1.4.3 Release Notesedit

The changes listed below have been released for the first time in Loud ML 1.4.3.

Note

Please raise issues on Github if you experience compatibility issues on your system.

Data sinkedit

Time series prediction can now be saved into a specific data sink (issue #58): - the name of the datasink can be passed to ._predict endpoint with the parameter datasink - the name the default data sink can be defined in model settings (default_datasink)

Model descriptionedit

The . character is now allowed in feature fields and measurements (issue #40).

Two formats are supported to define model features. The first (now deprecated, but still supported):

"features": {
    "io": [
       {... feature ...},
       {... feature ...},
       ...
    ],
    "i": [...],
    "o": [...]
}

and the second, new, format:

"features": [
   {... feature ...},
   {... feature ...},
   ...
]

In this new format, the input/output information is given by the parameter io inside the feature.

  • i = input
  • o = output
  • io = input/output

Example of feature:

{
    "name": "myfeature",
    "io": "i"
    ...
}

InfluxDBedit

  • Retention policy can now be configured in InfluxDB datasource settings.
  • Add create_database to InfluxDB datasource settings for enabling/disabling database creation (default = true).

Miscellaneousedit

Multiprocessing optimizations.

Bug fixesedit

Loud ML now serializes null instead of nan in the mse field of the prediction result. (issue #52).

Loud ML now avoids crashes when attempting to activate a model that is in the middle of training. (issue #53).