Models need to be trained on historical data. You should define a market (or a set of markets) for your model to train at. “Market” is a combination of a symbol, chart time frame (like D) and time range (like from date A to date B). You can use any symbols TrendSpider supports, including your own uploaded data, Breadth symbols or Composite symbols. Use any date frames TrendSpider supports. Use up to 10K candles for each date range. You can use up to 3 markets when training a model.
Under the hood, all the market data you select will be combined into a single learning data set. The last 20% of the data set will be used as a test data set. That is, it won’t participate in model deriving relations while training. Test data sets are only used for the purpose of assessing the model quality.
Approximately ~100 candles in between of “training data set” and “learning data set” will be omitted completely. These candles won’t participate in neither training data set nor test data set. We do it in order to avoid knowledge leaking between data sets because of lagging indicators.
Using multiple data sets
Training your model on multiple sets of data might result in better generalization capabilities of the model. That’s cool. However, it comes at a cost of increased complexity of the overall task of finding a good model. In general, avoid using multiple data sets unless you are sure that it's exactly what you want. The more markets you have, the greater are the odds of your model getting lost in noise, so you’ll need better inputs and more thoughtfully picked model type and model hyperparameters.
If you want multiple data sets, then select markets with similar volatility. Otherwise a fixed set of TP/SL (see below) won’t make a lot of sense.
One good case for using multiple markets is to use the same symbol and time frame, but pick different date ranges representing the same “kind of a market”. For example, you can pick all the ranges which are good examples of bullish markets. Once you train a model like that, you’d have to remember that it’s trained for bullish markets and backtest and apply it accordingly. The ranges still should be diverse enough, in order for a model to have a chance to see cases when it should NOT enter.
Order of data points in the training data set matches the order you've got the markets listed on your screen.
Note on how historical data is used for training
Please remember that market data does nto participate in machine learning directly. It just serves the purpose of being a foundation for computing your input values. Please refer to documentation related to model inputs to learn more.