0.37.0

• Add optional base_path argument to NCS reader 📖 (#553)
• Support no resampling in TimeSeriesDataset 🕐 (#539)
• Workflow: Add start ml-server after ensure-single-workflow Ⓜ️ (#572)
• Add ml-server as dependency for model-building 🤞 (#572)
• Make client have Mutex for setting endpoints 💻 (#519)
• Fix grafana metrics query ❓ (#582)
• Add istio virtualservice to workflow 🥅 (#563)
• Add more logging to server 🌳 (#576)
• Version bumps 🇻🇮
  • Bump pytest from 5.2.1 to 5.2.2 (#584)
  • Bump azure-datalake-store from 0.0.47 to 0.0.48 (#585)
  • Bump nbconvert from 5.6.0 to 5.6.1 (#583)
  • Bump pytest-mock from 1.11.1 to 1.11.2 🐦 (#575)

0.37.0-rc2

User-facing changes:

• Support row filter buffering (#542)
• Url encode requests to datalake (#562)
• Increase default client batch size from 1000 to 100000 (#545)

NB: #562 implies an important change in how you write the tag-names for Equinor tags. If the tag-name contain a slash, as in e.g. ABC-f2612/garPaD/DAMN then you previosly wrote the tag as ABC-f2612%2FgarPaD&2FDAMN, which is how the tag is listed in the datalake. You shall now write the tag with the actual tag-name (i.e. ABC-f2612/garPaD/DAMN)

Other changes:

• Allow parametrization of image sources in workflow (#548)
• Add and fix Client component docs (#551)
• Add datalake path for 1230-Visund
• Update gordo server to run with Gunicorn (#536)
• Set clusterIP: None for gordo-notification-svc (#566)
• Workflow: remove influx env var mounting (#568)

0.37.0-rc1

🎉 Move workflow generator and related parts (#534) 🎉
Fix exponential time test (#543)
Fix CLI connection to watchman-to-sql command (#546)

0.34.0

Bug fixes:

• Disable CircleCI Docker layer caching (#528)
• Fix threading error on server with TensorFlow models (#524)

Features:

• Remove get_model from model.py to avoid uneccesary import of tensorflow (#518)
• Condense LSTMAutoEncoder and LSTMForecast (#510)
• Add asset-to-path for IOC assets (#470)

0.33.0

• Remove GET request support from ML Server (#498)
• Upgrade to Tensorflow 2.0, drop Keras dependency. (#500)
• Various requirement updates (#515, #509, #517, #511, #505, #508, #503, #504, #499, #494, #489, #493)

0.32.0-rc3

• Optionally (default=True) use parquet for data transfer between client and server (#472)
• Fix issue where models with functions as default parameters did not load properly (#477)

0.32.0-rc2

0.32.0-rc2 fixes a minor issue regarding parsing of gordo-components own version number.

0.32.0-rc1

• Change sleep time from xor to exponentiation (#465)
• Builder: always save model to output_dir (#461)
• Change retries in client to have exponential sleep time (#464)
• Expose evaluation from config in model builder (#458)
• Make ML Server serve multiple models (#460)

0.31.0

• Update requirements (#440)
• Make autoencoder API parameters match and add some input validation (#431)
• Expose optimizer and optimizer_kwargs in feedforward_model factory (#429)
• Change interactive logic (#428)
• Cli dockstring fix (#439)
• Add model offset to metadata (#444)
• Model builder - Support a pure CV/HP mode (#395)
• Switch to CircleCI and push to Dockerhub (#442, #451, #450, #454)
• Add EV, R2, MSE and MAE to metadata (#377)
• Client adjust for offset (#445)
• Add more explicit docs about resolution strings (#457)

0.30.0

• Add simple dataframe serialization ability to server serializer utils (0b2b84d)
• Remove logic that y must be a subset of X for anomaly detection. (6d75c37)