Why Mosey?

Experiments leave me wanting

• to compare models at a glance
• to run save models that are good
• to track and save experimentation parameters
• to run models async asynchronously
• to track data used for specific models

So I built mosey an lightweight experimentation pipeline that runs on the MOSI dataset but is generalizable to any machine learning task.

It's of course experimental!

Key Features

• Name and save experiments
• Track data used in an experiment -Track parameters as will
• Save model and experimentation statistic
• Run mutiple models (in sequence for now) but will not bstop on single failures.
• Simple (ish), clean and fast.

Coming

• Serialization for models parameters, datetime objects; the works
• Tests!
• Handle custom data preperation schemes in Data
• Compare over experiments, currently compare only over runs
• Load Comparision!