Releases: equinor/gordo
Release 0.14.0 of gordo-components
- Pretty Sphinx-generated html documentation, available with
make docs
. (#140) - Add
lstm_symmetric
andlstm_hourglass
as possible "kinds" toKerasLSTMAutoEncoder
(#159) - IROC reader: Ignore empty files (#171)
- Support
Troll A
in NCS-reader (#176) - Save and serve Keras model build history (loss/acc for all epochs) (#148)
- Make Watchman report status & logs of model builders / servers to allow debugging without kubectl (#164)
- Add ability to cache and re-use old models per project for faster model-deployments (#184)
- Move gordo-demo into gordo-components with entry-point
gordo-components client
(#165)
Release 0.13.0 of gordo-components
Release 0.12.1 of gordo-components
This is a bugfix release which fixes a performance degradation where tag-fetching started taking very long time (#149)
Release 0.12.0 of gordo-components
Version 0.12 of gordo-components contains the following user-observerable changes:
- Support ability to download model from server via route GET /download-model (#133). The model can be loaded using
gordo_components.serializer.loads
- Add cross validation to build_model process w/ metadata (#134). The default is to do a 3-fold CV with
explained-variance
as scoring. The values are available in the metadata under model.cross-validation.scores - Add /prediction GET route to gordo ml server (#100). Takes 'start' and 'end' query parameter timestamps and returns an array of dicts specifying details information about each request such as time range interval
named tags and their values, along with absolute abnormality score. Requires that access to an influx-database with raw values have been set up for the project. - Add IROC reader to DataLakeBackedDataset (#135). This allow gordo to transparently use IROC-data in the same way as NCS (Norwegian continental shelf)-data.
Release 0.11.0 of gordo-components
Introduces a concept of DataProvider
and Dataset
, where dataset uses data providers to get data. (2adcc3b)
Release 0.10.0 of gordo-components
- Rename feedforward_symetric to feedforward_model
- Add model constructor feedforward_hourglass
- Change default KerasAutoEncoder kind to hourglass
Release 0.9.0 of gordo-components
- No longer require
TRAIN_START/END_DATE
orTAGS
to be passed tocli.build
routine as environment variables, will come in asdata_config
(41de274)
Release 0.8.0 of gordo-components
-
/predictions
->/prediction
for gordo ml server (baf37bf) -
/metadata
's JSONversion
->gordo-server-version
(67d05e9) -
gordo_components.dataset._datasets
is now public, (1c7522d) -
gordo_components.dataset.get_dataset
is now private (a4e776d) -
gordo_components.builder.build_model
will returntuple(model, metadata)
and will not save, and thus nooutput_dir
parameter (00c0561) -
Add example notebooks (e8db53e)
Release 0.7.0 of gordo-components
-
Support dumps and loads for gordo serializer (7774f7a)
- Can dump a pipeline into bytes
serializer.dumps()
and loadserializer.loads()
- Can dump a pipeline into bytes
-
Update watchman to just use target names (43dac9b)
- Running watchman no longer requies a list of sanitized machine/target names. Only the project name
-
Pickup METADATA passed in from infrastructure (5927f4c)
- Allows users to specify
metadata
key mapping in each machine config
- Allows users to specify
-
Make Watchman return metadata from servers & report naked endpoints (18b0528)
endpoints
now reportsendpoint
as the base path to the server and not/metadata
as the endpoint.- Returns the metadata from the endpoints
metadata
route as well ashealthy
status
-
Update ambassador host to match namespace it's in (b3911d0)
- Ambassador was moved into its own namespace, fix for watchman contacting servers.
-
Add DataLakeBackedDataset as a possible data getter (227eade, 48f383b, 4385aba)
Release 0.6.1 of gordo-components
- Fix of Swagger UI when deployed on Kubernetes behind Ambassador (3e563e3)