________________________________________________________________________

Version History

________________________________________________________________________

April 7th 2019 7PM = Release of MAC server .exe and MAC client script

March 26th 2019 7:08PM - Release server v1 script v1

March 26th 2019 7:28PM - Release script v2 - added the web-help functionality, fixed problem with buttons

________________________________________________________________________

Theory

________________________________________________________________________

what's a neural network for newbies by a newbie

you have a set of input values, numbers

and another set of output values, also numbers

you present hundreds of examples of input/output values

to the neural network

the neural net is now an expert

at giving you output values for any given input values

in our example, the input values are the positions of a moving target ( a null node named mark ) in Gabrielle's personal space ( meaning the target is parented to Gabrielle's hip joint ) and the output values are Gabrielle's shoulder and foreArm joints rotation angles

________________________________________________________________________

Instructions

________________________________________________________________________

----------------------------------------------------------------------------------------------------------------------------------------------

select all the nodes that produce your input/output values

----------------------------------------------------------------------------------------------------------------------------------------------

in our example,

- Gab's rShoulder joint

- Gab's rForeArm joint,

- A special node named Vernier*

- The target node named mark

* the vernier creator script may eventually be published on this site

----------------------------------------------------------------------------------------------------------------------------------------------

launch the mcjNeuralNetConsole script

----------------------------------------------------------------------------------------------------------------------------------------------

----------------------------------------------------------------------------------------------------------------------------------------------

Specify which NeuralNet server/service will do the maths

----------------------------------------------------------------------------------------------------------------------------------------------

for this software kit, the computations are done in an external app named mcjTinnServer. Which you should have installed by now since you did not skip past the "installation" section of this web page. Click on the [Select NeuralNet Server] button and select mcjTinnServer in your download folder. This only needs to be done one time, the script will remember the location.

----------------------------------------------------------------------------------------------------------------------------------------------

Build your lists of Inputs / Outputs

----------------------------------------------------------------------------------------------------------------------------------------------

from the list of selected nodes, i selected Gabrielle's rShoulder joint

all of Gabrielle's rShoulder properties appeared in the ListBox titled "channels"

there's 3 checkboxes that let you hide/show morphs, hidden properties or non-transforms properties

so i selected in turn the xRotate, yRotate and zRotate properties of the rShoulder joint

and added them to the output list using the [Add Output] button

then i selected the rForeArm joint and added the yRotate( bend) property to the outputs list

then i selected the target/mark, and added it's x, y and zTranslate properties to the inputs list

note that since the marker is parented to Gabrielle's hip bone so those are relative coordinates

and added them to the input list using the [Add Intput] button

Saving/retrieving your Input/Output selections

using the [Save I/O Selection] button you can save the contents of your input/output lists in a text file

Next time you run the script, you can recall this selection, as long as it's the same scene with the same nodes

----------------------------------------------------------------------------------------------------------------------------------------------

Creating the NeuralNetwork

----------------------------------------------------------------------------------------------------------------------------------------------

in the field named "Num Hidden Layer Nodes" enter the number of hidden layer nodes, example: 100

i suppose a larger number of nodes lets the NeuralNet solve problems that are more complex

Click the [Create NeuralNet for listed IOs] button

the script will start/run the NeuralNet server/service ( invisible on-screen but visible in task-manager )

the initial neural net has the same number of inputs and outputs as your Input/Output lists specify.

----------------------------------------------------------------------------------------------------------------------------------------------

Training the Neural Network -

----------------------------------------------------------------------------------------------------------------------------------------------

1 - We need training data

----------------------------------------------------------------------------------------------------------------------------------------------

The A.I. is untrained, it's an Artificial Idiot. It will give you totally random outputs for whatever inputs you give. You need to show it examples of what intelligent outputs look like.

In my example, i used mcjAutoLimb2015 to animate the arm over a 200 frames range,

The index finger tracks the position of the marker/target, so the rotations of the rShoulder and the rForeArm are intelligent output values for the 200 positions of the target/marker

So i specify that the timerange for the training data is the timelin "playRange" i.e. frames 0 to 199

I click the [Save IO Animation as training data] button and save it as vernierTraining.data

Note: contrary to Gustav Louw's Tinn training files, the first line specifies the number of inputs and outputs

----------------------------------------------------------------------------------------------------------------------------------------------

2 - The data does the training

----------------------------------------------------------------------------------------------------------------------------------------------

in the Training Specs section of the script, you can specify the number of iterations, the number of times the A.I. will use and re-use your 200 ( or more or less ) training data samples. You can also specify a training 'rate' and an annealing rate. But since i'm not familiar with that, you'll have to learn it or ball-park it. the default values for those is 1.0

Click the [Train NeuralNet from file] button

select the training data file you saved a moment ago

it will take a few seconds and the message in the log window will be TrnedTinn, OK

if a problem occurs it bay be because you used a training data file that was not created for the same number of inputs and outputs as what your Input/Output lists specify

----------------------------------------------------------------------------------------------------------------------------------------------

Saving/Retrieving your super intelligent A.I.

----------------------------------------------------------------------------------------------------------------------------------------------

Click the [Save NeuralNet to file(.data)] button

Later on if your Inputs and Output lists are compatible with this saved A.I. you can use the [Load NeuralNet from file] button to retrieve it

----------------------------------------------------------------------------------------------------------------------------------------------

Allow the A.I. to rule the world

----------------------------------------------------------------------------------------------------------------------------------------------

Specify a range of frames to be processed, ex: the whole playrange

Click on the [Solve Animation Range Using NeuralNet] button

the script will transmit one by one the positions of the marker and the vernier reading to the trained neural network and apply the response to Gabrielle's rShoulder and rForeArm rotations

keep an eye on the log window, when all the frames of the animation will have been solved by the A.I. the message "completed task" will appear

________________________________________________________________________

License that came with Gustav Louw's Tinn Library at Github

https://github.com/glouw/tinn/blob/master/LICENSE#L1

________________________________________________________________________

________________________________________________________________________

// License for the script and mcjTinnServer.exe

________________________________________________________________________

// You can use this script freely for personal or commercial use.

// You may not sell, resell, sub-license or rent this script in any way.

// you may credit this script to mCasual/Jacques

// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,

// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES

// OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND

// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT

// HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,

// WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING

// FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR

// OTHER DEALINGS IN THE SOFTWARE.

__________________________________________