V0.3.3 - Simpler error handling

• Slight change to the TrainingDataGenerator interface but I don't think this is enough to warrant a new minor version
• Added MustXxx functions to the model, which gives the developer a choice to either handle errors correctly or panic upon an error. There are situations where panic is very appropriate (for example, if you know a model should fit some data, any error that has been raised is likely severe, and probably not recoverable, such as having the wrong-shaped input data)
• Added a new example for FitGenerator which fits a model to a sin curve then plots the results

V0.3.2 - Small addition to prevent duplicate node names

Now Build will return an error if there re two nodes or two layers with the same name. This should prevent further bugs with me forgetting to name nodes in graphs.

V0.3.1 - Bug fix to do with naming of nodes

There were lots of nodes in the graph Goras was generating that had the same name, which was causing bugs. Fixed now I think.

v0.3.0 - Better Typing

• Models now support different layers having different types. You now specify a type for an Input layer, and that type is propagated through the model.
• New layer, called OneHot: OneHot encodes the tensor of inputs (currently only 1D is supported) to a 2D tensor
  • This currently uses a custom op that probably should get moved to the Gorgonia package

Version 0.2.0

• Multiple outputs are now supported!

• Added L2 regularisation, and now have to ability to add L1 in a future release.

• Loss functions have been revamped to support much more whacky losses.

  • Before, a loss could only have a single target, which had to be the same shape as the output layer.
  • Now, a loss can have any number of any shaped target nodes).

Small change to make PredictBatch more consistent with rest of code

v0.1.1

predictbatch now returns a slice of tensors rather than just one

First somewhat useable version

I have now fixed Softmax (yay!). I am still a little confused why the Gorgonia one did not work though.

v0.0.5

The main changes in this release are:

• Fixed batching bugs and added a batched Predict method
• Added another example which classifies cats and dogs
• Added a summary method for easy debugging of your models

The package is now very much usable for small and simple projects.

v0.0.4

The main feature in this release is Conv2D and MaxPool layers.

v0.0.3

The main changes in this release are improvements to error handling. Many functions now return errors rather than panicking, but they may also have a corresponding Must...() function that will panic if an error is detected. This is to try to bring Goras more in line with the way Gorgonia handles errors.