With manual training, you train a strategy by marking a few example trades on the chart. These example trades make up the data from which the machine learning system generates its trading algorithm. The training set should show by example the way you want the strategy to work. The training set should be just big enough to include common cases. The number of training cases you need depends on the indicators on the chart. The minimum size of the training set is one trade. If you only have a candle detector that flags buy signals then just one training case is needed on one of the buy signals. If you have more indicators, the learning algorithm will be more complex and so will need more training inputs.
Select a Data Range
Open the chart page, and set it to your required symbol. Add any indicators, and other inputs to the chart that your strategy will be using. Once that’s done, decide on the range of data you’ll be training with. If you want to do blind testing, it’s best to decide in advance which parts of the data you’ll be using for training, optimization and for testing.
To switch on training mode, click the hat icon on the toolbar. Then go back to the chart to start marking training trades. To mark a training trade, use the mouse and click the open point followed by the close point on the chart. When both points are accepted, the markers should display on the chart as in the screen below. The markers will automatically show if the trade is a buy or a sell. If you hover over either of the points, it’ll show the open and close times.
You can review the training points for the strategy in the training panel on the right of the screen. To display the training panel, click the hat icon in the top toolbar. The first tab displays the full list of training points. You can review any of these points by clicking on any individual row. This will highlight that particular point on the chart.
To delete a point, click the cross icon at the end of the row. To delete all training points, click the Clear button in the bottom menu.
Hints for Creating a Good Training Set
Mark a few example cases for each of your ideal trade entry points. For example, if your strategy is to buy at an oversold indicator level, mark a few points before and slightly after that signal.
The training set should capture just the typical cases. You don’t need to mark every case on the chart, especially when they are very similar. This will lead to over training. A machine learning system that’s over trained will not generalize well to new cases. This means it will work very well on the data you’ve trained on but will find it difficult to recognize new cases, as being valid trade signals.
Blind testing will help to show if your strategy has been over trained.
Under training is the opposite problem and usually happens when the training set is too small for the amount of variability in the data. A strategy that’s under trained will not be constrained enough will usually produce too many trade signals.
The right balance between under training and over training will depend on the number and types of indicators you are using. Using a single basic indicator, such as a candle detector, will require very little training data. If it produces just buy or sell signals, but not both, then just one point is sufficient. If it produces buy and sell signals, you will need to mark one buy and one sell, if you want your strategy to follow both signals.
Once you have created all of your training points, click on the Train button in the right panel. Any changes to the strategy, such as new indicators, or new training points will not take effect until it has been retrained.
When training is complete, you should see a success message. You can now go back to the strategy page, and run a back test to see how it performs.