-
20-Feb-2015
-
- Maybe make UML class diagram or something
-
17-Feb-2015
-
- Work on the Training module - perhaps have a single Teacher class with a couple of lists of functions
-
-
- Update the weights - gradient, momentum, weight reduction / normalization, etc.
-
-
-
- Stopping criteria - minimized error, change in error, etc.
-
-
- Modify LinearRegression model to match NeuralNetwork and LogisticRegression
-
- Merge LinearRegression and LogisticRegression into one model?
-
-
- LinearRegression == LogisticRegression with square error function and linear activation function
-
-
-
- RegressionModel, SimpleLinearRegression, SimpleLogisticRegression, etc.?
-
-
- Logger ideas:
-
-
- Real time graphics update - plot cost vs. iteration
-
-
-
- Test / validation subset classification - plot predictions for a set of test data
-
-
-
- Log to a file
-
-
-
- Composite logger - has multiple loggers and delegates to each
-
-
-
- Plot weights / bias functions (visualize receptive field)
-
-
16-Feb-2015
-
- Make Logger class which we can subclass from--perhaps have the model, training and test data as class attributes, so that nothing needs to be passed when the log methods are called. Or would a Listener pattern be better?
-
- Have a function (functions?) to create confusion matrices
-
- Look into statistical tests like ROC, AUC, F-score, etc. These may be good to put into a logger class, or have as external functions called by individual logger classes.
-
- matplotlib outupt?