2022.4

• TIM Platform LTS 2022.4
• Released on 19.4.2022

---

TIM API

TIM DB

New features:

• Implemented the ISO 8601 standard throughout the REST API
  • Added timeZone as an optional parameter for uploading and updating of datasets
  • timeZone information is returned as part of dataset and dataset version metadata
  • All returned timestamps are in UTC format
    • Sliced data (download)
    • Metadata parameters
  • Data aggregation happens in the original dataset timezone
  • Switched to using ISO 8601 Durations for durations, such as samplingPeriod and calculationTime
• Panel data (or grouped time series)
  • Dataset type in TIM DB
  • Group key dataset columns
  • Category filters for group keys
    • Simple filter
    • Complex filter
• CSV upload
  • Automatic detection of CSV upload and update parameters (decimalSeparator and csvSeparator)
• SQL upload
  • Added samplingPeriod parameter to SQL dataset upload
  • SQL dataset upload automatically recognizes the timestamp column

Changes:

• Disallowed duplicate headers (variable names) in uploaded datasets
• Removed GET /datasets/supported-formats
• Improved timestamp validation during dataset upload
• Improved timestamp format validation and added supported formats to Swagger schema

Bug fixes:

• Improved error message on attempting to upload an empty file as a dataset
• Added possibility to delete datasets and dataset versions in status Registered, Queued and Running
• Remedied requirement on order of JSON and CSV in POST /datasets/csv and PATCH /datasets/{id}/csv multipart requests

TIM Forecast

New features:

• Panel data (or grouped time series) support
• Endpoint for model upload over any compatible dataset: POST /forecast-jobs/upload-model (available for forecasting models of v5.3 and newer)
• New relative ranges type First (for forecasting in-sample rows and out-of-sample rows)
  • Default behavior can now be explicitly set by type Last

Changes:

• Made type of offsetLimit optional in configuration in registration bodies of POST /forecast-jobs/build-model and POST /forecast-jobs/{id}/rebuild-model

Bug fixes:

• Allowed empty request body for POST /forecast-jobs/{id}/rebuild-model and POST /forecast-jobs/{id}/predict

TIM Detect

New features:

• Endpoint for model upload over any compatible dataset: POST /detection-jobs/upload-model (available for anomaly detection models of v5.4 and newer)
• Accuracy measures for (labeled) anomaly detection jobs
  • Endpoint for retrieving accuracy measures
  • labelColumn to data section of request body in POST /detection-jobs/build-model
  • Accuracy measures for anomaly detection jobs returned in GET /detection-jobs and GET /detection-jobs/{id}

TIM Explain

Changes:

• Updated and improved model results in Swagger
• Improved descriptiveness of error messages

TIM MLOps

New features:

• Telemetry endpoints
  • GET /ops/dataset-calls
  • GET /ops/job-calls

Bug fixes:

• Fixed importing of dataset during use case import
• Fixed exporting of use case with anomaly detection experiment(s)

General

Changes:

• Removed unused fields from token payload
• Changed expiresIn of a token to an integer representing seconds

Bug fixes:

• Fixed handling of a lost AMQP connection
• Improved error message when sending a valid JSON request body that does not match the expected schema
• Improved protection against SQL injection attacks

TIM Studio

New features:

• ModelZoo Browser (not available for forecasting models of v5.0 or older)
  • Model hierarchy
  • Model download functionality
  • Model Complexity
  • Table view
  • Variable insights
    • Variable offsets
      • Used offsets
      • Available offsets
    • Unused variables
  • Forecasting: the dataset's difficulty is displayed
  • Forecasting: each model's quality is displayed
  • The sampling period/timescale of the dataset the ModelZoo was built on is displayed
• Timescale and aggregation
  • on experiment page
  • on dataset detail page
• Data availability visualization
  • on experiment page
  • on dataset detail page
• The navigation bar
  • A user's current location in the application is followed
  • The experiment icons are replaced with forecasting icons or anomaly detection icons in the navigation bar
• Enabled downloading of job results from the iteration table on an experiment page (forecasting and anomaly detection)
• User group owner rights can be granted and removed
  • on the user group detail page
  • on the user group add page
  • on the user group edit page
• The treemap labels are expanded to include both the feature/term importance and coefficient
• For forecasting model building jobs
  • Allow offsets configuration
  • Offset limit configuration
• Enabled deleting of datasets from a dataset's detail page

Changes:

• Ranges
  • Absolute out-of-sample range shown by default (instead of relative out-of-sample range)
  • Default selected in-sample range is removed on initial manual selection of in-sample range(s) in line chart
  • Default selected out-of-sample range is removed on initial manual selection of out-of-in-sample range(s) in line chart
  • Improved consistency in visualization of multiple ranges in the iterations table between forecasting and anomaly detection jobs
• Sampling periods are displayed in the largest relevant unit instead of in seconds or months only
• Numbers are rounded to three decimals in feature/term names
• Visualized selected state of iterations in iterations table immediately upon selection (instead of after results are loaded)
• The landing page redirects to default workspace (thereby including workspace editing and use case importing functionalities)
• Removed navigation bar and top bar from 404 page

Bug fixes:

• Forecasting jobs
  • Fixed timezone offset for jobs with absolute ranges
  • Fixed displayed in-sample ranges in iteration table
  • In the RCA residuals line chart, values are shown as missing when the corresponding historical actual value is missing
• Anomaly detection jobs
  • Displayed ranges set as relative range in iterations table
• Fixed normalization of results: relative to target/KPI statistics

Python Client

New features:

• Forecasting interface
  • create_forecast
  • create_forecast_and_execute
  • rebuild_forecasting_model
  • rebuild_forecasting_model_and_execute
  • Added allowOffsets and offsetLimit parameters to the configuration of forecasting jobs of type build model
    • build_forecasting_model
    • build_forecasting_model_and_execute
    • clean_forecast
• Anomaly detection interface
  • rebuild_anomaly_detection_model
  • rebuild_anomaly_detection_model_and_execute

Changes:

• Added a reference to the release notes in the listing description
• Updated the listing description to mention 'InstantML'

Bug fixes:

• Fixed error on calling upload_dataset or clean_forecast multiple times in the same session
• Fixed docstrings:
  • Response of upload_dataset: dataset_metadata Dict instead of ID and Dict separately
  • Response of clean_forecast: forecasting_logs are returned before dataset_logs
  • Response of get_anomaly_detection_results: no accuracies are returned

Architecture

• Removed TIM Studio Apollo dependency and GraphQL layer
• Merged forecasting and anomaly detection microservices into one (job management) microservice
• 'Always running' workers are no longer managed by the autoscaler