dzitkowskik / stockpredictionrnn Goto Github PK

Create a method of comparing performance of two neural networks. Use t-student metric to compare, and its pvalue to say if difference is significant. Use mean of error rates to say which one performed better. As input use a vector of error rates from cross validation ( #2 ) of two neural networks on the same data.

Try to use as many plots as possible from here: https://github.com/dzitkowskik/Introduction-To-Machine-Learning-And-Data-Mining/blob/master/Projects/project_2/plotting.py

Create features vector in OrderBook process method with features:

1. Time - time since last event (ask or bid)
2. Price - price of ask or bid
3. Mid Price - mean of highest bid and lowest ask
4. Volume
5. Side - ask of bid
6. AskPriceDiff - difference in price from last ask
7. BidPriceDiff - difference in price from last bid
8. AvgBidPrice - average bid price in orderbook
9. AvgAskPrice - average ask price in orderbook

Hi,

The link http://pages.mini.pw.edu.pl/~stokowiecw/CIBA/slides/lab2.pdf is 404
Any chance you have another copy? I'm curious.

Create a method of performing a feature selection with method Greedy forward selection using cross validation from #2 it should get testing data and neural network in parameters and return a list of numbers of features to use.

Write some words about how things were implemented and show early results of our working program after implementing issues #1 #2 #3 #4 and #5

Create a method of using cross validation with N passes that can be used with different neural network using different features. It should return a vector of error rates or a vector of vectors of predictions of test data. Data should be split in proportion defined in parameter.