TIMi is Fast
Efficiency is Money!
Using cumbersome, code intensive tools means less time to focus on your data and, ultimately,
less accurate results, less meaningful insights that leads to, finally, less value generated.
Inside TIMi, everything is designed to make your more efficient: The interfaces are intuitive and
requires no code. The tools are several orders of magnitude faster than any competing tool.
Put simply, TIMi reduces the pain associated with Analytic Projects:
With TIMi, it’s becoming fast&fun to discover the knowledge hidden in your data.
During analytical projects, the analysts are typically working on two tasks:
(1) Data Management
Fast Data Management
Inside TIMi, all Data Management tasks are performed using Anatella.
Typical Data Management tasks include: data cleaning, ETL, KPI creation, feature engineering.
During a typical predictive analytic project done with TIMi, 90% to 95% of the time is spent inside Anatella.
(and thus only from 5% to 10% is spent using “TIMi Modeler”).
Working with Anatella is faster than with any other Data Management tool because:
Most of the data transformation operators are “meta-data free”. It means that it is not necessary to define “metas-datas” to use 99% of the various transformations available in Anatella. This feature is very important because, during predictive analytics projects, it’s very common to manipulate tables of tens of thousands of columns and it is impossible to specify “by hand” the “meta-datas” of all these columns (…and that exactly what’s all the other ETL tools force you to do).More about ''Meta-data-free''
Built for Iterative Work
Analytics Projects are characterized by their “exploratory” nature: i.e. Before starting such projects, you don’t know in advance the “right” KPI’s to compute or the “right” features to create. This means that Analytics Projects usually involves several “iterations” (e.g. at each iteration your KPI’s are “refined”, your “features” discriminates more, your dataset gets cleaner). Thus, you need a Data Management Tool that allows you to quickly iterate: i.e. it should allow your to re-order the “little boxes” in your Data Transformation Graph in anyway your want, to re-parametrize the boxes differently, etc. with as little “suffering” as possible. At our current knowledge Anatella is the only solution that allow that.More about ''Meta-data-free''
Instantaneous Data Preview
Click any point inside your Data Transformation Graph and instantaneously see the *FULL* data-table that goes through this point. This functionality can only be made possible through a user-friendly “cache” system that is unique to Anatella. This unique cache system allows you to create (and debug) extremely complex data transformations including several hundreds of operations (i.e. of “little boxes”) because you can, at any time, check if your data flow behaves “as expected”. All other “visual” tool are limited to simple data transformations (above a few dozens boxes, they becomes really cumbersome to use) because their “debugging” capability is limited because of the lack of a good “Instantaneous Data Preview”.
Fast ETL Engine
Never wait anymore for a data transformation. Anatella handles tables with billions of rows and thousands of columns even on small infrastructures.More about the ''ETL'' engine
No code & Code re-usability
You can develop 99% of your data transformations using only your mouse. Forget about complex R, Python, Scala code: Embed you code inside “little boxes” that everybody can use. You can create with your mouse complex data transformations that use all the power of the Anatella,R,Python&Hadoop engines. Anatella already contains hundreds of pre-made “little boxes”, ready to be used. As a developer, I love to code but I hate to always “re-invent the wheel”. Thanks to the intuitive, mouse-based interface, I can re-use in a few mouse clicks very complex transformations that would otherwise costs me several hours of coding. Also, I can directly use the “little boxes” produced by my colleagues to gain even more efficiency. Anatella is the tool that offers an ”Abstraction Layer” that completely encapsulates your R/Python/JS code for easy re-use.More about ''R & Python Integration'' More about the ''Abstraction Layer''
Built for Predictive Analytics
Predictive Analytics projects are characterized by the size of the manipulated tables: i.e. It’s very common to have “analytical datasets” (i.e. tables) that contains several thousands of columns/variables. For example, our pre-made solution for Telecom contains a customer-view with 3000 variables. The Predictive Analytic solution developped by a well know insurance company contains a customer-view with 23000 variables. Manipulating such tables (with a high-column count) through standard SQL expressions is just a nightmare (I remember “assembling” such SQL queries using a lot of copy/paste in Excel! Aaargh!): This is error-prone and very inefficient. Since Anatella is built from the ground-up for Predictive Analytics, Anatella handles flawlessly tables with more than 30000 variables (this is unique): The Anatella’s interface is built for high-column-count tables: No more copy/paste in Excel!More about ''Machine Learning''
Fast Predictive Modeling
When creating a predictive model with TIMi, everything is automated: (1) The creation of variable recodings, (2) The selection of the optimal modeling algorithm, (3) The correct parametrization of the algorithms to avoid over-fitting, (4) The creation of charts that allows to easily check for Over-fitting (just in case), (5) The creation of MS-Word (and MS-Excel) Reports that allows you to easily understand and explain your model (e.g. to check for “leakage from the future” and other bad scenarios). This allows you to easily understand how your model works and quickly take the right action to improve or correct your model.
For example, Bouygues Telecom went from 1 churn model (refreshed once per year) to more than 100 TIMi models (refreshed once per month), while having less data scientists.
Fast Deployment (scoring), Fast Maintenance (rebuild)
I don’t count how many times a good predictive model was not used because it was (nearly) impossible to “put it in production”. It’s very common to spend days, sometime weeks or months, to obtain a production system that computes in reliable way the correct predictions. On the other hand, deploying one TIMi model (or several thousands TIMi models), is a matter of minutes. TIMi offers dozens of options to easily use your models. For example, you can use a model just by “dragd&droping” your “model file” inside an Anatella windows: It’s that easy! …or you can export in one mouse-click your models to SQL (all SQL dialects from all databases are supported), to easily use your models everywhere, any time.
No data cleaning required/Not sensitive to missing data
We optimized the implementation of the algorithms used inside TIMi to that they are mostly insensitive to dirty data (outliers) or missing data. Using other tools, it’s quite common to loose months in data cleaning processes.
Forget costly, lengthy and code intensive data quality & pre-processing, or laborious and tedious modeling procedures: The creation of new model is simple, intuitive & versatile, with just a few mouse-clicks.
Fast & Scalable Engine
The Timi modeling engine is extremely fast (it’s 100% multithreaded, vectorized, cache-optimized). For example, BBVA noticed that TIMi creates in 40 minutes, on one 2000$ laptop, a better predictive model (more accurate) than an expensive competitor running for 7 hours on a very large server (the hardware cost of the server is 13 millions euros). Another example: For BNP-Paribas Belgium, we created with TIMi in 8 minutes, on a 2000$ laptop, a better predictive model than a (linear) predictive model computed with the same comon competitor that was running for 5 days on their largest solution dedicated server from BNP-Paribas Belgium.
TIMi has a proprietary in-memory compression engine that allows to store a 1GB dataset in less than 100 MB RAM. TIMi also works in high-dimension: i.e. TIMi handles easily datasets with more than 30000 variables. The combination of high speed, high in-ram-compression and high-dimension allows TIMi to analyze practically any dataset size without any effort from the data scientist. Other tools requires you to spend several weeks (sometime months!) to get a subsample of the rows/variables that can be “digested” by your limited modeling tool. No such things with TIMi! No more time lost for nothing…
Automatic Variable/Feature Selection
Not only TIMi delivers predictive models with superior accuracy (see our recurrent outstanding results in machine learning competitions) but it also delivers precise “insights”. For example: You want to know: What are the main drivers behind the churn of a group of customers? What are the main “sensor-readings” that decides if your truck/plane/machine will fail in the upcoming minutes/hours/weeks? We got you covered. The algorithm behind TIMi’s variable selection is one of the most advanced and accurate available today: i.e. It uses an enhanced “wrapper approach” that surclasses all other algorithms in this field.
Despite his superiority, the “wrapper approach” is practically never used because it implies very large running times. Thanks to a unique implementation, we reduced the running time to a minimum. This means that TIMi is one of the very few tool (maybe the only one) that is using such superior approach for variable selection.
This means that you can give, in a few seconds, very accurately answers to questions like “What are the main drivers/sensors/variables behind the observed phenomenon?” while other tools will use either more rudimentary/imprecise algorithms or takes literally hours of computation times.
Practically no risk of Over-Fitting, ever
Over-fitting control is always a big concern, especially for the “difficult” models (i.e. the models with “little lift” such as most churn models). 99% of the time, TIMi automatically delivers a perfect model without any over-fitting. For the very few cases where you still observe over-fitting (you can easily check for over-fitting thanks to the many charts automatically generated by TIMi), the TIMi interface contains a few meta-parameters that allows you to easily control the modeling process to remove any over-fitting.
Create now un-rivaled Timi predictive models on large graph based data sets using LinkAnalytics. The ultimate solution to extract advanced Social Network Algorithms metrics out of gigantic social data graphs.
We reduced by 10% the churn on the customer-segment with the highest churn rate.
The Timi Suite includes a very flexible ETL tool that swiftly handles terabyte-size datasets on an ordinary desktop computer.