Comet is a quant algo fuzzer.
Here is a minimal .cm script formulating a ratio between a raw time-series feature and its moving average:
Behavior Comparator(signal: DataFrame, eps: Float, reference: DataFrame) {
weights="behavior_1_compare.pth", train=true, supervised_epochs=100
} -> DataFrame
Flow volume_spike {
volume = data("volume")
adv20 = data("adv20")
Comparator(volume, 0.1, adv20)
}
Transformer pretraining
cargo run -- examples/behavior_1.cm