features
will denote the name of a data column that is utilized during the training. In our dataset, each column holds an integer value ranging from 0 to 112. This value represents the id of a selected hero for that player in that match. For instance, p_2_hero_id
holds the index of the hero Player 2
has selected. Furthermore, Player 0
to Player 4
form the first team, whereas Player 5
to Player 9
form the second.radiant_win
, which holds the value of either 0
or 1
, depending on the outcome of the game. Ultimately, the model is evaluated on its accuracy.split
key is used to configure how the pipeline reads the data from the source and splits it into a training and an evaluation dataset. The first part of the block is quite self-explanatory - it is an 80-20 split. However, the second part needs some explanation.all pick
or all draft
normal
game or a ranked
gametrainer
configuration. I will not go over the details, but in a short explanation, it creates a simple feedforward neural network dealing with a binary classification problem. As explained before, the input goes through an embedding layer first, followed by an average pooling layer. The output is fed into fully connected layers, which ultimately handle the prediction.