mess with iris dataset. use different algorithms to identify the flower types.
[x] pre-developed NN
[x] bayesian network
[x] random forest
150 total samples, 30 for testing
NN achieved about 0.933 accuracy
G-NB achieved about 0.933 accuracy
RF achieved about 0.933 accuracy
final lessons learned:
RF is very flexible, but takes a LOT of effort to program correctly.
with enough tuning, networks can achieve similar performance, but are optimized for different tasks
with high variance comes overfitting to training data, per Elements of Statistical Learning. fix by ending training early or adding more bootstrapped trees to average.
naive bayes classifier is nice and simple to program, with very few hyperparameters, but makes several assumptions (esp independent variables)
input: training and testing data
output:
--- "performance: 0.XXX"
mess with iris dataset. use different algorithms to identify the flower types.
[x] pre-developed NN
[x] bayesian network
[x] random forest
150 total samples, 30 for testing
NN achieved about 0.933 accuracy
G-NB achieved about 0.933 accuracy
RF achieved about 0.933 accuracy
final lessons learned:
input: training and testing data output: --- "performance: 0.XXX"