[Michele Simionato (@micheles)]

• In some cases applyToSources was giving a fake error about the source
  not being in the source model even if it actually was

[Chris Van Houtte (@cvanhoutte)]

• Adds the Van Houtte et al. (2018) significant duration model for New
  Zealand

[Michele Simionato (@micheles)]

• Added a way to compute and plot the MFD coming from an event based
• Storing the MFDs in TOML format inside the datastore

[Robin Gee (@rcgee)]

• Moves b4 constant into COEFFS table for GMPE Sharma et al., 2009

[Graeme Weatherill (@g-weatherill)]

• Adds functionality to Cauzzi et al. (2014) and Derras et al. (2014)
  calibrated GMPEs for Germany to use either finite or point source distances

[Michele Simionato (@micheles)]

• Restored the ability to associate site model parameters to a grid of sites
• Made it possible to set hazard_curves_from_gmfs=true with
  ground_motion_fields=false in the event based hazard calculator
• Introduced a mechanism to split the tasks based on an estimated duration
• Integrated oq plot_memory into oq plot
• Removed NaN values for strike and dip when exporting griddedRuptures
• Fixed oq reset to work in multi-user mode
• Extended the source_id-filtering feature in the job.ini to multiple sources
• Supported WKT files for the binary perils in the multi_risk calculator
• Added an early check on the coefficients of variation and loss ratios of
  vulnerability functions with the Beta distribution
• Made sure that oq engine --dc removes the HDF5 cache file too
• Removed the flag optimize_same_id_sources because it is useless now
• Introduced a soft limit at 65,536 sites for event_based calculations
• Fixed a performance regression in ucerf_classical that was filtering
  before splitting, thus becoming extra-slow
• Improved the progress log, that was delayed for large classical calculations
• Exported the ruptures as 3D multi-polygons (instead of 2D ones)
• Changed the aggregate_by exports for consistency with the others
• Changed the losses_by_event exporter for ebrisk, to make it more
  consistent with scenario_risk and event_based_risk
• Changed the agglosses and losses_by_event exporters in scenario_risk,
  by adding a column with the realization index
• Changed the generation of the hazard statistics to consume very little
  memory
• Fixed a bug with concurrent_tasks being inherited from the parent
  calculation instead of using the standard default
• Removed the dependency from mock, since it is included in unittest.mock
• For scenario, replaced the branch_path with the GSIM representation in
  the realizations output
• Added a check for suspiciously large source geometries
• Deprecated the XML disaggregation exporters in favor of the CSV exporters
• Turned the disaggregation calculator into a classical post-calculator
  to use the precomputed distances and speedup the computation even more
• Fixed the disaggregation calculator by discarding the ruptures outside
  the integration distance
• Optimized the speed of the disaggregation calculator by moving a statistical
  functions outside of the inner loop
• Changed the file names of the exported disaggregation outputs
• Fixed an export agg_curves issue with pre-imported exposures
• Fixed an export agg_curves issue when the hazard statistics are different
  from the risk statistics
• Removed the disaggregation statistics: now the engine disaggregates only on
  a single realization (default: the closest to the mean)
• Forbidden disaggregation matrices with more than 1 million elements
• Reduced the data transfer when computing the hazard curves
• Optimized the reading of large CSV exposures
• Fixed the --hc functionality across users
• Optimized the reduction of the site collection on the exposure sites
• Made more robust the gsim logic tree parser: lines like
  <uncertaintyModel gmpe_table="../gm_tables/Woffshore_low_clC.hdf5">
  are accepted again
• Added a check against duplicated values in nodal plane distributions and
  hypocenter depth distributions
• Changed the support for zipped exposures and source models: now the
  name of the archive must be written explicitly in the job.ini
• Added support for numpy 1.16.3, scipy 1.3.0, h5py 2.9.0
• Removed the special case for event_based_risk running two calculations

[Graeme Weatherill (@g-weatherill)]

• Adds the Tromans et al. (2019) adjustable GMPE for application to PSHA
  in the UK

[Michele Simionato (@micheles)]

• Optimized src.sample_ruptures for (multi)point sources and are sources
• Fixed a mutability bug in the DistancesContext and made all context
  arrays read-only: the fix may affect calculations using the GMPEs
  berge_thierry_2003, cauzzi_faccioli_2008 and zhao_2006;
• Fixed a bug with the minimum_distance feature
• Fixed a bug in the exporter of the aggregate loss curves: now the loss
  ratios are computed correctly even in presence of occupants
• Removed the (long time deprecated) capability to read hazard curves and
  ground motion fields from XML files: you must use CSV files instead

[Marco Pagani (@mmpagani)]

• Implemented a modified GMPE that add between and within std to GMPEs only
  supporting total std

[Michele Simionato (@micheles)]

• Added the ability to use a taxonomy_mapping.csv file
• Fixed a bug in classical_damage from CSV: for hazard intensity measure
  levels different from the fragility levels, the engine was giving incorrect
  results
• Serialized also the source model logic tree inside the datastore
• Added a check on missing intensity_measure_types in event based
• Fixed oq prepare_site_model in the case of an empty datadir
• Added a comment line with useful metadata to the engine CSV outputs
• Removed the long time deprecated event loss table exporter for event based
  risk and enhanced the losses_by_event exporter to export the realization ID
• Removed the long time deprecated GMF XML exporter for scenario
• IMT-dependent weights in the gsim logic tree can be zero, to discard
  contributions outside the range of validity of (some of the) GSIMs
• Now it is possible to export individual hazard curves from an event
• Added a view gmvs_to_hazard

[Daniele Viganò (@daniviga)]

• Fixed packaging issue, the .hdf5 tables for Canada were missing

[Michele Simionato (@micheles)]

• Fixed regression in the gsim logic tree parser for the case
  of .hdf5 tables

[Michele Simionato (@micheles)]

• Added a rlzi column to to sig_eps.csv output
• Accepted GMF CSV files without a rlzi column
• Accepted a list-like syntax like return_periods=[30, 60, 120, 240, 480]
  in the job.ini, as written in the manual
• Fixed a bug in the asset_risk exporter for uppercase tags

[Paul Henshaw (@pslh)]

• Fixed an encoding bug while reading XML files on Windows

[Michele Simionato (@micheles)]

• Added a view gmvs_to_hazard

[Giovanni Lanzano (@giovannilanzanoINGV)]

• Lanzano and Luzi (2019) GMPE for volcanic zones in Italy

[Michele Simionato (@micheles)]

• Now it is possible to export individual hazard curves from an event
  based calculation by setting hazard_curves_from_gmfs = true and
  `individual_curves = true (before only the statistics were saved)

[Graeme Weatherill (@g-weatherill)]

• Adds adaptation of Abrahamson et al. (2016) 'BC Hydro' GMPEs calibrated
  to Mediterranean data and with epistemic adjustment factors

[Chris Van Houtte (@cvanhoutte)]

• Added new class to bradley_2013b.py for hazard maps
• Modified test case_37 to test multiple sites

[Marco Pagani (@mmpagani)]

• Fixed a bug in the logic tree parser and added a check to forbid logic
  trees with applyToSources without applyToBranches, unless there is a
  single source model branch

[Michele Simionato (@micheles)]

• Removed the experimental parameter prefilter_sources

[Daniele Viganò (@daniviga)]

• Multiple DbServer ZMQ connections are restored to avoid errors under heavy
  load and/or on slower machines

[Michele Simionato (@micheles)]

• Removed the ugly registration of custom signals at import time: now they
  are registered only if engine.run_calc is called
• Removed the dependency from rtree
• Removed all calls to ProcessPool.shutdown to speed up the tests and to
  avoid non-deterministic errors in atexit._run_exitfuncs

[Marco Pagani (@mmpagani)]

• Added tabular GMPEs as provided by Michal Kolaj, Natural Resources Canada

[Michele Simionato (@micheles)]

• Extended the ebrisk calculator to support coefficients of variations

[Graeme Weatherill (@g-weatherill)]

• Adds Kotha et al (2019) shallow crustal GMPE for SERA
• Adds 'ExperimentalWarning' to possible GMPE warnings
• Adds kwargs to check_gsim function

[Michele Simionato (@micheles)]

• Fixed problems like SA(0.7) != SA(0.70) in iml_disagg
• Exposed the outputs of the classical calculation in event based
  calculations with compare_with_classical=true
• Made it possible to serialize together all kind of risk functions,
  including consequence functions that before were not HDF5-serializable
• Fixed a MemoryError when counting the number of bytes stored in large
  HDF5 datasets
• Extended asset_hazard_distance to a dictionary for usage with multi_risk
• Extended oq prepare_site_model to work with sites.csv files
• Optimized the validation of the source model logic tree: now checking
  the sources IDs is 5x faster
• Went back to the old logic in sampling: the weights are used for the
  sampling and the statistics are computed with identical weights
• Avoided to transfer the epsilons by storing them in the cache file
  and changed the event to epsilons associations
• Reduced the data transfer in the computation of the hazard curves, causing
  in some time huge speedups (over 100x)
• Implemented a flag modal_damage_state to display only the most likely
  damage state in the output dmg_by_asset of scenario damage calculations
• Reduced substantially the memory occupation in classical calculations
  by including the prefiltering phase in the calculation phase

[Daniele Viganò (@daniviga)]

• Added a 'serialize_jobs' setting to the openquake.cfg
  which limits the maximum number of jobs that can be run in parallel

[Michele Simionato (@micheles)]

• Fixed two exporters for the ebrisk calculator (agg_curves-stats and
  losses_by_event)
• Fixed two subtle bugs when reading site_model.csv files
• Added /extract/exposure_metadata and /extract/asset_risk
• Introduced an experimental multi_risk calculator for volcanic risk

[Guillaume Daniel (@guyomd)]

• Updating of Berge-Thierry (2003) GSIM and addition of several alternatives
  for use with Mw

[Michele Simionato (@micheles)]

• Changed the classical_risk calculator to use the same loss ratios for all
  taxonomies and then optimized all risk calculators
• Temporarily removed the insured_losses functionality
• Extended oq restore to download from URLs
• Removed the column 'gsims' from the output 'realizations'
• Better parallelized the source splitting in classical calculations
• Added a check for missing hazard in scenario_risk/scenario_damage
• Improved the GsimLogicTree parser to get the line number information, a
  feature that was lost with the passage to Python 3.5
• Added a check against mispellings in the loss type in the risk keys
• Changed the aggregation WebAPI from
  aggregate_by/taxonomy,occupancy/avg_losses?kind=mean&loss_type=structural to
  aggregate/avg_losses?kind=mean&loss_type=structural&tag=taxonomy&tag=occupancy
• Do not export the stddevs in scenario_damage in the case of 1 event
• Fixed export bug for GMFs imported from a file
• Fixed an encoding error when storing a GMPETable
• Fixed an error while exporting the hazard curves generated by a GMPETable
• Removed the deprecated feature aggregate_by/curves_by_tag

[Michele Simionato (@micheles)]

• Compatibility with 'decorator' version >= 4.2

[Giovanni Lanzano (@giovannilanzanoINGV)]

• Contributed a GMPE SkarlatoudisEtAlSSlab2013

[Michele Simionato (@micheles)]

• Changed the event loss table exporter to export also rup_id and year
• Extended the ebrisk calculator to compute loss curves and maps

[Rodolfo Puglia (@rodolfopuglia)]

• Spectral acceleration amplitudes at 2.5, 2.75 and 4 seconds added

[Marco Pagani (@mmpagani)]

• Improved the event based calculator to account for cluster-based models

[Michele Simionato (@micheles)]

• Removed the now redundant command oq extract hazard/rlzs

[Daniele Viganò (@daniviga)]

• Fixed 'oq abort' and always mark killed jobs as 'aborted'

[Michele Simionato (@micheles)]

• Made it possible to use in the Starmap tasks without a monitor argument
• Stored the sigma and epsilon parameters for each event in event based
  and scenario calculations and extended the gmf_data exporter consequently
• Fixed the realizations CSV exporter which was truncating the names of the
  GSIMs
• Deprecated the XML exporters for hcurves, hmaps, uhs
• Introduced a sap.script decorator
• Used the WebExtractor in oq importcalc
• Restored validation of the source_model_logic_tree.xml file
• Raised an early error for missing occupants in the exposure
• Added a check to forbid duplicate file names in the uncertaintyModel tag
• Made it possible to store the asset loss table in the ebrisk calculator
  by specifying asset_loss_table=true in the job.ini
• Added a flag oq info --parameters to show the job.ini parameters
• Removed the source_name column from the disagg by source output

[Rao Anirudh]

• Fixed wrong investigation_time in the calculation of loss maps from
  loss curves

[Robin Gee (@rcgee)]

• Added capability to optionally specify a time_cutoff parameter to
  declustering time window

[Michele Simionato (@micheles)]

• Merged the commands oq plot_hmaps and oq plot_uhs inside oq plot
• Changed the storage of hazard curves and hazard maps to make it consistent
  with the risk outputs and Extractor-friendly

[Chris Van Houtte (@cvanhoutte)]

• Added necessary gsims to run the Canterbury Seismic Hazard Model
  in Gerstenberger et al. (2014)
• Added a new gsim file mcverry_2006_chch.py to have the Canterbury-
  specific classes.
• Added a new gsim file bradley_2013b.py to implement the
  Christchurch-specific modifications to the Bradley2013 base model.

[Michele Simionato (@micheles)]

• Added a check on the intensity measure types and levels in the job.ini,
  to make sure they are ordered by period
• Reduced the number of client sockets to the DbServer that was causing
  (sporadically) the hanging of calculations on Windows
• Extended the WebAPI to be able to extract specific hazard curves, maps
  and UHS (i.e. IMT-specific and site specific)
• Removed the realization index from the event loss table export, since
  is it redundant
• Forced all lowercase Python files in the engine codebase
• Removed the dependency from nose

[Robin Gee (@rcgee)]

• Updated GMPE of Yu et al. (2013)

[Michele Simionato (@micheles)]

• Added an Extractor client class leveraging the WebAPI and enhanced
  oq plot_hmaps to display remote hazard maps
• Added a check when disaggregation is attempted on a source model
  with atomic source groups
• Implemented serialization/deserialization of GSIM instances to TOML
• Added a check against mispelled rupture distance names and fixed
  the drouet_alpes_2015 GSIMs
• Changed the XML syntax used to define dictionaries IMT -> GSIM
• Now GSIM classes have an .init() method to manage notrivial
  initializations, i.e. expensive initializations or initializations
  requiring access to the filesystem
• Fixed a bug in event based that made it impossible to use GMPETables
• Associated the events to the realizations even in scenario_risk: this
  involved changing the generation of the epsilons in the case of asset
  correlation. Now there is a single aggregate losses output for all
  realizations
• Removed the rlzi column from the GMF CSV export
• Introduced a new parameter ebrisk_maxweight in the job.ini
• For classical calculations with few sites, store information about the
  realization closest to the mean hazard curve for each site
• Removed the max_num_sites limit on the event based calculator

[Valerio Poggi (@klunk386)]

• Added an AvgSA intensity measure type and a GenericGmpeAvgSA which is
  able to use it

[Michele Simionato (@micheles)]

• Introduced the ability to launch subtasks from tasks
• Stored rupture information in classical calculations with few sites

[Chris Van Houtte (@cvanhoutte)]

• Adding conversion from geometric mean to larger horizontal component in
  bradley_2013.py

[Michele Simionato (@micheles)]

• Fixed a bug in applyToSources for the case of multiple sources
• Moved the prefiltering on the workers to save memory
• Exported the aggregated loss ratios in avg losses and agg losses
• Removed the variables quantile_loss_curves and mean_loss_curves: they
  were duplicating quantile_hazard_curves and mean_hazard_curves
• Only ruptures boundingbox-close to the site collection are stored

[Marco Pagani (@mmpagani)]

• Added cluster model to classical PSHA calculator

[Michele Simionato (@micheles)]

• Fixed a bug in scenario_damage from ShakeMap with noDamageLimit=0
• Avoided the MemoryError in the controller node by speeding up the saving
  of the information about the sources
• Turned utils/reduce_sm into a proper command
• Fixed a wrong coefficient in the ShakeMap amplification
• Fixed a bug in the hazard curves export (the filename did not contain
  the period of the IMT thus producing duplicated files)
• Parallelized the reading of the exposure

[Marco Pagani (@mmpagani)]

• Fixed the implementation on mutex ruptures

[Michele Simionato (@micheles)]

• Changed the aggregated loss curves exporter
• Added an experimental calculator ebrisk
• Changed the ordering of the events (akin to a change of seed in the
  asset correlation)

[Robin Gee (@rcgee)]

• Fixed bug in tusa_langer_2016.py BA08SE model - authors updated b2 coeff
• Fixed bug in tusa_langer_2016.py related to coeffs affecting Repi models

[Michele Simionato (@micheles)]

• Added a check to forbid to set ses_per_logic_tree_path = 0
• Added an API /extract/event_info/eidx
• Splitting the sources in classical calculators and not in event based
• Removed max_site_model_distance
• Extended the logic used in event_based_risk - read the hazard sites
  from the site model, not from the exposure - to all calculators
• In classical_bcr calculations with a CSV exposure the retrofitted field
  was not read. Now a missing retrofitted value is an error

[Robin Gee (@rcgee)]

• Fixed bug in tusa_langer_2016.py BA08SE model - authors updated b2 coeff

[Michele Simionato (@micheles)]

• Fixed a bug in scenario_damage from ShakeMap with noDamageLimit=0
• Avoided the MemoryError in the controller node by speeding up the saving
  of the information about the sources
• Fixed a wrong coefficient in the ShakeMap amplification
• Fixed a bug in the hazard curves export (the filename did not contain
  the period of the IMT thus producing duplicated files)

[Michele Simionato (@micheles)]

• Fixed the GMF exporter to export the event IDs and not event indices

[Robin Gee (@rcgee)]

• Fixed bug in tusa_langer_2016.py related to coeffs affecting Repi models

[Graeme Weatherill (@g-weatherill)]

• Adds GMPE suite for national PSHA for Germany

[Daniele Viganò (@daniviga)]

• Added a warning box when an unsupported browser is used to view the WebUI
• Updated Docker containers to support a multi-node deployment
  with a shared directory
• Moved the Docker containers source code from oq-builders
• Updated the documentation related to the shared directory
  which is now mandatory for multi-node deployments

[Matteo Nastasi (@nastasi-oq)]

• Removed tests folders

[Stéphane Drouet (@stephane-on)]

• Added Drouet & Cotton (2015) GMPE including 2017 erratum

[Michele Simionato (@micheles)]

• Optimized the memory occupation in classical calculations (Context.poe_map)
• Fixed a wrong counting of the ruptures in split fault sources with
  an hypo_list/slip_list causing the calculation to fail
• Made the export of uniform hazard spectra fast
• Made the std hazard output properly exportable
• Replaced the ~ in the header of the UHS csv files with a -
• Restored the individual_curves flag even for the hazard curves
• Implemented dGMPE weights per intensity measure type
• Extended --reuse-hazard to all calculators
• Fixed a bug in event_based_risk from GMFs with coefficients of variations

[Graeme Weatherill (@g-weatherill)]

• Adds magnitude scaling relation for Germany

[Michele Simionato (@micheles)]

• Used floats for the the GSIM realization weights, not Python Decimals
• Added a flag fast_sampling, by default False
• Added an API /extract/src_loss_table/<loss_type>
• Removed the rupture filtering from sample_ruptures and optimized it in
  the RuptureGetter by making use of the bounding box
• Raised the limit on ses_per_logic_tree_path from 216 to 232;
• Added a parameter max_num_sites to increase the number of sites accepted
  by an event based calculation up to 2 ** 32 (the default is still 2 ** 16)
• Added a command oq compare to compare hazard curves and maps within
  calculations
• Extended the engine to read transparently zipped source models and exposures
• Restored the check for invalid source IDs in applyToSources
• Extended the command oq zip to zip source models and exposures
• Parallelized the associations event ID -> realization ID
• Improved the message when assets are discarded in scenario calculations
• Implemented aggregation by multiple tags, plus a special case for the
  country code in event based risk

[Marco Pagani (@mmpagani)]

• Added two modified versions of the Bindi et al. (2011) to be used in a
  backbone approach to compute hazard in Italy
• Added a modified version of Berge-Thierry et al. 2003 supporting Mw

[Michele Simionato (@micheles)]

• Changed the way loss curves and loss maps are stored in order to unify
  the aggregation logic with the one used for the average losses
• Now it is possible to compute the ruptures without specifying the sites
• Added an early check for the case of missing intensity measure types
• Deprecated the case of exposure, site model and region_grid_spacing all
  set at the same time
• Implemented multi-exposure functionality in event based risk
• Changed the event based calculator to store the ruptures incrementally
  without keeping them all in memory
• Refactored the UCERF event based calculator to work as much as possible
  the regular calculator
• Optimized the management and storage of the aggregate losses in the event
  based risk calculation; also, reduced the memory consumption
• Changed the default for individual_curves to "false", which is the right
  default for large calculations
• Optimized the saving of the events
• Removed the save_ruptures flag in the job.ini since ruptures must be saved
  always
• Optimized the rupture generation in case of sampling and changed the
  algorithm and seeds
• Fixed a bug with the IMT SA(1) considered different from SA(1.0)
• Removed the long-time deprecated GMF exporter in XML format for event_based
• Added a re-use hazard feature in event_based_risk in single-file mode
• Made the event ID unique also in scenario calculations with
  multiple realizations
• Removed the annoying hidden .zip archives littering the export directory
• Added an easy way to read the exposure header
• Added a way to run Python scripts using the engine libraries via oq shell
• Improved the minimum_magnitude feature
• Fixed the check on missing hazard IMTs
• Reduced substantially the memory occupation in event based risk
• Added the option spatial_correlation=no correlation for risk calculations
  from ShakeMaps
• Removed the experimental calculator ucerf_risk
• Optimized the sampling of time-independent sources for the case of
  prefilter_sources=no
• Changed the algorithm associating events to SESs and made the event based
  hazard calculator faster in the case of many SESs
• Reduced substantially the memory consumption in event based risk
• Made it possible to read multiple site model files in the same calculation
• Implemented a smart single job.ini file mode for event based risk
• Now warnings for invalid parameters are logged in the database too
• Fixed oq export avg_losses-stats for the case of one realization
• Added oq export losses_by_tag and oq export curves_by_tag
• Extended oq export to work in a multi-user situation
• Forbidden event based calculations with more than max_potential_paths
  in the case of full enumeration
• Saved a large amount of memory in event_based_risk calculations
• Added a command oq export losses_by_tag/<tagname> <calc_id>
• Extended oq zip to zip the risk files together with the hazard files
• Changed the building convention for the event IDs and made them unique
  in the event loss table, even in the case of full enumeration
• Optimized the splitting of complex fault sources
• Fixed the ShakeMap download procedure for uncertainty.zip archives
  with an incorrect structure (for instance for ci3031111)
• Disabled the spatial correlation in risk-from-ShakeMap by default
• Optimized the rupture sampling where there is a large number of SESs
• Extended the reqv feature to multiple tectonic region types and
  removed the spinning/floating for the TRTs using the feature
• Reduced the GMPE logic tree upfront for TRTs missing in the source model
• Fixed the ShakeMap downloader to use the USGS GeoJSON feed
• Improved the error message when there are more than 65536 distinct tags
  in the exposure
• Turned vs30measured into an optional parameter

[Chris Van Houtte (@cvanhoutte)]

• Added siteclass as a site parameter, and reference_site_class as
  a site parameter than can be specified by the user in the ini file
• Added new classes to mcverry_2006.py to take siteclass as a predictor
• Updated comments in mcverry_2006.py
• Added new mcverry_2006 test tables to account for difference in site
  parameter
• Added qa_test_data classical case_32

[Michele Simionato (@micheles)]

• Fixed the rupture exporter for Canada
• Extended the oq prepare_site_model to optionally generate the
  fields z1pt0, z2pt5 and vs30measured
• It is now an error to specify both the sites and the site model in the
  job.ini, to avoid confusion with the precedency
• Implemented a reader for site models in CSV format
• Made the export_dir relative to the input directory
• Better error message for ShakeMaps with zero stddev
• Added a source_id-filtering feature in the job.ini
• Added a check on non-homogeneous tectonic region types in a source group
• Fixed the option oq engine --config-file that broke a few releases ago
• Replaced nodal_dist_collapsing_distance and
  hypo_dist_collapsing_distance with pointsource_distance and made
  use of them in the classical and event based calculators

[Graeme Weatherill (@g-weatherill)]

• Fixes to hmtk completeness tables for consistent rates and addition of
  more special methods to catalogue

[Michele Simionato (@micheles)]

• Restricted ChiouYoungs2008SWISS01 to StdDev.TOTAL to avoid a bug
  when computing the GMFs with inter/intra stddevs
• Raised an error if assets are discarded because too far from the hazard
  sites (before it was just a warning)
• Added an attribute .srcidx to every event based rupture and stored it
• Fixed an issue with the Byte Order Mark (BOM) for CSV exposures prepared
  with Microsoft Excel
• Reduced the site collection instead of just filtering it; this fixes
  a source filtering bug and changes the numbers in case of GMF-correlation
• Added a command oq prepare_site_model to prepare a sites.csv file
  containing the vs30 and changed the engine to use it
• Added a cutoff when storing a PoE=1 from a CSV file, thus avoiding NaNs
  in classical_damage calculations
• Reduced the data transfer in the risk model by only considering the
  taxonomies relevant for the exposure
• Extended oq engine --run to accept a list of files
• Optimized the saving of the risk results in event based in the case of
  many sites and changed the command oq show portfolio_loss to show
  mean and standard deviation of the portfolio loss for each loss type

[Marco Pagani (@mmpagani)]

• Added a first and preliminary version of the GMM for the Canada model
  represented in an analytical form.
• Added a modified version of Atkinson and Macias to be used for the
  calculation of hazard in NSHMP2014.
• Added support for PGA to the Si and Midorikawa (1999).

[Michele Simionato (@micheles)]

• Made it possible to run the risk over an hazard calculation of another user
• Worked around the OverflowError: cannot serialize a bytes object larger
  than 4 GiB in event based calculations
• Started using Python 3.6 features
• Fixed the check on vulnerability function ID uniqueness for NRML 0.5
• Ruptures and GMFs are now computed concurrently, thus mitigating the
  issue of slow tas...

[Matteo Nastasi (@nastasi-oq)]

• specified 'amd64' as the only architecture supported by ubuntu packages

[Michele Simionato (@micheles)]

• Changed the source writer: now the srcs_weights are written in the XML
  file only if they are nontrivial
• Changed the algorithm assigning the seeds: they are now generated before
  the source splitting; also, a seed-related bug in the splitting was fixed
• For event based, moved the rupture generation in the prefiltering phase

[Daniele Viganò (@daniviga)]

• Fixed a bug with CTRL-C when using the processpool distribution

[Robin Gee (@rcgee)]

• Raised the source ID length limit in the validation from 60 to 75 characters
  to allow sources with longer IDs

[Michele Simionato (@micheles)]

• Introduced a multi_node flag in openquake.cfg and used it to
  fully parallelize the prefiltering in a cluster
• Used the rupture seed as rupture ID in event based calculations
• Changed the deprecation mechanism of GSIMs to use a class attribute
  superseded_by=NewGsimClass
• Solved the pickling bug in event based hazard by using generator tasks
• Improved the distribution of the risk tasks by changing the weight

[Pablo Heresi (@pheresi)]

• Contributed the HM2018CorrelationModel

[Michele Simionato (@micheles)]

• Restored the individual_curves flag that for the moment is used for the
  risk curves
• Introduced two experimental new parameters floating_distance and
  spinning_distance to reduce hypocenter distributions and nodal plane
  distributions of ruptures over the corresponding distances
• Optimized the parsing of the logic tree when there is no "applyToSources"
• Made the IMT classes extensible in client code
• Reduced the hazard maps from 64 to 32 bit, to be consistent with the
  hazard curves and to reduce by half the download time

[Graeme Weatherill (@g-weatherill)]

• Implements a fix of Montalva et al (2016) for new coefficients (now
  Montalva et al. (2017))

[Michele Simionato (@micheles)]

• Parallelized the reading of the source models
• Optimized oq info --report by not splitting the sources in that case
• Speedup the download of the hazard curves, maps and uhs
• Honored concurrent_tasks in the prefiltering phase too
• It is now legal to compute uniform hazard spectra for a single period
• Added command oq plot_memory
• Introduced a MultiGMPE concept
• Saved the size of the datastore in the database and used it in the WebUI

[Graeme Weatherill (@g-weatherill)]

• Adds geotechnical related IMTs

[Michele Simionato (@micheles)]

• Renamed /extract/agglosses -> /extract/agg_losses and same for aggdamages
• Supported equivalent epicentral distance with a reqv_hdf5 file
• Fixed the risk from ShakeMap feature in the case of missing IMTs
• Changed the way gmf_data/indices and ruptures are stored
• Added experimental support for dask
• Added 11 new site parameters for geotechnic hazard
• Changed the SiteCollection to store only the parameters required by the
  GSIMs

[Robin Gee (@rcgee)]

• The number of sites is now an argument in the method _get_stddevs()
  in the GMPE of Kanno, 2006

[Michele Simionato (@micheles)]

• Changed the serialization of ruptures to HDF5: the geometries are now
  stored in a different dataset
• Bug fix: the asset->site association was performed even when not needed
• Made it possible to serialize to .hdf5 multipoint sources and
  nonparametric gridded sources
• Added a check on source model logic tree files: the uncertaintyModel
  values cannot be repeated in the same branchset
• Added a flag std_hazard_curves; by setting it to true the user can
  compute the standard deviation of the hazard curves across realizations

[Marco Pagani (@mmpagani)]

• Added Thingbaijam et al. (2017) magnitude-scaling relationship

[Michele Simionato (@micheles)]

• Added an /extract/ API for event_based_mfd
• Fixed a bug in the classical_damage calculators: multiple loss types
  were not treated correctly

[Marco Pagani (@mmpagani)]

• Adding tests to the method computing decimal time

[Michele Simionato (@micheles)]

• Removed the event_based_rupture calculator and three others
• Added a field size_mb to the output table in the database and made
  it visible in the WebUI as a tooltip
• Added a command oq check_input job.ini to check the input files
• Made the loss curves and maps outputs from an event based risk calculation
  visible to the engine and the WebUI (only the stats)
• Added a check on duplicated branchIDs in GMPE logic trees

[Daniele Viganò (@daniviga)]

• Fixed a bug when reading exposure with utf8 names on systems with non-utf8
  terminals (Windows)
• Changed the openquake.cfg file and added a dbserver.listen parameter
• Added the hostname in the WebUI page. It can be customize by the user
  via the local_settings.py file

[Michele Simionato (@micheles)]

• Added a Content-Length to the outputs downloadable from the WebUI
• Fixed a bug when extracting gmf_data from a hazard calculation with a
  filtered site collection
• Stored an attributed events.max_gmf_size
• Added a check on exposures with missing loss types
• Added a LargeExposureGrid error to protect the user by tricky exposures
  (i.e. France with assets in the Antilles)
• Changed the event_based_risk calculator to compute the loss curves and
  maps directly; removed the asset_loss_table
• Changed the event_based_risk calculator to distribute by GMFs always
• Optimized the memory consumption in the UCERF classical calculator
• Added a parameter minimum_magnitude in the job.ini
• Added an utility utils/combine_mean_curves.py

[Marco Pagani (@mmpagani) and Changlong Li (@mstlgzfdh)]

• Added a version of the Yu et al. (2013) GMPE supporting Mw

[Michele Simionato (@micheles)]

• Reduced the data transfer in the UCERF calculators
• Stored the zipped input files in the datastore for reproducibility
• Fixed a regression when reading GMFs from an XML in absence of a sites.csv
  file

[Robin Gee (@rcgee)]

• Extend oq to_shapefile method to also work with YoungsCoppersmithMFD
  and arbitraryMFD MFD typologies.

[Michele Simionato (@micheles)]

• Now the hazard statistics can be computed efficiently even in a single
  calculation, i.e. without the --hc option
• Added a check on the Python version in the oq command
• Reduced the data transfer when sending the site collection
• Changed the default filter_distance

[Daniele Viganò (@daniviga)]

• Fixed a bug where the PID was not saved into the database
  when using the command line interface
• Made it impossible to fire multiple CTRL-C in sequence
  to allow processes teardown and tasks revocation when Celery is used

[Michele Simionato (@micheles)]

• Used scipy.spatial.distance.cdist in Mesh.get_min_distance
• Prefiltered sites and assets in scenario calculations
• Made it possible to specify the filter_distance in the job.ini
• Made rtree optional again and disabled it in macOS
• Optimized the SiteCollection class and doubled the speed of distance
  calculations in most continental scale calculations
• Fixed an ordering bug in event based risk from GMFs when using a
  vulnerability function with PMF
• Replaced Rtree with KDtree except in the source filtering
• Parallelized the source prefiltering
• Removed the tiling feature from the classical calculator
• Undeprecated hazardlib.calc.stochastic.stochastic_event_set and
  made its signature right
• Removed the source typology from the ruptures and reduced the rupture
  hierarchy
• Removed the mesh spacing from PlanarSurfaces
• Optimized the instantiation of the rtree index
• Replaced the old prefiltering mechanism with the new one

[Daniele Viganò (@daniviga)]

• Managed the case of a dead controlling terminal (SIGHUP)

[Michele Simionato (@micheles)]

• Removed Decimal numbers from the PMF distribution in hazardlib
• Fixed another tricky bug with rtree filtering across the international
  date line
• Added a parameter prefilter_sources with values rtree|numpy|no
• Removed the prefiltering on the workers, resulting in a huge speedup
  for gridded ruptures at the cost of a larger data transfer
• Changed the losses_by_event output to export a single .csv file with
  all realizations
• Added a cross_correlation parameter used when working with shakemaps
• Now sites and exposure can be set at the same time in the job.ini
• Introduced a preclassical calculator
• Extended the scenario_damage calculator to export dmg_by_event
  outputs as well as losses_by_event outputs if there is a consequence
  model
• Unified region and region_constraint parameters in the job.ini
• Added a check to forbid duplicated GSIMs in the logic tree
• Introduced some changes to the realizations exporter (renamed field
  uid -> branch_path and removed the model field)
• Added a command oq celery inspect
• Reduced the check on too many realizations to a warning, except for
  event based calculations
• Improved the hazard exporter to exports only data for the filtered
  site collection and not the full site collection
• Extended the BCR exporter to export the asset tags

[Catalina Yepes (@CatalinaYepes)]

• Revised/enhanced the risk demos

[Michele Simionato (@micheles)]

• Added a warning about the option optimize_same_id_sources when the user
  should take advantage of it

[Daniele Viganò (@daniviga)]

• celery-status script converted into oq celery status command
• Removed Django < 1.10 backward compatibility
• Updated Python dependices (numpy 1.14, scipy 1.0.1,
  Django 1.10+, Celery 4+)

[Michele Simionato (@micheles)]

• Implemented scenario_risk/scenario_damage from shakemap calculators
• Exported the asset tags in the asset based risk outputs
• Fixed a numeric issue for nonparametric sources causing the hazard curves
  to saturate at high intensities
• Added an utility to download shakemaps
• Added an XML exporter for the site model
• Slight change to the correlation module to fix a bug in the SMTK
• Added a distribution mechanism threadpool