NeuroMachine – trading software, designed to automatically search for hidden price patterns of financial instruments. This programme is a self-learning model that simulates the work of the human brain.
NeuroMachine using neural technologies is able to find the interdependence of the price behavior of various trading instruments and build a fully-functional trading algorithm.
NeuroMachine is designed to work in conjunction with the most popular trading terminals, it does not require special knowledge in computers and programme codeing.
In the first window of the project master, you must select the account number of the trading terminal, specify the timeframe of the input data, select trading instruments for downloading the quotes history.
In the next window you need to set the depth of historical data injected and the time difference between the computer and the terminal. When you click "Next" data will be loaded into the program for further analysis.
The next step is to configure the parameters of the neural network. It should be noted trading instruments that will provide input to the analysis; depth of the analyzed data; an instrument that will be predicted by the neural network and the type of forecast. Two variants of the forecast are supported: "NEXTBAR" - the forecast of growth or fall in the price of the predicted bar, "RISE AND FALL" - the trend forecast, during which the price will not increase/decrease by a specified amount as a percentage. In the presented picture, we will build a forecast of the trend of EURUSD, during which the price will not "squeeze" more than 0.03% (~ 0.00034 points), using historical data 10 bars deep.
Next you need to establish the type and complexity of the neural network, the type of neurons. The neural network consists of internal layers (1, 2, 3 ...) containing neurons. For each layer, you must specify the number of neurons. Then you need to determine the type of activation function for neurons "Sigmoid" or "RelU". Sigmoid is more suitable in most cases. Determine the type of neural network: "Classification" or "Regression". The network of the "Classification" type allows you to build a probabilistic forecast of the rise or fall in the price of the predicted instrument, the type "Regression" predicts the value of the rise or fall in the price.
In this window, you can either go directly to the neural network training by clicking the "Train" button, or you can finish the master work and save the data in order to return to the training later.
When you click the "Train" button, the neural network will launch the setup master. The master includes the following steps:
The graph of the predicted instrument (red line) is displayed in the first window of the master. For successful learning of a neural network, it is necessary to specify two data intervals:
The remaining part of the data (white area on the graph) does not participate in the learning process, however, the trading results will still be calculated on this data, which allows evaluating the performance of the strategy in the "future" on unfamiliar data.
The "Auto-select" button helps to automatically select a balanced range of learning in such a way that it would have an equal number of up trends and downtrends.
In the next window, you need to set Becktester parameters for calculating the trading results: trading instrument, broker commission, etc.
Here you can set the parameters of the learning algorithm. Selecting "Deep learning" activates the function of deep learning. It can be used for faster and better learning of the neural network, especially when there is a large amount of input data. Setting the checkbox "Compute trades opening/closing levels" activates the automatic selection of levels for opening/closing deals.
After pressing the (Start learning) button, the neural network begins the process of building a predictive model at the training interval. In the process of learning, the system constantly tests the model at the validation interval in order to avoid over-optimization and "fitting" the results. The best model found is preserved, and the results of the work are shown by the yield graph, which takes up most of the masters’s window. Two other graphs, located to the right, show the graph of profitability of the current model and the training error. You can stop training at any time until the end of the cycle by pressing the (Stop) button.
The button (Cycle repeat) allows you to repeat the last training cycle. This can be useful if you are not satisfied with the automatically selected best model and you want to stop the learning process manually.
Pressing the (Continuous Learning) button leads to an endless repetition of training cycles. At the same time, the system will select the best model among all cycles. You can stop the learning process at any time by pressing the (Stop) button.
Pressing the "Finish" button saves the created neural network and trading strategy. After that, the trading algorithm is ready for analysis or automated trading.
To test the strategy on historical data, simply select the menu item "Script -> Script backtester", as a result of which the strategies backtester window will open.
On the first tab of the backtester, you must specify:
After clicking the "Test" button, the backtester will perform calculations and on the "Report" tab will display general information about the trading strategy, on the "Transactions" tab - information about completed transactions, and on the "Chart" tab - a balance chart.
To launch the obtained strategy for real trading, you must perform the following steps: set the value of the lot of the instrument being traded, establish a connection with the trading terminal and launch the automatic trading.
By default, the lot size for a trading instrument is set at 0.01. To change this value, you need to execute the menu command – "Settings -> Composite instrument settings" and in the opened window edit the lot value of the desired instrument.
To launch NeuroMachine in automated trading, you need to establish a connection with the trading terminal – "Connection -> Connect to Forex terminal" and start the automatic trading – "Script -> Run script". The "Messages" tab of the main screen displays all events during program operation.
Examples of the real work of the program NeuroMachine and trading results can be found in the section Video Reports.