Model Performance over Time
Basic comparison of time series models with Weave and PanelPlot
Created on June 1|Last edited on June 7
Comment
Compare performance by model type & year
The charts below show the mean absolute error on the validation dataset for several training scenarios: training on data up to one of three years (through 2009, 2010, and 2012) or validating on data up to one of two years (through 2014 or 2015). These charts compare three different model types: the simple baseline in blue (repeat the previous time step), a linear model in red, and a dense model in orange. The dense model (which is the most complex) performs best, and the error increases across models as more data is added in this toy scenario.
Flexible evaluation on subsets
Evaluate model performance on different time slices by specifying the relevant Artifact alias (any tag formatted as a string: e.g. the full date, month, year, precise timestamp)
Predictions across models for June & July (M6 and M7)
WIP: Weave to diff two time series
With a Weave expression, you can compute the delta over time between any two lines. We're working on enabling this for more than two lines.
Add a comment