matrix functor", but when I do it, it say "Sum of related transition rates exceed 1", how is that possible?

Thanks

Math

Hermann Rodrigues

unread,
Oct 25, 2012, 2:52:55 PM10/25/12
to Dinamica EGO on behalf of Mathieu
Hi Mathieu,

On Mon, Oct 22, 2012 at 4:52 PM, Dinamica EGO on behalf of Math
<dinamica-ego+noreply-APn2wQcFJiZ...@googlegroups.com>
wrote:

> So after that I tried to add my simple matrix to the "calculate transition
> matrix functor", but when I do it, it say "Sum of related transition rates
> exceed 1", how is that possible?

The sum of all transitions coming from the same classe must not exceed
1, i.e, the amount of cells changing from one given class must not
exceed 100%. Can you post your matrix here? Is it this the matrix
produced by Dinamica itself or have you created it by hand?

Best,

----------------------------------------------------------------------
Hermann Rodrigues
her...@csr.ufmg.br
Skype: hermann.rodrigues
Centro de Sensoriamento Remoto / UFMG
http://www.csr.ufmg.br/dinamica

Mathieu

unread,
Oct 29, 2012, 1:38:29 PM10/29/12
to dinami...@googlegroups.com
Hi,

Thanks for responding my question.

I figured out the two images did not have the same amount of values (some pixels of 2010 were null in 1984). After corrected it on arcgis, the simple transition matrix (made with dinamica) works when I import it as a parameter, but the multiple one doesn't yet works. Anyway, I want to try to generate a predicted map and then change parameters more carefull, that's why I'm actually trying the simple example H0.

I put some transition function parameters matrix randomly, corresponding to my transition matrix. When I lauch the process, it still say "The layer #1 is not defined". Do Dinamica has a maximum number of categories per map and do they need to be the same for my two maps (one category doesn't exist for 1984)?

Thank you

Mathieu


Le jeudi 25 octobre 2012 14:53:27 UTC-4, Hermann Rodrigues a écrit :
Hi Mathieu,

On Mon, Oct 22, 2012 at 4:52 PM, Dinamica EGO on behalf of Math
<dinamica-ego+noreply-APn2wQcFJiZdFLskgV77EPFOftDIhRSAaLlU...@googlegroups.com>
simple_extrait84.csv
error_layer1.jpg

Hermann Rodrigues

unread,
Oct 29, 2012, 4:18:13 PM10/29/12
to Dinamica EGO on behalf of Mathieu
Hi Mathieu,

Each transition must have a corresponding probability layer in the
probability map. And these layers must be named "probability_X_to_Y
where X and Y are classes for each transition X to Y.

The H0 example is extremely basic and it is assuming a single
transition. That is why this example uses a Calculate Map to create
the probability map.

If you are planning to use several transitions, you should use a
Create Cube Map to create a probability map from individual
probability layers or use more specialized tools like the operators
designed to generate probability maps using the Weights of Evidence
method.

Probability maps must respect some properties. For example, for a
given cell, the sum of all probabilities from concurrent transitions,
i.e, transitions from the same class, must be less than or equal to
1.0. If the probability map has a cell type different from IEEE 754 32
Bit Real, all probabilities must be scaled so that 0 becomes 0 and 1.0
becomes the maximum value represented by the given cell type.

Internally, Dinamica EGO uses a signed integer value to represent each
class, so classes range from -2147483648 to 2147483647. If you are
planning to save the transition matrix as a lookup table, the limits
are tighter and a class must be a value between 0 and 999.

And probably, the most important advice: Usually it is hard to
calibrate a model with dozens of different transitions.

Best,

----------------------------------------------------------------------
Hermann Rodrigues
her...@csr.ufmg.br
Skype: hermann.rodrigues
Centro de Sensoriamento Remoto / UFMG
http://www.csr.ufmg.br/dinamica


On Mon, Oct 29, 2012 at 3:38 PM, Dinamica EGO on behalf of Mathieu
<dinamica-ego+noreply-APn2wQcFJiZ...@googlegroups.com>
wrote:
Message has been deleted
Message has been deleted

Mathieu

unread,
Nov 1, 2012, 8:45:18 AM11/1/12
to dinami...@googlegroups.com
Hi Hermann,

Thanks for the precisions. I tried to run the model with "create cube map", but I have some difficulties to create the probability maps (you can see on the attached image).

The transition from the same class, are less than or equal to 1.0, the multiple matrix now works.

Do I need to put a "calculate map" in the "repeat" module and should I put out the "create cube map" so every loop a new map is added to it?

Thank you

Mathieu
WebRep
Évaluation globale
 

Hermann Rodrigues

unread,
Nov 1, 2012, 12:15:50 PM11/1/12
to Dinamica EGO
Hi Matheu,

Unfortunately, I could not find the images attached to your post.

We really need to define an operator to incrementally add new layers to a map, but in the current version there is no such operator.

The only way you can construct a cube map using the current Dinamica release is statically defining the number of layers in the output map. This poses a challenge to define a map using dozens or maybe hundreds of layers.

See the example http://csr.ufmg.br/dinamica/dokuwiki/doku.php?id=raster_cube. It shows how to define a cube map using other four maps.


Best,

----------------------------------------------------------------------
Hermann Rodrigues
her...@csr.ufmg.br
Skype: hermann.rodrigues
Centro de Sensoriamento Remoto / UFMG
http://www.csr.ufmg.br/dinamica


--
You received this message because you are subscribed to the Google Groups "Dinamica EGO" group.
To view this discussion on the web visit https://groups.google.com/d/msg/dinamica-ego/-/6q4_JfzYnLAJ.

To post to this group, send email to dinami...@googlegroups.com.
To unsubscribe from this group, send email to dinamica-ego...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/dinamica-ego?hl=en.

Mathieu

unread,
Nov 1, 2012, 12:36:50 PM11/1/12
to dinami...@googlegroups.com
Hi,

sorry I forgot the attached the image.

When you say layers you mean the probability map for each transition?
error_cubemap.jpg

Hermann Rodrigues

unread,
Nov 1, 2012, 4:20:59 PM11/1/12
to Dinamica EGO
Hi,

Yes. When I say layer I mean the probability "layer" for each transition.

Each transition needs a corresponding probability layer in the probability map. And each probability layer has to be named  "probability_X_to_Y". So, if you have a transition from class 2 to class 1, you need a layer in the probability map named "probability_2_to_1". Then, all layers corresponding to all the transitions must be stacked to form a single probability map.

If you are using *only one* transition, your probability layer does not have to follow the "probability_X_to_Y" name convention.

See "Dinamica EGO\Dataset\Examples\advanced\sojourn_time\simulate_deforestation_and_abandonment_using_sojourn_time" in the folder of examples for an example how to construct the probability map using two different layers. In this model, the functor named "Updated Probabilities" is a Create Cube Map.


Best,

----------------------------------------------------------------------
Hermann Rodrigues
her...@csr.ufmg.br
Skype: hermann.rodrigues
Centro de Sensoriamento Remoto / UFMG
http://www.csr.ufmg.br/dinamica


On Thu, Nov 1, 2012 at 2:36 PM, Dinamica EGO on behalf of Mathieu <dinamica-ego+noreply-APn2wQcFJiZ...@googlegroups.com> wrote:
Hi,

sorry I forgot the attached the image.

When you say layers you mean the probability map for each transition?


Mathieu
WebRep
Évaluation globale

Reply all
Reply to author
Forward
0 new messages