Release Notes

What's new in version 6.3.2

Released: December 19, 2024

General:

Added story name and id in driver overview export.
Improved layout and font sizes in pair-wise plots.
Improved styles of data indicator icons in drivers overview slide.

Bug fixes:

Model export was not working correctly due to new version of Pandas 2.
New slider component to resolve resizing issues.
Global field mutation in parser issue, caused model validation to return wrong results.
Predicted vs. actual slide changes to resolve inconsitentcies in metrics.
"Change drivers" modal does not always display the proxies.
Missing data bars in do not always display in Driver Overview.

What's new in version 6.3.1

Released: November 8, 2024

Bug fixes:

Unnamed columns in Excel with blank string caused crash.
Expand KPI names did not work correctly under drivers plot.
Share button was disabled inside of the story.
Share button did not appear in stories not owned by the user.
Driver Overview Table was resizing badly.
Full page dropzone "progress" did not display.
Uploading corrupted file does not display error.
Scatter plot of KPI slide did not display for specific story.
Regressions on various plots.

What's new in version 6.3

Released: October 11, 2024

General:

Average importances are added to the Excel export of the driver overview.
Improvements to story sharing. Users will now be able to share a story with someone that does not have an account yet. The new user will get an email with the shared story and the request to create an account. Once they created the account, the shared story will be available. This is useful for sharing results with clients during an engagement, as most clients do not have an account yet, consultants will not have to be waiting for the account creation to share results. Additionally, new permissions have been added, like the ability to prevent users from cloning and resharing the story.
New functionality in drivers overview to indicate the data availability for each driver.
Adapted pairwise plots that allow the user to customise the graphs. Users can change the width of the labels section, bar plots, adapt the text sizes and opacity of the points. There is also a high-contrast mode for presentations and screenshots.

Bug fixes:

Removed incorrect filtering info on imported stories.
What-ifs slides do not handle some negative drivers correctly.
Ordering of KPI not coherent in KPI plot and Driver plot.
Confusion matrix is not symmetric.
Cloning of filtered story switches histograms to filtered data when adding a filter.
TableParser removes successive white spaces.
Pairwise plot and Linear vs non linear not agreeing on rounding.
Import of stories with version 5.2.2 or older fail.
Filtering fails when you change the data source for some data sets.
Cloning an imported story has more records than before.
Filtering components flickering issues.
Disappearing additional tooltips selection after filtering.

What's new in version 6.2

Released: March 28, 2024

General:

Linear regression lines and p-values are now available on the pair-wise plots.
Additional statistics on the KPIs plot. User can now view the min, max, mean and median on each plot.
Mark filtered out data and outliers in scatter plot of KPI plot.

Bug fixes:

Thumbnails turned bold on What-Ifs slide.
Histograms do not follow the Y-axis when the option Align KPI axis is selected in the KPI Slide.
Change driver option in driver overview results in a new story that cannot be exported.
Pagination is not correct for “drivers + selected columns” in data summary table.
Select/Unselect does not select as intended.
Data health wheel-filters are disconnected from table.
Checkbox “highlight outliers” has inconsistent behaviour.
Report dialog for filtering is not clear.
Cloning an imported story from SDK fails.
Toggle styles not correct in story creation page.
Dropdown keeps appearing after changing the data source.
Description field difficult to edit on story overview page.
Align axis on KPI plot is not correct.

What's new in version 6.1.5

Released: December 20, 2023

General:

Show .dsx filename as data source when importing stories

Bug fixes:

Issues during filtering and story creation, user KPI and input selection was lost after update of data summary table.
KPI selection lost after filtering of new datasource.
Driver overview in story overview uses file from wrong run.
Wrong information in StoryOverview when creating two separate stories.
Story rename doesn’t works in story overview.
Prediction and validation slide metrics text cropped.
Datasource panel wrongly displays filtered datasources.
Timestamp column in data set makes code crash (SDK related).
Avatar selection no longer working after components upgrade.
Datasource not found for shared story.
Remove meta information from imported stories.

What's new in version 6.1.4

Released: September 21, 2023

General:

Simplified the simple correlations slide, now there is just one checkbox “Hide variables without connections”.

Bug fixes:

Resolved issue with parsing of csv files with special characters in front-end. Some slides, like pair-wise plots failed with these data-sets.
Model validation slide bug-fixes and improvements.
Front-end issue on what-ifs page.
Index was used instead of column names for selection and driver labels when changing a data source during cloning.
Unnecessary bar chart plot for categorical inputs on kpi page.
Histogram labels were not rounded in data summary table/data health. Note: Will only be resolved for new stories.

What's new in version 6.1.3

Released: July 13, 2023

General:

Ordering of the drivers in Driver Overview is now more consistent. In the past it was only sorted on “overlap”, but because the overlap was often 0, the order of the drivers was mostly random for less important drivers. Now it is sorted on importance as well.

Bug fixes:

Resolved bug in sampling of data. Finding the min and max was computed wrong for columns with missing values.
Resolved confusing behaviour when deleting shared story and datasource, a wrong message was generated when a shared stories was deleted by one user and cloned by another.
Cloning a story with filtered data skipped the title tab and went directly to the data summary table.
.XLS (old excel files) support was no longer available in predicted vs actual slide, is now working again.
Resolved what-Ifs missing the actual value when select a row from the table for data sets with special characters.
Error notifications during import of several stories via import page.

What's new in version 6.1.1

Released: April 13, 2023

General:

Toggle the export of PPTX, it will be off by default. This way we can avoid the long generation times, but still have it as an option for some users.
Better error handling, stories with some failed KPIs will indicate this in the story overview. Stories with all KPIs failed are now also better handled.

Bug fixes:

Cloning broken for data sets with special characters.
File with special characters not parsed on PVA slide.
Cloning of old story is not possible.
We are trying to load drivers overview of failed story.
Model validation slide is failing with specific test data set.
Story cloning: wrong driver labels for specific data set.
Story fails when some runs fails (models cannot be merged).

What's new in version 6.1

Released: April 4, 2023

General:

All new stories are now using the Multi-KPI story framework from the DataStories SDK. You can now import and export any story from and to the SDK, share it with colleagues or customers between standalone versions or create an offline backup of your most important stories.
New functionality to allow the filtering of a data source before story creation. Remove records that outside of a certain range or only use records that have a certain label on the fly with only a few clicks during story creation.
Native UTF-8 support for all columns and values, this should make easier to work with international data sets. If you want to make your data look cool, be sure to add emojis :mushroom::potato::+1::skin-tone-2::clock3::thermometer:
New functionality to change the drivers of an existing story with other inputs that are strongly correlated (for example replacing an average with a set-point that is controllable). You can also remove a driver that you do not like, while preserving all the other drivers.
Model quality and driver importances are now available directly from the story overview, just expand the story and access them immediately.
Upgraded our worker cluster, users can expect significant improvements in the time it takes to compute a story, sometimes up-to twice as fast.
Improvements into tagging of multi-kpi stories, it is now easier to find a story back irrespecutful if it was a single or multi-kpi story.
Improvements to PPTX export for multi-kpi stories.
You can now access the slides as soon as they are generated, no need to wait on the exports or PPTX.
Improved Binary KPI error metrics for Multi-KPI stories.
New functionality in story cloning. Users can select to only show the drivers or previously selected inputs.
Improvements to sampling policy when showing larger data sets. Platform will now be able to handle up-to 10k rows in all plots.
All statistics on slides are now based on the “modeling dataset”, which is the data that is left after removing out all invalid rows, ie. rows having a missing KPI values.

Bug fixes:

Resolved failing statistics on columns containing only NaNs.
Model validation slides gives strange error when predicting a particular test data set.
Record info was missing in Driver Plots.
Driver Overview Slide does not load for multi kpi when a kpi failed.
Fix performance of downloading predictions in the Prediction and Validation slide.
Frequency plot for binary KPI is using sampled data instead of full data set.
Histogram does not preserve order of bins/bins are not correct.
Solved non descriptive error message.
Story fails on DataHealth when no numeric columns are included.
Single driver scenarios cause typing errors.
Interface issues with wrapping text that has no spaces.
Issues with deleting data sources.
Fix the size and metrics spacing for the confusion matrix.
Highlight model driven outliers is not working when clicking connect the dots for multi-kpi stories.
Pairwise plots are not behaving correctly on smaller screens.
Pair-wise plot crashes when single quote in record info names present.
Outlier elimination is on for all imported stories.
Data summary table does not sort the KPI column correctly.
Align both graphs of pairwise plot at the top.
Zoom does not always reveals all points.
Data distribution quality indicators broken.
Error when downloading predicted-vs-actual file.

What's new in version 6

Released: March 24, 2021

General:

Added the ability to import and export stories between Platform, SDK, and local Platform installations
New computational engine with speed improvements of upto 6x during model training and better handling of smaller data sets
Improvements of the jitter functionality in pairwise plots.
Add linear-vs-non-linear slide to the multi KPI story
Improved driver overview table with driver proxy functionality
Improved the performance of the Predict vs Actual plot (for larger data sets)
Extended KPI slide for multi-kpi stories
Added Driver-plots for multi-kpi
Improved plots to handle larger data sets and subsample when needed to keep the display responsive

Bug fixes:

More tags than kpis are created if there are commas in the kpi name
Cloning is not preserving the selected inputs
404 error when calling HeadObject operation: Not Found
Fix sorting on the "Percent Available" column in the data summary
Datasources are coupled upon cloning
Execution exception in production story
Downloaded prediction files are missing columns
Missing outliers column in DataSummary component
Story creation warning upon cloning
Multi KPI slide - wrong order in barchart
Number of runs parameter resets to 3 when cloning a multi kpi story
Issues with what-ifs and data table
Unaccounted empty categories in box plots
Fix graphics related problems for lcertain data sets
Sorting doesn't work and crashes the plot on Prediction and Validation slide for specific story
Missing values should not share the valid values axis
Pairwise plots inconsistencies
Clicking the vertical axis in the control chart makes plots dissappear
Model validation graphics crash under large data sets
Slow loading WhatIfs data table with large datasets
Stories with a large number of KPIs crash the layout
Driver-plots "Show Outliers" button broken.
Data not visible in the single KPI plot
Align KPI axis messes histograms in multi KPI plots
Missing categorical driver plot
Exporting an imported story
Import story re-run does not update [Driver Overview] slide
Fix the formatting of model based outliers in multikpi Model Validation
Fix the Conclusion slide layout with many KPIs
The 'connect the dots' functionality freezes the page for specific data set
What-ifs keeps showing the actual value during the optimisation step.
Optimizer desirability ignored when typing in the values instead of moving the sliders
Proxy analysis in driver overview misses candidates
Story import misses columns in filtered data

What's new in version 5.2.3

Released: December 21, 2020

General:

Improvements in story cloning when changing the data source
Visual improvements of driver overview slide

Bug fixes:

Table under What-ifs not aligned for certain data sets
Resolved issues for certain data sets in the histograms of the KPI overview slide
Missing values not shown correctly in drivers plot

What's new in version 5.2.2

Released: September 28, 2020

General:

Versions of Cloud Platform, Standalone Platform and SDK are now aligned at the same version

Bug fixes:

Export to DataStories (RSX) was not always available for Pro pack
Issue with vertical axis for binary KPIs in WhatIfs
Issue during model validation when subsampling
Multi-kpi What-Ifs did not always change the reported predicted value
At least one KPI should be selected was returned by mistake during cloning

What's new in version 5.2.1

Released: September 2, 2020

General:

KPI plot added record info
Model validation for Multi-KPI, date time is now after title of data set
Story type is now shown in Story Overview
Improvements of clone story with new data source
Filter for highly correlated variables during modelling
Improved what-ifs performance for multi-kpi
Report the datahealth before auto-scaling, so it corresponds with the histograms

Bug fixes:

Ordinal variables are not working correctly when exported to Excel
Outlier detection bug, caused issues when modeling stories
Filtering of tags not working correctly for single vs multi-kpi stories
Error when exporting binary KPI to Excel
Story rename not working as expected
What-ifs, sliders behave strangely when “hide irrelevant drivers” is active
Issue with ordinal categories in pair-wise plots
Model validation failing for specific data set
Model export of multi-kpi is not working when there are special characters in the KPI names
Create story assumes that the user is changing data source when going back from step 3 to 2
Escape is going wrong in what-ifs
Issues for certain Multi-KPI stories
Histograms bars on KPI slide are not in the correct order
Align KPI does not go below zero
Driver plots are missing for some variables

What's new in version 5.2

Released: July 3, 2020

General:

Binary KPI: Calculate error metrics and values for the confusion matrix for both Positive and Negative cases.
Optimization now available in what-ifs slide
Multi-KPI:
- KPI Slide with histograms and scatter plot (only when 2 KPIs)
- Even when one of the KPIs fails, a story is still generated about the other KPIs
- Data health slide
- Model prediction and validation slide
- Pair-wise plots slide
- Conclusion slide
- Multi-objective optimizer with trade-offs in what-ifs slide
Textual improvements to Platform slides
Handle trivial correlations with KPI better

Bug fixes:

Some older what-ifs no longer work (before 2019)
Tag in story overview is not retrieved upon searching
Improvement of description and white space
Story fails and then gives a 500 error response.
Issues with UTF-8 character in Pair-wise plots

What's new in version 5.1.13

Released: May 27, 2020

General:

Allow export to DataStories SDK in What-Ifs slide
Introduced possibility to select/upload a new data source when cloning a story, inputs and KPIs will be matched.
UI improvements in Simple Correlations slide
UI improvements in What Ifs slide
Improved stability of algorithms for data-sets with very low correlation.
[For Beta users] Multi-kpi story

Bug fixes:

On Drivers slides, plot not sticky when scrolling
Typo in the error metrics
Resolved missing tooltips that provide record info
Formatting of predicted value on What-Ifs slide was not behaving correctly
Workaround for Chrome bug with background filters

What's new in version 5.1.12

Released: February 24, 2020

General:

Improvements to API for story creation
Better exception handling to recoverable errors
Stability improvements to the engine
Textual improvements on error pages

Bug fixes:

Plot on the right in Pair-Wise Plots slide not always rendering correctly
Actual value on What-ifs slide is not displayed
What-if table problems for larger data sets
Model error plot was causing issues in Chrome
Model validation failed on special formatted CSV

What's new in version 5.1.11

Released: November 28, 2019

General:

Introduced an option to generate API keys in the account for integration with external resources
Removed a requirement for min 30 rows in the dataset to start a story, for now we require only 20
Made the design adaptive for smaller screens (like iPad)

Bug fixes:

Fixed a bug when it was impossible to login at direct link on sign-in page
Fixed a bug when contact information is not saved in Settings
Fixed a bug when a sharing a story with not-existing user, the story didn’t pop-up in the account of this user after his account creation
Fixed a bug on what-ifs when labels of plots overlapping each other.

What's new in version 5.1.10

Released: October 1, 2019

KPI slide:

Added a histogram of distribution for NUM KPI with a possibility to check the outlying values as records in the original data (available both for old & new stories)
Switched a plot color from pink to cyan blue

Linear vs Nonlinear:

Started to use original variables names instead of internal ones (taking special characters like ' into account)

Pair-Wise plots:

Introduced new types of plots to show relations between the KPI and a variable, where taking into account variables type (available only for new stories, for old stories the scatter plots remain):
Introduced jitter option for scatter plots when shown for BIN or CAT variables
Started to use subsampled data file, subsampling the data with more than 5000 records up to 5000 randomly. Outliers elimination is taken into account, so outlying records are not there in case they were removed

Model Validation:

Introduced better visualization for the discrete NaN points (previously the neighbouring available points where connected ignoring NaNs present between them which gave the wrong visualization)

What-Ifs:

Made a limit of 10 MB for the table size to be loaded with the full data, otherwise loading only the columns of KPI, drivers, special columns, $predicted, $predicted score, $record number, $confidence
Introduced new 7.0 version of the slide (technical improvements), including processing of errors for autotests
Introduced “lazy loading” of the components on the slide
Fixed a bug when data set with special characters in column names broke the slide (data with ” )

Driver plots:

Fixed a sorting of labels for NUM values (by value) and TXT categories (lexicographically)

General:

Removed sign-up & sign-in with LinkedIn account options
Removed a default account creation when signing-in with Google account option
Fixed “Forgot password” option on sign-in page (pop-up window didn’t act)
Fixed cookie banner

What's new in version 5.1.8

Released: June 26, 2019

KPI slide:

Added the option to highlight outliers in the KPI plot (available only for new stories)
Changed the way of subsampling the KPI records when there is > 1000 records present in the data (from step-wise to random-wise)
Introduced grey color to missing category in histogram for binary KPI

Your Stories page:

Added a possibility to filter by a data source
Added a type of the search attribute seen in searching results (id, tag, title, author, KPI or source)
Added "Interrupt story" button with a possibility to terminate the running or stuck story
Fixed a bug with tooltip hidden for the first story in a list

Story creation:

Reverted the tag-input back
Fixed a bug with deselected hovered KPI button

What-Ifs:

Introduced a warning when clicking a row of the table with N/A prediction caused by missing or outlier value in a driver (available only for new stories created after the release date)
Introduced buttons refactoring (aggregated export buttons + moving the buttons to different place)
Fixed a bug when actual value was shown by clicking on Maximize / Minimize buttons
Improved the behaviour when too small value of a driver was shown in the inout field (for example, 4.9872173e-10 instead of 0 )
Fixed issues when showing incorrect predicted and actual value by clicking a row of the table

Model Validation:

Allowed downloading the predictions for the validation data without the KPI column
Fixed a bug when "id" column chosen as column to be used in tooltips was used as "Record ID" to sort the plot
Fixed a bug when the green line stopped being shown after meeting the missing value of actual (for Prediction per Record plot)
Fixed a bug when incorrectly calculated the number of records where error-metrics were calculated

Other improvements:

Fixed a bug when some old stories (created before June 2018)
Introduced a banner for the IE
Introduced a new optimisation algorithm to deliver MAX/MIN values of the KPI on the What-Ifs
Fixed a bug on Pair-Wise and Driver Plots when showing missing values as a separate category
Changed a label of the scale for Linear vs Nonlinear for consistent ones

What's new in version 5.1.7

Released: June 5, 2019

Model Validation:

Allowed to download the prediction file for the empty KPI column
Changed the text of error when uploading validation data file without a driver, showing a missing driver name
Allowed to upload validation data file in a format different from training data (excel vs csv and opposite)
Fixed a confusing behaviour when replacing the KPI with const = 1 when uploading validation data file with empty KPI column
Fixed a bug for classification problem when producing incorrect matrix when binary KPI contains missing (interpreted as ?-category)
Fixed a bug with calculating the number of records not predicted being outliers in drivers (which is showed in the text)

Plots:

Removed the transparency
Introduced bigger fonts and consistent labels
Introduced better margin for a watermark on the plots
Reduced the width of the plots (Linear vs Nonlinear) to be consistent to the style of other slides

Story creation:

Introduced the colors for bars representing missing values (consistent with Data Health page: red/yellow/green)
Fixed a bug when missing and total number of cells were not changing when selecting/deselecting columns
Fixed a bug when story could not be created from Data Source page

Other improvements:

Fixed a bug when outlier elimination option was ignored when cloning a story
Fixed a bug with broken What-Ifs plots and table for data with ' character in a header
Fixed a bug with broken Driver Plots for data with ' character in a header
Fixed Restart button in Admin UI
Added a sorting option (by KPI value, by record number) on the KPI slide

What's new in version 5.1.6

Released: April 16, 2019

Pair-wise plots:

Introduced pair-wise plot in vue.js
Changed the layout of the dots for better density visualisation
Added search-field to the plot with the option to filter on variables
Added the possibility to fix a dot and track its change through the variables change
Fixed a bug with zoom option on the plot and added the Reset zoom option

Data Health slide (the changes are available only for new stories):

Changed the average per column metrics to be calculated only on the story inputs, not in the whole data and also:
- If TXT column was included, the values there are not converted to missing values any more
- Corrected outliers statistics calculation
- Distribution quality calculation is based on fairness, not on balance any more
Added tooltips to the metrics
Added outliers percentage and absolute value per column with gradation to BAD/ REASONABLE /GOOD in the data table
Added distribution quality per column with gradation to BAD/ REASONABLE /GOOD in the data table
Added gradation to BAD/ REASONABLE /GOOD to Percent Available column in the data table
Added a possibility to filter the table values by clicking the donuts parts.

Story creation page:

Introduced the whole page in vue.js
Introduced small frontend changes, such as:
- Changed the error texts and introduces some hints on data requirements
- Better drop-down list of data sources

Other Improvements:

Allowed models export for binary classification
Fixed a threshold applied for the validation data set
Improved mapping of classes for binary KPI (in particular, 0 is never mapped to 1 any longer)
Fixed a bug with exporting story to pptx, when export starting by clicking 2-3 times in a row
Fixed a bug with impossibility to log in after logging-out
Made more clear error-message when uploading a test data file without KPI column on Model Validation slide
Added drivers importance on What-Ifs slide
Fixed a bug when impossible to run a story with all BIN variables
Fixed a bug on What-Ifs with broken plots when clicking a row with a missing value in CONST driver
Fixed a problem with long names in tooltips on Driver Plots
Introduced the same style for all the bar charts on platform

Technical improvements:

Updated the docker up to python 3.7.

What's new in version 5.1.5

Released: March 21, 2019

General Improvements:

Made the left menu panel expandable for more convenient usage of story slides
Introduces the improved slide of story conclusions
KPI plot: introduced "Connect dots" and "Reset zoom" buttons
Made a tag-section on Story Creation page as an input
Introduced a new (vue.js based) table of the data summary table on Data Source page
Changed minimum and maximum radius for Simple Correlations graph to avoid glitches
Added an option to clear all the items at once on the search field on Simple Correlations slide
Fixed a bug on driver plots with incorrect tooltip values for date variables
Fixed a constraining behaviour for predicted value on What-ifs slide when clicking on the table rows

Technical improvements:

Introduced the integration of python API in worker of the platform

What's new in version 5.1.4

Released: February 12, 2019

Graphs and charts implemented with new components:

KPI slide (distribution histogram and KPI plot):
- implemented this graph in vue.js component
- added reset zoom option to KPI pink plot
- added percentages to each class in distribution histogram
Linear vs Nonlinear slide (bar chart with linear correlation and mutual content):
- implemented this graph in vue.js component
- added percentages to each variable of the plot
- added gradient to the bars
- added X and Y axises for plot to look more pretty
- adapted the width between the charts to the width of the variables names
Predictive models slide (bar chart with drivers importance):
- implemented this graph in vue.js component
- adapted the plot and the text on the slide
- added ?-icon with the AUC explanation for binary classification problems
Model validatio slide (confusion matrix for binary classification problems):
- implemented this graph in vue.js component

What-Ifs:

Fixed a bug when table crashed because the original data table contains the same name as internal name of column in What-ifs table
Fixed a problem with rounding float numbers in the table
Fixed a bug when "Actual" value was not displayed for some data sets

Other Improvements:

Corrected the values for types of the columns on the table of Data Overview slide
Fixed a limitation for binary classification problems when KPI distribution histogram displayed not all values of the KPI classes (when there are more than 500 values in total)
Fixed issues with activity plot on Dashboard (bug in Firefox browser, excluded shared stories from counter of created)
Improved layout of blocks on Dashboard page (fixed fonts, blocks alignment, added a mark and tooltip of account creation event)
Fixed a bug with FPR metric not shown when 0-value on Model Validation slide
Fixed a bug when deselected columns were shown as selected after submit on Story Creation page
Removed sub-menu for "Export" option in top menu panel, created separate pptx and toolbox exports
Removed IP column from settings page
Fixed a bug, when in histogram for a column with 10 unique numeric values was shown range on Data Summary table
Rounded and sorted in logical order the values in box plots of Data Summary table
Rounded the error metrics to 3 digits after comma, hid the tooltips for the metrics that depend on 1-positive class
Fixed a bug when empty column was defined as categorical on Story Creation
Replaced login button on sign-in page to be intuitively correct.

What's new in version 5.1.3

Released: January 23, 2019

Your stories page:

Story description made to be saved online instead of after page refreshing

Dashboard:

Introduced a new plot of statistics on stories created
Introduced DataStories Help in 'Did you know' section

Data Overview slide:

Changed 'Not useful data term' into 'Percentage of missing cells'
Added formatting of numbers on a slide and introduced some text remarks

Other improvements:

Changed a text of the general error slide
Added KPI name to the top panel on every slide of the story
Made more clear text for progress-bar when story is in queue
Removed border from the element with additional settings on story creation page
Introduced more clear text below what-ifs table

Bugfixes:

Fixed a problem with numeric column name
Fixed a bug when what-ifs table was not connected to the plots for some data sets
Fixed a bug when it's impossible to delete a story from example account.

What's new in version 5.1.2

Released: December 24, 2018

Your stories page:

implemented a different color for newly shared stories for improved visibility;
introduced the AUC instead of the F1-score metric as the story accuracy for classification problems;
added a warning when entering an existing tag to a story;
made the filter field visible while scrolling and added search by author name, title, and story ID;
improved the loading of stories on page while scrolling;
fixed the issue with filtering by 2 tags in a row.

Data Health slide:

implemented a new summary story table with fixed histograms and a better searching option;
improved the description of summary statistics of data (missing data and balanced / distributed data) on the Data health slide.

Dashboard:

improved layout for for bigger screen resolutions (up to 2560px);
outliers statistics added.

General:

changed the icons associated with the slides.

Other improvements:

pair-wise plots: added a fix for the plot tooltip, where the variable name contains “-“ character and fixed bug with overlapping of labels;
added a check and error message preventing users to run a story with less than 2 input variables besides the KPI;
fixed pop-up message, when licence is expired.

What's new in version 5.1.0

Released: November 27, 2018

Story sharing:

sharing stories between users with email notification;
sharing with non-users of platform along with inviting them to create a trial account;
revoking access from certain users.

Your stories page:

page is sub-devided for tabs for better navigation;
filtering of stories by title, id and KPI is available;
story title and descriptions were made editable;
drivers are added to the story overview element.

Story creation:

new data overview table integrated with improved UI;
warning added about the minimum number of rows in the data set.

Model validation:

added the information with number of records (total, predicted, with missings and outliers) for training and validation data set;
added error metrics for binary classification problems;
warning added in case when it's impossible to calculate the error metrics.

Other improvements:

navigation panel improved for better navigation;
search field on edge-bundling graph added;
fixed what-ifs plots boundaries and when having 1 driving variable;
fixed what-ifs logic when binary numeric driver is observed;
fixed Predictive Model slide when having 1 driving variable;
required story title (impossible to create stories with empty title);
added a check that at least 2 variables selected not to be KPI;
added a check that there are at least 20 values non-missing in column to be selected as KPI;
fixed read-only view after submit on story creation page;
improved explanation when it's not possible to calculate model accuracy;
KPI plot: adapted initial ranges for the graph;
Pair-wise plots: bug fixes with overlapping labels;
settings page: small bug fixes.

What's new in version 5.0.3

Released: October 01, 2018

Validate your model:

We created a new slide for a story called Model Validation, where you can test how the final model ensemble performs on unseen data (called 'test data') by uploading a new data file and evaluating the model on every record of the file:
- The top plot will show you the difference between the actual and predicted values of the KPI;
- Enabling 'Highlight model-based outliers' will show you the records for which the prediction was far of the actual value;
- The error plots (or confusion matrixes in case your KPI is binary) will allow you to compare the behaviour of the model on the train data vs. the test data;
- You can download the results evaluated on your test data set in the same format as your original data (either CSV or Excel) using the download button. We add some columns in the new data with predicted values for the KPI, the confidence about our prediction and some extra characteristics;
Please, remember that the structure of the test data set should be similar to the story's data source. That means, all drivers should be present as well as the KPI and the columns naming should be exactly the same. If you have selected any additional columns as record identifiers to display on tooltips, those columns should also be present in the test data set.

Learn how the KPI and Drivers change together:

We created a new slide called Driver Plots for individual scatter plots of the KPI, all driver variables and linked them together, because it is often useful to explore how individually observed values were changing together in the dataset. You can change the view of the points: sort them by record ID as they appear in the data or by the KPI or any of the drivers.

Clone your story:

We added “Clone story” button on “Your Stories” page. By the click you can analyse the same data source from the other side: to choose another KPI(s), exclude any variables from the analysis or change other settings. By default a new story will inherit the configuration of its parent and you will be able to change it before the run.

Tag your Story:

From now you can tag your stories: on story creation page, in the story overview and even in the story results - look for the tag-symbol in the header of the report.
- after you tagged your stories, you use the dropdown selector at the top of 'Your Stories' page to display only the stories with certain tags. If you select multiple tags, only stories that contain all of the selected tags will be shown. Tip: you can click on a tag directly to filter by it.
The state of selected tags is remembered on “Your stories” page. You can move within a platform without worrying to lost your current story in the list of stories. As well as in a case of lost connection you won't have any problems to recover your filters when you load the page. The state is remembered for the time of your browser session.

Improvements to story creation page:

We added the optional section with 'Additional Settings', where you can set up the configuration of parameters before running your story:
- tag your story
- select up to 5 columns to show in the tooltips (which may convey special meaning to you about the records in your data)
- set the number of modelling iterations (explain for what is is)
- enable or disable 'Eliminate outliers' option (by default we clean up the original data set and remove all far outliers from the data. It may help to make the evaluation less sensitive to non-standard records that appear in your data)

More information about your data:

On Data Overview slide we added the number of outliers that were found in original data set before modelling in case the option “Eliminate outliers” was enabled.

Improvements to What-ifs:

We changed the layout of plots and fixed a lot of small bugs on What-ifs slide.
The predictive value is now included in summary table of the What-ifs slide so that you could easily compare it to the actual KPI value.

Other Improvements:

Simple correlation slide has got a new layout. Among others, some settings moved to a special menu (collapsed by default). We also added some small features to make the management of settings more convenient.
On Pair-wise plots slide the bar charts has got different color for negative values of linear correlation.