HyperparameterOptimization is None

[webbug2005]

df_conc.head(5)
index | raw
2021-07-07 09:00:01 | 0.48
2021-07-07 09:30:06 | 0.46
2021-07-07 10:00:01 | 1.39
2021-07-07 10:30:06 | 0.84
2021-07-07 11:00:02 | 1.01

index    datetime64[ns]
raw             float64
dtype: object
```



FOr this data when i try to run 

```
from luminaire.optimization.hyperparameter_optimization import HyperparameterOptimization
hopt_obj = HyperparameterOptimization(freq='H', detection_type='OutlierDetection')
print(type(hopt_obj))
opt_config = hopt_obj.run(data=df_conc)
print(opt_config)

```

<class 'luminaire.optimization.hyperparameter_optimization.HyperparameterOptimization'>
None

my frequency is actually every 30 mins.. but every freq= gives me same None .

Any help is appreciated.

[sayanchk]

@webbug2005 Currently, Luminaire batch models only supports certain pandas offsets (link) and 30 mins is currently not supported. That being said, you can still use the WindowDensityModel which supports any arbitrary time frequency. The difference with the batch model is window based model mostly focus identifying anomalies over time windows instead of individual outlying time points.

[webbug2005]

my end goal is the use streaming data, that comes every 30 mins.. so will skip to directly testing WindowDensity. But this means i cannot use optimization for streaming data , as it doesnt support 30 mins ? (if i keep a window of 3 months)

[sayanchk]

Optimization is only for the batch models in Luminaire. WindowDensityModel has an embedded optimization precess to pick the right parameters, so, you don't need to use the optimization module separately.

[webbug2005]

@sayanchk , I tried with many data sets.. kept it limited to 30 days data , with every row at 30 mins interval. I also split the timestamp to just HH:MM as seconds was never exact. But keep hitting below error.

None {'success': False, 'ErrorMessage': 'Invalid number of FFT data points (0) specified.'}

[sayanchk]

@webbug2005, would you be able to share any reproducible code with a dummy data in colab? It will be easier for me to debug that way!