- Pulls financial data (historical prices of various indices) from the Yahoo Finance API
- Constructs a recurrent neural network with an architecture suitable for time-series forecasting
- Trains the network on that data
- Displays graphs and output data detailing the results
Output graph of %-change in price over time of various indices, with prediction for the S&P500 in red, actual S&P500 in blue:
Output graph of neural network training and validation errors over training epochs:
Example output text:
train-errors: [ 0.043748 0.041089 0.037072 0.036456 0.035725 0.034317 0.033760 0.032625 0.032160 0.032116 0.031686 0.031163 0.030988 0.030825 0.029538 0.029586 0.029203 0.028406 0.028104 0.027824 0.027243 0.030032]
valid-errors: [ 0.117359 0.045864 0.042122 0.041731 0.043724 0.039824 0.039880 0.038656 0.039171 0.041646 0.043562 0.035770 0.039031 0.039117 0.040060 0.043550 0.039386 0.039182 0.043095 0.043065 0.041201 0.044457]
Net Topology: 24-12-1
On Fri, Aug 23, 2013 the market will increase.
53.397% Directional Accuracy