Random Number node #91373

New Issue

Jacques Lucke · 2021-09-13T13:32:04+02:00

Jacques Lucke commented

2021-09-13 13:32:04 +02:00

Since we remove the Attribute Randomize node, we need a viable alternative to generate random numbers based on a seed. This node might replace or extend the Random Float node.
It's unclear whether we want different nodes for different data types or a single node for all data types.

Other related topics:

Should the node itself have a seed? Then the same input seed can be used in different nodes to generate different numbers. Without this, the user often has to manually change seeds.
How does this relate to the White Noise texture node?

Since we remove the `Attribute Randomize` node, we need a viable alternative to generate random numbers based on a seed. This node might replace or extend the Random Float node. It's unclear whether we want different nodes for different data types or a single node for all data types. Other related topics: * Should the node itself have a seed? Then the same input seed can be used in different nodes to generate different numbers. Without this, the user often has to manually change seeds. * How does this relate to the White Noise texture node?

Jacques Lucke commented

2021-09-13 13:32:04 +02:00

Added subscriber: @JacquesLucke

Paul Larson commented

2021-09-14 23:19:36 +02:00

Added subscriber: @GeorgiaPacific

Charlie Jolly commented

2021-09-15 01:33:06 +02:00

Added subscriber: @CharlieJolly

Charlie Jolly commented

2021-09-15 01:33:06 +02:00

Some thoughts:

Rename Random Float to Random Number and base on the legacy Attribute Randomize node
Keep as a single high level node with Data Types from legacy Attribute Randomize node (Vector In > Vector Out etc)
Remove Operation menu from Attribute Randomize, this is no longer needed in fields (use Math node instead)
Add a checkbox for Auto seed which hides seed value (uses a seed based on some derived value)
Add alternative distribution options from D10459
Port White Noise (hash) shader node and keep as a separate and alternative node for unsigned random values (assume this is implemented with a different hash function)

Some thoughts: - Rename Random Float to Random Number and base on the legacy Attribute Randomize node - Keep as a single high level node with Data Types from legacy Attribute Randomize node (Vector In > Vector Out etc) - Remove Operation menu from Attribute Randomize, this is no longer needed in fields (use Math node instead) - Add a checkbox for Auto seed which hides seed value (uses a seed based on some derived value) - Add alternative distribution options from [D10459](https://archive.blender.org/developer/D10459) - Port White Noise (hash) shader node and keep as a separate and alternative node for unsigned random values (assume this is implemented with a different hash function)

Zijun Zhou commented

2021-09-15 01:41:29 +02:00

Added subscriber: @Eary

Zijun Zhou commented

2021-09-15 01:41:29 +02:00

In #91373#1220171, @CharlieJolly wrote:

Add a checkbox for Auto seed which hides seed value (uses a seed based on some derived value)

I like the checkbox but not sure about hiding the seed value. I previously discussed this in the devtalk thread:
https://devtalk.blender.org/t/fields-and-anonymous-attributes-proposal/19450/148?u=eary

I also mentioned a checkbox for making it output a field instead of a constant, but the reason for this thought was "connect the index to the seed is actually not ideal, since you just lose the control over the seed by doing that." For example, you want two different random floats but their max and min are the same range, you would still need to manually change seed, and if you hide the seed value, that's more of a pain in this situation.

I would suggest instead of hiding the seed value after checking the box, simply change the value from the seed itself to a "Seed Offset" after "Auto Seed" is checked, so it would be a field output that would have different seed based on the index, and the user can still control the seed's offset in an intuitive way.

If you think the naming to be confusing, then we can have the logic of seed offset under the hood but have the UI be, like the seed input would jut stay being displayed as seed, the checkbox can be named "Field", or if you are not comfortable exposing the field term in the interface, maybe "Constant" and have it checked by default. Once you uncheck it, the node will auto seed it under the hood and outputs a field. The seed input would just stays the same in the UI while actually being an offset of the actual seed. If you don't like the checkbox to be named "Constant" then we can figure something else but the central idea is the seed offset.

> In #91373#1220171, @CharlieJolly wrote: > - Add a checkbox for Auto seed which hides seed value (uses a seed based on some derived value) I like the checkbox but not sure about hiding the seed value. I previously discussed this in the devtalk thread: https://devtalk.blender.org/t/fields-and-anonymous-attributes-proposal/19450/148?u=eary I also mentioned a checkbox for making it output a field instead of a constant, but the reason for this thought was "connect the index to the seed is actually not ideal, since you just lose the control over the seed by doing that." For example, you want two different random floats but their max and min are the same range, you would still need to manually change seed, and if you hide the seed value, that's more of a pain in this situation. I would suggest instead of hiding the seed value after checking the box, simply change the value from the seed itself to a "Seed Offset" after "Auto Seed" is checked, so it would be a field output that would have different seed based on the index, and the user can still control the seed's offset in an intuitive way. If you think the naming to be confusing, then we can have the logic of seed offset under the hood but have the UI be, like the seed input would jut stay being displayed as seed, the checkbox can be named "Field", or if you are not comfortable exposing the field term in the interface, maybe "Constant" and have it checked by default. Once you uncheck it, the node will auto seed it under the hood and outputs a field. The seed input would just stays the same in the UI while actually being an offset of the actual seed. If you don't like the checkbox to be named "Constant" then we can figure something else but the central idea is the seed offset.

Simon Thommes commented

2021-09-23 15:57:57 +02:00

Added subscriber: @SimonThommes

Simon Thommes commented

2021-09-23 15:57:57 +02:00

Following the outcome of the design discussion:
I didn't have a chance to participate myself and am largely happy with what was decided.
There is one thing I do want to point out though that I would really change about the node and that is the naming of the seed inputs.
The decision was to call them Seed 1 and Seed 2, while seed 1 is prepopulated with the Index field as default.

because of this difference in the default the two inputs (while holding the same potential functionality) are semantically different, in my opinion
there is a good reason why the first seed is using the index seed as a default and it is very useful to communicate this reason with appropriate naming
due to the fact that the first seed is using the Index field by default it will look and behave differently in the UI, as it is marke as a field socket and has no input slider
most users that are not very experienced with procedural systems or programming will have a hard time making the direct connection between a
it is argued that the Stable ID naming of the point distribute node output makes its functionality clear enough to not default to a stable id in the random value node. the same naming logic should apply here, in my opinion

For all of those reasons I propose to call the first seed input ID instead. That marks it clearly as an identifier, which it is. Same ID results in same random value. And it does not narrow it down do be an index or an integer, so it is general enough. But it communicates much more clearly how it is utilized (by default).

Following the outcome of the design discussion: I didn't have a chance to participate myself and am largely happy with what was decided. There is one thing I do want to point out though that I would really change about the node and that is the naming of the seed inputs. The decision was to call them `Seed 1` and `Seed 2`, while seed 1 is prepopulated with the Index field as default. - because of this difference in the default the two inputs (while holding the same potential functionality) are semantically different, in my opinion - there is a good reason why the first seed is using the index seed as a default and it is very useful to communicate this reason with appropriate naming - due to the fact that the first seed is using the Index field by default it will look and behave differently in the UI, as it is marke as a field socket and has no input slider - most users that are not very experienced with procedural systems or programming will have a hard time making the direct connection between a - it is argued that the `Stable ID` naming of the point distribute node output makes its functionality clear enough to not default to a stable id in the random value node. the same naming logic should apply here, in my opinion For all of those reasons I propose to call the first seed input `ID` instead. That marks it clearly as an identifier, which it is. Same ID results in same random value. And it does not narrow it down do be an index or an integer, so it is general enough. But it communicates much more clearly how it is utilized (by default).

Zijun Zhou commented

2021-09-23 16:22:01 +02:00

In #91373#1224666, @SimonThommes wrote:
The decision was to call them Seed 1 and Seed 2, while seed 1 is prepopulated with the Index field as default.
...
For all of those reasons I propose to call the first seed input ID instead. That marks it clearly as an identifier, which it is. Same ID results in same random value. And it does not narrow it down do be an index or an integer, so it is general enough. But it communicates much more clearly how it is utilized (by default).

I like it. So there will be an ID input that has Implicit index field, and a Seed input with a slider. This sounds good.

> In #91373#1224666, @SimonThommes wrote: > The decision was to call them `Seed 1` and `Seed 2`, while seed 1 is prepopulated with the Index field as default. > ... > For all of those reasons I propose to call the first seed input `ID` instead. That marks it clearly as an identifier, which it is. Same ID results in same random value. And it does not narrow it down do be an index or an integer, so it is general enough. But it communicates much more clearly how it is utilized (by default). I like it. So there will be an `ID` input that has Implicit `index` field, and a `Seed` input with a slider. This sounds good.

Jacques Lucke commented

2021-10-15 15:15:20 +02:00

Changed status from 'Needs Triage' to: 'Resolved'

Jacques Lucke closed this issue

2021-10-15 15:15:20 +02:00

Jacques Lucke self-assigned this 2021-10-15 15:15:20 +02:00

Sign in to join this conversation.

No Label

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

Random Number node #91373