Basic Model Builder
Alviss AI has a basic model builder designed to quickly build models for you, it is available at /models/new. It consists of a couple of steps
Select dataset¶
You can select which dataset to use when building the model
Which modeling combinations¶
For each combination of country, region and grouping a dedicated model is buildt. Select the combinations that you want to build a model for.
Training parameters¶
In order for the model to train as well ass possible there are some parameters one can tweak such as
- Training period: data from this period will be used to update the model
- Hold-Out period: Data from this period is used to evaluate the performance of the model, the data can overlap with the training period if one chooses.
- Samples: The number of samples used to quantify the uncertainty during the training, higher number of samples makes it more stable but also takes longer.
- Learning rate: How big steps the parameters of the model is updated with.
- Epochs: How many times the model should be updated with the new data before it is done training.
Set specific effects¶
In the model there are various variable groups(depending on what is in the data set):
- Media
- Season
- Trend
- Pricing
- ....
Here one can turn of some of the variable groups if one is not interested in including them in the model.
One can also state how big impact the variable group is expected to have on the sales, note the more of these one use the slower the model will be to train.
