MACHINE LEARNING &
AI HACKATHON
SANTA CLARA
30 SEP ~ 01 OCT 2017
What is DAVinCI LABS?
It is
The easiest & quickest machine learning modeling tool for both experts and non-experts.
In a nutshell
If you know what you want to predict, and have the data that includes it,
you can build models like a machine learning scientist
Why use DAVinCI LABS?
You are ready to model
You have clean data set, and know exactly what you want to predict.
No time
There isn't enough time for you to experiment with various data transforms and ML algorithms.
You need deliverables
You need to quickly test & evaluate your models for reporting and implementation.
OR maybe
Your current model isn't performing well.
Download User Manual
Click to download the complete user manual
Get Started

Unique Features
Convenient User Interface
DAVinCI LABS requires zero coding. It automates and optimizes key data analytics/modeling functions with AI. By minimizing user intervetion, DAVinCI LABS minimizes human error and works like a skilled data scientist.
Accessible via Web Browser
DAVinCI LABS is accessible via web browser. Official browsers are Internet Explorer 11+ and Chrome 58+. Users login to the DAVinCI LABS website with their own username and password.
Automatic Data Transforms
DAVinCI LABS automatically and optimally transforms numeric and categoric features to achieve better training of the models. The transform is processed based on interaction between each variable and the selected target. The transform includes several processes including data smoothing, selection of relevant categories from categorical variables, imputation of missing values.
Stability Index
DAVinCI LABS provides stability index of every variable, showing how reliable the variable is over time in terms of predictive power. DAVinCI LABS supports both categorical and numeric variables and visualizes predictive power of each variable for considered periods of time.
Automatic Removal of Bad Features
DAVinCI LABS automatically detects variables which have very low/no predictive power with respect to the chosen target variable. Automatic removal of bad features supports both categoric and numeric features.
Auto Selection of Algorithm
DAVinCI LABS provides automatic selection and tuning of the best algorithm given user-defined dataset settings. Auto-seleciton supports Gini, MSE, MAE and K-S as optimization criteria. Result of auto selection lists all algorithms ordered by their quality and each algorithm can be tested.
Flexible Auto Tuning
DAVinCI LABS supports Gini, MSE, MAE and K-S as optimization criteria for automatic tuning of hyperparameters. Automatic tuning is provided in 2 modes: simple and complex. Simple tuning guarantees faster search while complex provides more comprehensive search in the hyper-parameters space.
Ensembles
DAVinCI LABS supports training weighted combination of models (a.k.a "stacking") with use of K-fold cross validation for more reliable selection of coefficients when combining the models.
Classification of Asymetric Data
When performing classification algorithms, if the class distribution in the target field is imbalanced, classification performance could be poor. Especially, in case of severe data asymmetry with more than 99: 1, all the input values are classified into one class, resulting in bad performance. In order to solve such problem, the prediction value of the missing class is weighted so as to classify it into the corresponding class, and the weight value is optimized by data using the optimization technique.
Detailed Report of the Training
DAVinCI LABS simultaneously calculates MSE, MAE, K-S, Gini, AUPRC and PSI for training and validation sets after the training. The calcuation and detailed report can be downloaded in Excel format.
Automatic Comparison of Data Distribution
DAVinCI LABS checks whether the train dataset and the validation dataset come from statistically similar distribution. Several statistical techniques are used including t-test. DAVinCI LABS sends a warning signal if the distribution of the fields with significant impact on the trained model is different from the distribution of training dataset and validation dataset.
Exportable Prediction Function (Available for paying customers)
DAVinCI LABS allows user to export a trained model, including pre-processing for missing values and custom formulas, to a separate .dlm file that can be loaded and executed with a standalone Java library bundled with the .dlm file. The exported model is executed locally without any need of an external server or service, allowing for flexible integration into the user's own infrastructure. Further, a C language wrapper is provided for executing the model from native code through the Java JNI if the user prefers integrating with C instead of Java directly.
Feature Importances
DAVinCI LABS calculates feature importances for all models including neural networks and deep neural networks. The result presents relative importance of every feature used by the model.
Prediction Explanation
Prediction function provides information on which features caused the biggest impact on the prediction value for every data sample. The result is represented as a list of feature impacts with strongest features going first.
Automated Clustering and Visualization
For the specified target field, clustering function automatically finds a cluster whose target average is substantially high or low relative to the global target average. The process of finding dozens of clusters is as follows; First, DAVinCI LABS finds one cluster, and then sequentially search for the next cluster by boosting. Each cluster is defined by several simple conditions (e.g. 0.5 < A < 1.2 & 0.2 < B).
You can visualize them in 3D space according to the conditions and specifications of each cluster (size, average of the target fields, etc.), from multiple angles.
PROCESS
UX that navigates like a data analyst
AI engine proven with excellent results
Discovering hidden patterns inside diverse datasets requires sophisticated data pre-processing and analytics technology.
In DAVinCI LABS, we have bundled years of research and know-how gained from enterprise projects by our world-class machine learning scientists
A
Data Upload
1. Structured relational data that includes target information
2. csv per model
3. Metadata feature available
B
Target setting & data preprocessing
1. Set the target, and then transform with one click
2. Check input field information
C
Model training
1. Set data splits, and model output options
2. Training options: auto-selection, single algorithm, and ensemble
D
Model testing & reporting
1. Input a .csv file of test data to produce predictions
2. Download predictions to perform evaluations on your analytics tool
3. Download reports
PHASE 1 Prediction model generation via Machine Learning
(Training models based on past data)
 
Client DB (Big Data)
Finance, commerce, manufacturing, medical, etc.

Model training & analysis
Data pre-processing &
pattern recognition

Prediction function (scoring model) generation

Mount prediction function on existing system
 
PHASE 2 Operate & update prediction function
(Compute prediction for new data / algorithm retraining)
 
Accumulation of new incoming data

Real-time operation of prediction function

Apply predictions to business strategies

Algorithm retraining & prediction function update
 
Video GUIDE
Get Started
Click to download the complete user manual
Download User Manual
If you have any questions about the DAVinCI LABS,
Send us an email at contact@solidware.io

©2017  DAYLI Intelligence. All rights reserved.