Julia BRT Pipeline and Priority Map #99

gottacatchenall · 2023-10-01T22:20:51Z

This pipeline (./pipelines/julia_sdm.json) runs on my machine, few notes before trying to run:

This requires rebuilding the Docker image to update to Julia v1.9.3 with docker compose build
Currently this doesn't mask out open-water, which should probably be fixed because the SDM assigns positive probability to some open-water regions, though it fits (or overfits?) much closer to the occurrence points this happens now

jmlord · 2023-10-02T15:10:43Z

About masking open water, could the BinaryLayer script be useful?

Closes #91

Made them a bit larger by default. Closes #97

through pipeline editor right pane

This reverts commit d3b17a2.

This reverts commit 851a9b3.

jmlord

Ok on my side with these minor changes.
We'd need an SDM person to review now...

runners/julia-dockerfile

scripts/SDM/julia_sdms/generateBackground.yml

scripts/SDM/julia_sdms/fitBRT.yml

scripts/SDM/julia_sdms/generateBackground.yml

scripts/SDM/julia_sdms/loadCHELSA.yml

pipelines/julia_sdm.json

glaroc · 2023-10-02T20:20:18Z

I think it would be important to have spatial resolution as a pipeline input for all SDMs.

…julia_brt

gottacatchenall · 2023-10-02T21:58:03Z

I think it would be important to have spatial resolution as a pipeline input for all SDMs.

I agree, and this also further highlights why we need a pipeline that is only about validating inputs (i.e. hypothetically the user thinks the spatial res is defined in meters when its in degrees so it goes around the earth several times)

+ Small comments on pipeline IO, but it needs way more!

frousseu · 2023-11-16T19:05:42Z

The BRT pipeline can now be run on my Windows computer. For now, the two main things to improve would be:
1 - The ability to choose the species and the predictors
2 - There are duplicated inputs in the Input form (mask, weight matrix) and it is not clear why the weight matrix is there. Shouldn't it be in the pipeline for the priority map?
Otherwise, it would be nice to connect the julia brt with the createBackground function to get a biased background.

frousseu · 2023-11-16T19:16:02Z

Otherwise, I'm getting this error from the block3.json in the priority map pipeline:

$ids
[1] "esacci-lc-2020"

Error: HTTP status '400'. Invalid parameters provided 1 validation error for SearchPostRequest
bbox
  Bounding box must be within (-180, -90, 180, 90) (type=value_error)
No traceback available 
Parent job is Cancelling: killing server process...
Parent job is Cancelling: done.

gottacatchenall · 2023-11-16T20:33:55Z

Otherwise, I'm getting this error from the block3.json in the priority map pipeline:

$ids
[1] "esacci-lc-2020"

Error: HTTP status '400'. Invalid parameters provided 1 validation error for SearchPostRequest
bbox
  Bounding box must be within (-180, -90, 180, 90) (type=value_error)
No traceback available 
Parent job is Cancelling: killing server process...
Parent job is Cancelling: done.

This is because the bounding box has to in long/lat coordinates for Julia right now as everything is in WGS84 (between -180 and 180 long and -85,85 lat)

jmlord · 2024-02-01T14:32:50Z

The BRT pipeline can now be run on my Windows computer. For now, the two main things to improve would be: 1 - The ability to choose the species and the predictors 2 - There are duplicated inputs in the Input form (mask, weight matrix) and it is not clear why the weight matrix is there. Shouldn't it be in the pipeline for the priority map? Otherwise, it would be nice to connect the julia brt with the createBackground function to get a biased background.

I think Guillaume's recent modifications should have fixed these.

jmlord

@gottacatchenall A few clarifications in documentation please before we merge.

Also, most important: the priority map was empty when I ran it.

jmlord · 2024-03-11T13:55:06Z

scripts/SDM/julia_sdms/pcaLayers.yml

@@ -0,0 +1,22 @@
+script: pcaLayers.jl
+name: PCA Layers
+description: "PCAs the layers"


Lacking documentation

jmlord · 2024-03-11T14:01:22Z

pipelines/SDM/SDM_BRT_julia.json

+      "label": "Bounding box",
+      "description": "Bounding coordinates of the extent in the order xmin, ymin, xmax, ymax, in degrees",
+      "type": "float[]",
+      "example": ["-78", "43", "-68", "65"]


The default bounding box is a bit weird.

jmlord · 2024-03-11T14:28:08Z

pipelines/Block3/block3.json

+    },
+    "pipeline@6": {
+      "label": "collections_items (for climate uniqueness)",
+      "description": "Vector of strings, collection name followed by '|' followed by item id",


Please explain a bit how we are supposed to choose these.

jmlord · 2024-03-11T14:29:03Z

pipelines/Block3/block3.json

+    },
+    "pipeline@43": {
+      "label": "k",
+      "description": "the value of k for the k-means algorithm",


Please explains what this means, how to choose it, what impacts it has.

jmlord · 2024-03-11T14:31:37Z

pipelines/Block3/block3.json

+      "example": 5
+    },
+    "Block3>weightLayers.yml@44|uncertainty": {
+      "description": "uncertainty in the inference of some value across space",


More doc here and in the script. I know this is meant to be generic, but maybe giving an example : an indicator we want to optimize for?
Also, what effect in the priority map?

jmlord · 2024-03-11T15:07:13Z

scripts/Block3/climateVelocity.yml

Should be in the folder along with the other climate layers. There should be an explanation differentiating it with Juan's work.

jmlord · 2024-03-11T15:07:49Z

scripts/SDM/julia_sdms/generateBackground.yml

Any reason not to use the R version?

jmlord · 2024-03-11T15:09:04Z

scripts/SDM/julia_sdms/fitBRT.yml

+    type: image/tiff;application=geotiff
+  fit_stats:
+    label: fit_stats
+    description: JSON of model fit stats and threshold


Guidance for interpretation of the fit statistics would be appreciated.

jmlord · 2024-03-11T15:10:58Z

scripts/SDM/julia_sdms/fitBRT.yml

+    type: image/tiff;application=geotiff
+  sdm_uncertainty:
+    label: sdm_uncertainty
+    description: map of relative uncertainty 


Guidance for interpretation: 1 uncertain and 0 certain?

jmlord · 2024-03-11T15:11:44Z

scripts/Block3/weightLayers.jl

When I ran this, the output priority map was empty.

Other comments were added to GitHub PR.

jmlord · 2024-04-03T20:40:02Z

@gottacatchenall Can you investigate why this produces no result?
See also documentation clarification requests in review above.

gottacatchenall · 2024-05-07T14:41:05Z

Once PoisotLab/SpeciesDistributionToolkit.jl#227 is merged, I will refactor this to be merged.

tpoisot · 2024-05-07T14:43:55Z

Once

yup

jmlord · 2024-05-28T19:58:35Z

Hi folks,
Any update on this?
Laura's lab intends to integrate a new bias correction algorithm and could use a BRT script in their pipeline.

tpoisot · 2024-05-28T20:18:45Z

I'm close but no ETA yet -- if all goes well, by the end of June? I'd need to see with @gottacatchenall how much work is needed to port his existing pipeline

glaroc · 2024-05-29T15:35:23Z

@gottacatchenall Do you have a description somewhere of the procedure you use to obtain the uncertainty layer for the BRT ? Thanks.

…s into mdc/julia_brt

gottacatchenall · 2024-06-15T17:03:23Z

I've added a description of the BRT methodology (including how uncertainty is computed) here .

A few things I noticed on revisiting this:

Crossvalidation splits are included in the BRT script, which isn't ideal because we would like different CV methods to be different scripts that can be used an inputs to different SDMs
This probably necessitates a subpipeline for repeated CV on a given model using a single CV method
Computation of fit metrics is inevitably going to be duplicated so should be moved somewhere else

gottacatchenall · 2024-09-19T20:36:24Z

replaced with #170

tpoisot · 2024-09-19T20:37:17Z

gottacatchenall added 6 commits October 1, 2023 11:35

no

6996ec0

🎉 Julia BRT pipeline

0c1c83f

Pipeline metadata and clipping SDM below threshold

121b9de

🐛 in fit stats dir

6e986cd

🐛 in clipping

15b7c6b

yaml data

fe5db3f

jmlord added 8 commits October 2, 2023 11:18

revert change that should not be merged to main

0171f24

Same IO Id condition for outputs than inputs.

15e5a42

Closes #91

removed "undefined" css class

bbb26da

Keeping the width set by the user.

c574a19

Made them a bit larger by default. Closes #97

Can now set pipelines input examples

08c6ece

through pipeline editor right pane

Revert "Fix alignment in chrome."

6ef47e6

This reverts commit d3b17a2.

Revert "Pointer over pen icon"

2814dda

This reverts commit 851a9b3.

Support user inputs as outputs.

68ede53

jmlord reviewed Oct 2, 2023

View reviewed changes

jmlord added 2 commits October 2, 2023 11:47

review of Inputs

918e86b

moved to SDM folder

c2f3a57

gottacatchenall requested a review from glaroc October 2, 2023 15:51

gottacatchenall and others added 8 commits October 2, 2023 14:35

🎉 Julia BRT pipeline

5c45d07

Pipeline metadata and clipping SDM below threshold

f725d8e

🐛 in fit stats dir

883c7e1

🐛 in clipping

cdbaaf0

yaml data

6c1e18e

revert change that should not be merged to main

b904ead

review of Inputs

4f41483

moved to SDM folder

d930d12

Merge branch 'mdc/julia_brt' of github.com:GEO-BON/biab-2.0 into mdc/…

085e1df

…julia_brt

jmlord added 3 commits November 9, 2023 14:09

Corrected accessibility

94fb30c

+ Small comments on pipeline IO, but it needs way more!

remove undefined output

6c2b732

removed a few output blocks, layout

671bc70

glaroc added 2 commits January 31, 2024 14:43

Remove compose.yml

5ddab46

Modified with proper inputs and viewer compat and merge from main

1b97730

glaroc and others added 3 commits March 7, 2024 17:13

Clean prioritization pipeline

aa73460

Tweak outputs

eefd9e7

Merge remote-tracking branch 'origin/main' into mdc/julia_brt

db938eb

jmlord requested changes Mar 11, 2024

View reviewed changes

Documentation review & layout

367fbbf

Other comments were added to GitHub PR.

gottacatchenall and others added 3 commits June 15, 2024 12:56

description 📝

edd9f68

Merge branch 'mdc/julia_brt' of github.com:GEO-BON/boninabox-pipeline…

56f0595

…s into mdc/julia_brt

Update README.md

ca6377c

Update README.md

a416ef9

jmlord mentioned this pull request Sep 11, 2024

Boosted Regression Tree SDM pipeline #168

Closed

Merge remote-tracking branch 'origin/main' into mdc/julia_brt

181a801

gottacatchenall closed this Sep 19, 2024

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Julia BRT Pipeline and Priority Map #99

Julia BRT Pipeline and Priority Map #99

gottacatchenall commented Oct 1, 2023 •

edited

Loading

jmlord commented Oct 2, 2023

jmlord left a comment

glaroc commented Oct 2, 2023

gottacatchenall commented Oct 2, 2023 •

edited

Loading

frousseu commented Nov 16, 2023

frousseu commented Nov 16, 2023

gottacatchenall commented Nov 16, 2023

jmlord commented Feb 1, 2024

jmlord left a comment

jmlord Mar 11, 2024

jmlord Mar 11, 2024

jmlord Mar 11, 2024

jmlord Mar 11, 2024

jmlord Mar 11, 2024

jmlord Mar 11, 2024

jmlord Mar 11, 2024

jmlord Mar 11, 2024

jmlord Mar 11, 2024

jmlord Mar 11, 2024

jmlord commented Apr 3, 2024

gottacatchenall commented May 7, 2024

tpoisot commented May 7, 2024

jmlord commented May 28, 2024

tpoisot commented May 28, 2024

glaroc commented May 29, 2024

gottacatchenall commented Jun 15, 2024

gottacatchenall commented Sep 19, 2024

tpoisot commented Sep 19, 2024

Julia BRT Pipeline and Priority Map #99

Julia BRT Pipeline and Priority Map #99

Conversation

gottacatchenall commented Oct 1, 2023 • edited Loading

jmlord commented Oct 2, 2023

jmlord left a comment

Choose a reason for hiding this comment

glaroc commented Oct 2, 2023

gottacatchenall commented Oct 2, 2023 • edited Loading

frousseu commented Nov 16, 2023

frousseu commented Nov 16, 2023

gottacatchenall commented Nov 16, 2023

jmlord commented Feb 1, 2024

jmlord left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jmlord commented Apr 3, 2024

gottacatchenall commented May 7, 2024

tpoisot commented May 7, 2024

jmlord commented May 28, 2024

tpoisot commented May 28, 2024

glaroc commented May 29, 2024

gottacatchenall commented Jun 15, 2024

gottacatchenall commented Sep 19, 2024

tpoisot commented Sep 19, 2024

gottacatchenall commented Oct 1, 2023 •

edited

Loading

gottacatchenall commented Oct 2, 2023 •

edited

Loading