Eval Commandedit

The eval-model command evaluates model prediction errors in a given time range, with a given model.

For example the command below will evaluate the model named avg_temp2 between yesterday and the current time:

loudml -e "eval-model --from now-1d --to now avg_temp2"

And this example will output result to stdout:

eval(test-model): 100%|███████████████████████████| 1/1 [00:01<00:00,  1.01s/it]
timestamp           @count_value         loudml.count_value
1569570000.0        15217.              15066.361
1569571200.0        14772.              14729.681
1569572400.0        14423.              14171.51
1569573600.0        226.                13703.53
...

The eval-model command supports the following options:

-f

(date) The from date in range query, supports Date Math notation.

-t

(date) The to date in range query, supports Date Math notation.

-s

Will save output data points to the bucket

-e

Calculate and print the mean square error

-v

Use the specified model version for evaluation. Use -v all with -e flag to evaluate and compare the error for all saved model versions

model_name

The model name