OpenQuake Engine 2.4.0

[Michele Simionato (@micheles)]

• Now the command oq export loss_curves/rlz-XXX works both for the
  classical_risk calculator and the event_based_risk calculator

[Daniele Viganò (@daniviga)]

• Remove the default 30 day-old view limit in the WebUI calculation list

[Michele Simionato (@micheles)]

• Fixed a broken import affecting the command oq upgrade_nrml
• Made it possible to specify multiple file names in
  in the source_model_logic_tree file
• Reduced the data transfer in the object RlzsAssoc and improved the
  postprocessing of hazard curves when the option --hc is given
• Changed the ruptures.xml exporter to export unique ruptures
• Fixed a bug when downloading the outputs from the WebUI on Windows
• Made oq info --report fast again by removing the rupture fine filtering
• Improved the readibility of the CSV export dmg_total
• Removed the column eid from the CSV export ruptures; also
  renamed the field serial to rup_id and reordered the fields
• Changed the event loss table exporter: now it exports an additional
  column with the rup_id
• Changed scenario npz export to export also the GMFs outside the maximum
  distance
• Fixed scenario npz export when there is a single event
• Replaced the event tags with numeric event IDs
• The mean hazard curves are now generated by default
• Improved the help message of the command oq purge
• Added a @reader decorator to mark tasks reading directly from the
  file system
• Removed the .txt exporter for the GMFs, used internally in the tests
• Fixed a bug with relative costs which affected master for a long time,
  but not the release 2.3. The insured losses were wrong in that case.
• Added an .hdf5 exporter for the asset loss table
• Loss maps and aggregate losses are computed in parallel or sequentially
  depending if the calculation is a postprocessing calculation or not
• Deprecated the XML risk exporters
• Removed the .ext5 file
• Restored the parameter asset_loss_table in the event based calculators
• Added a full .hdf5 exporter for hcurves-rlzs
• Removed the individual_curves flag: now by default only the statistical
  hazard outputs are exported
• Saved a lot of memory in the computation of the hazard curves and stats
• Renamed the parameter all_losses to asset_loss_table
• Added an experimental version of the event based risk calculator which
  is able to use GMFs imported from an external file
• Added a max_curve functionality to compute the upper limit of the
  hazard curves amongst realizations
• Raised an error if the user specifies quantile_loss_curves
  or conditional_loss_poes in a classical_damage calculation
• Added a CSV exporter for the benefit-cost-ratio calculator
• The classical_risk calculator now reads directly the probability maps,
  not the hazard curves
• Turned the loss curves into on-demand outputs
  for the event based risk calculator
• The loss ratios are now stored in the datastore and not in an
  external .ext5 file
• The engine outputs are now streamed by the WebUI
• Used a temporary export directory in the tests, to avoid conflicts
  in multiuser situations
• Added an .npz exporter for the loss maps
• Raised an error early when using a complex logic tree in scenario
  calculations
• Changed the CSV exporter for the loss curves: now it exports all the
  curves for a given site for the classical_risk calculator
• Fixed the save_ruptures procedure when there are more than 256
  surfaces in the MultiSurface
• Renamed the csq_ outputs of the scenario_damage to losses_
• Changed the way scenario_damage are stored internally to be more
  consistent with the other calculators
• Removed the GSIM from the exported file name of the risk outputs
• New CSV exporter for GMFs generated by the event based calculator
• The event IDs are now unique and a constraint on the maximum
  number of source groups (65,536) has been added
• Added an output losses_by_event to the scenario_risk calculator
• Changed the output ruptures.csv to avoid duplications
• Added an output losses_by_taxon to the scenario_risk calculator
• Fixed a performance bug in get_gmfs: now the scenario risk and damage
  calculators are orders of magnitude faster for big arrays
• Added an export test for the event loss table in the case of multiple TRTs
• Removed the experimental rup_data output
• Added an .npz export for the output losses_by_asset
• Exported the scenario_risk aggregate losses in a nicer format

[Daniele Viganò (@daniviga)]

• The 'oq webui' command now works on a multi-user installation
• Splitted RPM packages into python-oq-engine (single node) and
  python-oq-engine-master/python-oq-engine-worker (multi-node)

[Paolo Tormene (@ptormene)]

• The 'Continue' button in the Web UI is now available also for risk
  calculations

[Michele Simionato (@micheles)]

• Fixed a Python 3 bug in the WebUI when continuing a calculation: the
  hazard_calculation_id was passed as a string and not as an integer
• Changed to rupture storage to use variable length-arrays, with a speedup
  of two orders of magnitude
• Avoided storing twice the rupture events
• Optimized the serialization of ruptures on HDF5 by using a sids output
• Changed the Web UI button from "Run Risk" to "Continue"
• The avg field in the loss curves is computed as the integral of the curve
  again, and it is not extracted from the avg_losses output anymore
• Made the fullreport exportable
• Fixed the rup_data export, since the boundary field was broken
• Restored the output losses_by_taxon in the event_based_risk calculator
• Fixed the calculator event based UCERF so that average losses can
  be stored

[Daniele Viganò (@daniviga)]

• Added a check to verify that an 'oq' client is talking to the
  right DbServer instance
• Introduced an optional argument for 'oq dbserver' command line
  to be able to override its default interface binding behaviour

[Michele Simionato (@micheles)]

• Optimized the event based calculators by reducing the number of calls
  to the GmfComputer and by using larger arrays
• Added a check on missing vulnerability functions for some loss type
  for some taxonomy
• Now we save the GMFs on the .ext5 file, not the datastore
• Fixed bug in event_based_risk: it was impossible to use vulnerability
  functions with "PM" distribution
• Fixed bug in event_based_risk: the ebrisk calculator is required as
  precalculator of event_based_risk, not others
• Fixed bug in scenario_risk: the output all_losses-rlzs was aggregated
  incorrectly
• Now the ucerf_risk calculators transfer only the events, not the ruptures,
  thus reducing the data transfer of several orders of magnitude
• Added a view get_available_gsims to the WebUI and fixed the API docs
• Introduced a configuration parameter max_site_model_distance with default
  of 5 km
• Implemented sampling in the UCERF event based hazard calculator

[Daniele Viganò (@daniviga)]

• Use threads instead of processes in DbServer because SQLite3
  isn't fork-safe on macOS Sierra

[Michele Simionato (@micheles)]

• Fixed a TypeError when deleting a calculation from the WebUI
• Extended the command oq to_hdf5 to manage source model files too
• Improved significantly the performance of the event based calculator
  when computing the GMFs and not the hazard curves
• Stored information about the mean ground motion in the datastore
• Saved the rupture mesh with 32 floats instead of 64 bit floats
• Raised the limit on the event IDs from 2^16 to 2^32 per task
• Fixed classical_risk: there was an error when computing the statistics
  in the case of multiple assets of the same taxonomy on the same site
• Changed the UCERF event based calculators to parallelize by SES
• Fixed a site model bug: when the sites are extracted from the site model
  there is no need to perform geospatial queries to get the parameters
• Added a command oq normalize to produce good sites.csv files
• Introduced a ses_seed parameter to specify the seed used to generate
  the stochastic event sets; random_seed is used for the sampling only
• Changed the build_rcurves procedure to read the loss ratios directly from
  the workers