gramaziokohler / aixd_ara Goto Github PK

The model_summary method in controller breaks when the number of targets is larger than the number of features.

Observed in ARA version 0.9.3.

Currently we only have interface to the CAE model.

To add CVAE, check if it needs additional inputs (e.g. weights for the beta-CVAE) ?

If not, then both models (CAE and CVAE) could be handled from the same component with and additional input to choose model type.

Testing ARA on Mac and Rhino 7 reveals following issues:

• server - no terminal window shows up
• open folder (explorer) doesn't work
• plotters in static mode

Will not be investigated or fixed for now:

• RhinoPythonEditor: no module named Grasshopper (sometimes)
• plotters in interactive mode - hangs up (on some machines)
• no Bitmap+ for mac (static mode display doesn't work)

TODO:

• create new release
• adjust documentation

Examples:

• Urban massing, direct sun
• Shell Bridge

For each:

• grasshopper file
• readme
• pre-generated dataset ?
• checkpoint ?

The project resetter (that clears the session controller and the sticky) should be triggered every time a server is started (or stopped).

revise how the request is formulated

• match data type to the dataobject
• enable requesting ranges

generator browsing

• Implement "clicker" to browse through generated designs
• integrate the component that instantiates the design into generator? or add it as a new component (currently missing)
• clear the instance if output of the generator is cleared
• remove genselect component?

request-predict-real comparison --> to be continued in #31

• improve how the inputs for this are formatted and collected
• remove the manual setup to collect real values
• integrate retrieval of PAs #17

Currently some names/nicknames of GH components contain white spaces etc.
Make all follow the same naming convention.

Until #4 is in place, remove the port input from Server GH component because other components can't retrieve the custom port number and use the DEFAULT_PORT only.
Also, check all places where port is used.

Documentation structure:

• Introduction - a few sentences, mention aixd
• Installation guide
• Tutorial: workflow, components
• Examples --> link to example files
• Documentation of gh components (instead of API section)
• License
• Authors

Other tasks:

• build and publish (on https://gramaziokohler.github.io/ara ?) @gonzalocasas
• adjust colors on wepbage
• replace compas logo with ARA logo
• create ARA logo, update icons #21

• create logo for ARA
• pick colors for documentation
• adjust colors in icons

In documentation pages in dark mode:

• replace compas logo with ARA logo
• fix rounded corners of icons

Transfer the sketch into the lib.

Similar to #13 , the component should create the param components that are used to set design params and retrieve performance atrributes:

• the param needs to be of the correct type matching the data object
• has the right prefix
• warn if there are already components with this name on the canvas?

In the GH component Server:

• stop does not stop the server
• show_window does not modify after the server has started (one cannot show or hide window later)
• with show_window on, when the window is closed manually by the user, the runner key remains in the sticky

Add:

• simple one-click installer for lay users
• installation guide for advanced users
• installation guide for developers (needed?)

where?

• in README
• in documentation pages #19

Rename the default prefixes to set/retrieve values in parametric model:
GENERATED_ --> DP_
REAL_ --> PA_

Problem:

kaleido is missing:

On Windows 10, Rhino 7.0, ARA 0.9.3 installed using conda instructions.

Warning:

kaleido 0.2.1 causes issues on Windows (hangs up) as previously reported in https://gitlab.renkulab.io/ai-augmented-design/aixd/-/issues/70 .

A working solution is to pip install kaleido==0.1.0.post1 as suggested in plotly/Kaleido#134

TODO:

add the correct version of kaleido to the requirements

Currently, when setting design parameter values to components, every Param component expires instantly.
This triggers rerun of the downstream part of the parametric model every time.

It might be wiser not to expire them, but trigger recalculation upstream from the output (PA) components.
It needs to be checked what happens to other (non-PA) outputs of the model though.

continuing from #15

request-predict-real comparison

• improve how the inputs for this are formatted and collected
• remove the manual setup to collect real values
• integrate retrieval of PAs #17

We want to clear the sticky dictionary, and clear the components using it:

• on demand
• every time a project setup changes, to avoid misleading contents in different components.

In the sticky, only items referring to components in this GH document should be cleared.

Currently the ModelTrain takes the current state of the model for the new training campaing.

Should it automatically reset the model? Or should we add an option to reset?

To retrieve errors, warning and other messages from axid and background app:

• Incorporate Logger from AIXD.
• homogenize what, when and how messages sent back to GH

Also:

• Find a way to send to and display a progress (e.g. training epochs) in GH

Rename from axid_grasshopper or AIXD to ARA:

• prefix of ghcomponents: e.g. aixd_DataBool --> ara_DataBool @funkchaser
• gh toolbar name, also in ghcomponents (category) AIXD --> ARA @funkchaser
• repo in https://github.com/gramaziokohler/aixd_grasshopper --> aixd_ara? @gonzalocasas
• module name in ../src/aixd_grasshopper --> ../src/aixd_ara? @gonzalocasas
• setup.py @gonzalocasas
• config.py @gonzalocasas
• install.py @gonzalocasas
• all imports @gonzalocasas

In the Launch Server component:

• check if port is already open by another app
• enable user to pass a custom port number
  • all other components need to know this number (e.g. through sticky)

Make GH components turn orange when:

• non-optional inputs are missing
• on URL error (e.g. when trying to load a dataset but the project has not been set up)

• in components
• in examples
• in documentation
• internally in controller, api etc. where meaning variable names
• datablock names