Modeling Managed

Fay Kallel
Augmented ML Workflow, Experiment Management, Modeling Best Practices, Training & Tuning

Machine learning powers many of the services we use today from social-media feeds, to voice assistants, to media streaming and much more. Despite widespread use, modeling is uncertain, constrained and multi-variable in nature. With 51% of enterprises building their own models in house, standard operating procedures are quickly becoming a top priority for modeling reproducibility, compliance and explainability. 

SigOpt pioneered an intelligent approach to hyperparameter optimization that is widely adopted across industries and within the most sophisticated modeling environments, the likes of Two Sigma in algorithmic trading, US federal government and leading streaming services providers to name a few. Building on our leadership in model optimization, we’re excited to share with you a new solution for managing the modeling pipeline and enabling the emerging practice of machine learning operations. I hope you join our beta community to reap the benefits that many of our users are raving about. 

Over the last 18 months, we’ve worked closely with modelers across industries to solve their modeling pain points. What we learned centered around three key areas, best illustrated in these interview quotes: 

    • Managing the modeling journey 

“We need a training management dashboard to understand training errors, validation errors, showing feature engineering importance in the context of long training cycles. Also, with too many models, having a history of modeling is essential and needs to be curated the right way.” 
—Device manufacturer – Voice Assistance Modeling

    • Reaching the best model

“We need a solution that makes model training intuitive, what model to use for a particular use case? how to debug your models? Is every step doing what it is supposed to do? What additional experimentation do I need to reach the best model?” 
—Global e-tailer – marketplace modeling system

    • Enterprise system of record to manage all model types 

“We hired experts from Google, Amazon, Facebook, Microsoft and they have diverse skills across tools, we can’t afford to standardize on one stack so we need to make them productive through an enterprise platform that manages all of their models regardless of the modeling environment.” 
—Global Travel & Hospitality service provider – Booking recommender system

SigOpt’s Experiment Management solution helps you track, filter and reproduce your entire modeling workflow, with interactive visuals you can augment your intuition, understand your progress and explain your results. With just a few lines of code you can manage your modeling pipeline from feature selection to best model through collaborative and seamless training and tuning.

We believe our solution unlocks your modeling pain points, you can now:

    • Track and organize your entire model development workflow: SigOpt tracks and organizes your modeling iterations, including: architectures, metrics, parameters, hyperparameters, code snapshots and the results of feature analysis, training runs or tuning jobs. You’ll efficiently explore your problem space by filtering and comparing results and learning iterations.
    • Visualize, understand and explain your progress: Modeling is about tradeoffs, but it can be hard to see the options. With Interactive visuals you can compare training curves, metrics, and models quickly. This in turn helps you understand model performance, inform your intuition and explain results to your colleagues.
    • Seamlessly train and tune any model type: We know that transitioning between training and tuning can be expensive in time and resources. With SigOpt you don’t have to leave tuning until the last minute. Our solution fully integrates automated hyperparameter tuning with training run tracking to give you a sense of the bigger picture and the path to reach your best model. With features like automated early stopping, highly customizable search spaces, multimetric optimization and multitask optimization you can advance your model with easy hyperparameter tuning before taking it into production.

We built our solution in partnership with over 100 modelers who are tackling a variety of modeling problems from many industries. Here’s some of their feedback:

“Thank you so much for your work on this, the COOL factor is high, the fact that you are hovering over a metric point and showing me how that point is linked to all metrics and hyperparameters, that dashboard is giving me in one view great intuition across two types of metrics, I commend you for coming up with an interface like this.”
—National Insurance & Asset Management Provider – Financial modeling

“Saved views after filtering is awesome! It solved my current work-around where I write a script to pull from hundred of directories”
—Government Intelligence Research modeling

Today we are proud to share our solution with you all.  We want to give you superpowers to retrace and reproduce every modeling decision you’ve ever made with perfect recall.

We would love to get your feedback! Join our beta and see what more you can learn about your models. 

Also stay in the know and make sure to check out our deep dive blogs on how to:

    • Track and organize your entire model development workflow
    • Visualize, understand and explain your progress
    • Seamlessly Train and Tune any model type
Fay-Kallel1
Fay Kallel Head of Product

Want more content from SigOpt? Sign up now.