Overview

Add a model by clicking on the top-left button. Once your model is saved, it will be displayed on your model dashboard.

Step 1 : create the model

Name

Each model has a name. Model names must be unique. By default you are provided a random model name.

Algorithm

Each model uses an algorithm. Algortihms currently available :

• Classification (SVM)

LSTM will be released soon. You must configure your model according to the algorithm you choose.

Data source

The data your model is trained on. Sources currently available:

• Alpaca
• Binance
• Interactive Brokers
• FTX

You must be aware of the difference between how your model is trained and the exchange it is running on, as you can, for instance, train you model with Alpaca's data and deploy it on IB, or train on Binance and deploy it on FTX etc.

Granularity

Size of the candlesticks of the dataset.

Period

The time series period (historical data) your model will be trained on.

Asset

You model must be trained on data from one asset at a time. It lists only shortable stocks for Alpaca. Stocks are printed as <Company Legal Name> <TICKER> and FX & cryptos as <BASE/QUOTE>

Features

There is currently 2 types of features :

• OHLCV : Data related to Open, High, Low, Close and Volume metrics.
• Technical indicators : All the indicators from the TA-Lib library.

You can also choose return lags. For instance, 2 return lags means that for each dataset's row, there will be the two last asset's return. 1 return lag is 1 feature. Free plans have 20 features max, Premium have unlimited features.

Return discrimination

For medium, the model classifies as "BUY" returns higher than 66% of the distribution, and "SELL" those lower than 33%.
For low, respectively 56% and 44%.
For high, respectively 85% and 14%.

Training set

Split point (%) between the training set and the test set. If the training set is x%, the test set will be (100-x)% automatically.

Kernel

Kernel hyperparameter of your model, RBF by default. More info about kernels : https://scikit-learn.org/stable/auto_examples/svm/plot_svm_kernels.html

C

C hyperparameter of your model, you can add and and delete how many values you want. The best C value will be automatically chosen. "The C parameter trades off correct classification of training examples against maximization of the decision function’s margin." More info about C parameter : https://scikit-learn.org/stable/auto_examples/svm/plot_rbf_parameters.html

Gamma

Gamma hyperparameter of your model, you can add and and delete how many values you want. The best Gamma value will be automatically chosen. "Intuitively, the gamma parameter defines how far the influence of a single training example reaches, with low values meaning 'far' and high values meaning 'close'." More info about Gamma parameter : https://scikit-learn.org/stable/auto_examples/svm/plot_rbf_parameters.html

Step 2 : review the model

Once your model is saved, you can see its name, its status ("trained", "not trained", "training" or "error") and you can review all its details in the parameters section. Make sure all these details are correct. If not, you can delete the model by clicking on the bin icon and typing your model's name. If details are correct, you can go to the next step.

Step 3 : training and backtesting

Start the training

To start the process, simply click on the loading icon and click on the "start backtesting" button. After that, your model's status must become "training" and you will be invited to go to the chat section to be notified when your model is ready.

Interpret the results

Each model's report is made of 4 elements:

• A comparison graph of historical returns between the model's strategy and the market.
• An overview of the main model's details and the hyperparameters selected.
• A classification report.
• A confusion matrix.

A new yellow "download" icon also appears. In this tab, you can download the model and the scaler for external use, results checking etc.

The graph

The graph is composed of two lines: one yellow and one black. The yellow one represents the cumulative historical returns of the model's strategy. The black one represents the cumulative historical returns of the asset.

The classification report

To make it simple, it shows the accuracy of your model. More info about classification report : https://scikit-learn.org/stable/modules/model_evaluation.html#classification-report

The confusion matrix

An image that shows the difference between the prediction and the actual values. The Y axis is the prediction and X is the actual values. More info about the confusion matrix : https://scikit-learn.org/stable/modules/generated/sklearn.metrics.confusion_matrix.html