✔ Incremental Development
The natural iterative and incremental problem solving approach of forming, testing and validating hypotheses is present in kaos. Machine learning is similar to any problem with unknown dimensionality, which requires continual improvement towards a desired end goal (i.e. a specific metric or desired KPI).
kaos simulates this process by allowing any number of training inputs - code, data and/or params. A simplified conceptual example is presented below.
Inputs | kaos Command |
Code, Data, Params |
|
Resulting model trained with incorrect architecture
Inputs | kaos Command |
Code with correct architecture |
|
Resulting model trained with poorly labelled training dataset
Inputs | kaos Command |
Data with relabelled training data |
|
Resulting model trained with assumed learning rate
Inputs | kaos Commands |
Hyperparameters with range of learning rates |
|
Resulting model satisfies the desired KPI!
.png?generation=1571389057954072&alt=media)
kaos enables incremental processing when any or all updated inputs are desired
Check out the Training Pipeline for detailed information regarding its inputs and outputs.
