[1]:

# Import to be able to import python package from src
import sys
sys.path.insert(0, '../src')

[2]:

import ontime as on
import pandas as pd

Getting Started#

The library is divided in three parts :

  1. core for all basic features

  2. module for all features using core features. E.g. benchmarking, ml preprocessing, etc.

  3. context for all features related to the usage of onTime in an applied scenario

core Features#

This is a low level API. Most objects and functions are accessible in the base object :

ontime
├── detectors
├── generators
├── Model
├── plots
├── processors
└── TimeSeries

For instance :

[3]:

ts = on.generators.random_walk().generate(start=pd.Timestamp('01-01-2023'), end=pd.Timestamp('12-31-2023'))
ts[0:5]

[3]:

<TimeSeries (DataArray) (time: 5, component: 1, sample: 1)> Size: 40B
array([[[ 0.8042713 ]],

       [[ 1.58685749]],

       [[ 0.88496745]],

       [[ 1.07890811]],

       [[-0.41554812]]])
Coordinates:
  * time       (time) datetime64[ns] 40B 2023-01-01 2023-01-02 ... 2023-01-05
  * component  (component) object 8B 'random_walk'
Dimensions without coordinates: sample
Attributes:
    static_covariates:  None
    hierarchy:          None
[4]:

ts.plot()

[4]:

module and context features#

High level API with various features. Let’s load some data for an example :

[5]:

from darts.datasets import EnergyDataset
ts = EnergyDataset().load()

[6]:

df = ts.pd_dataframe()
df = df.interpolate()
cols = ['generation biomass', 'generation solar', 'generation nuclear']
df = df[cols]

[7]:

ts = on.TimeSeries.from_dataframe(df)

[8]:

ts_uni = ts['generation solar'].slice(pd.Timestamp('2015'), pd.Timestamp('2016'))
ts_multi = ts.slice(pd.Timestamp('2015'), pd.Timestamp('2016'))

module Features#

High level API with features related to data processing, ML/AI, etc.

[9]:

train, test = on.module.processing.common.train_test_split(ts_uni, test_split=0.3)

context Features#

High level API with features related to a physical machine or process.

Profiler#

[10]:

profiler = on.context.common.Profiler()

What does the common week looks like ?

[11]:

week_mean = profiler.profile(ts_uni, profiler.Period.WEEKLY, profiler.Aggregation.MEAN).rename({"value": "week_mean"})
week_median = profiler.profile(ts_uni, profiler.Period.WEEKLY, profiler.Aggregation.MEDIAN).rename({"value": "week_median"}) # variable is renamed cause error raises when set to "value"

[12]:

(
    on.Plot()
    .add(on.marks.line, week_mean)
    .add(on.marks.line, week_median)
    .show()
)

[12]:

Generic Predictor#

[13]:

model = on.context.common.GenericPredictor()

[14]:

model.fit(train)

[14]:

<ontime.context.common.generic_predictor.GenericPredictor at 0x7f775011cdf0>

What does the future looks like ?

[15]:

pred = model.predict(48)

[16]:

(
    on.Plot()
    .add(on.marks.line, train[-96:].rename({"generation solar": "Training set"}))
    .add(on.marks.line, pred.rename({"generation solar": "Prediction"}))
    .add(on.marks.line, test[:48].rename({"generation solar": "Truth"}), type="dashed")
    .properties(width=600, height=300)
    .show()
)

[16]:

Generic Detector#

[17]:

model = on.context.common.GenericDetector()

[18]:

model.fit(train)

[18]:

<ontime.context.common.generic_detector.GenericDetector at 0x7f775011e980>

Does the current signal has problem ?

[19]:

detected_test = model.detect(test)

[20]:

(
    on.Plot(test[:72])
    .add(on.marks.line)
    .add(on.marks.mark, data=detected_test[:72].rename({"generation solar": "Anomalies"}), type="dot")
    .properties(width=600, height=300)
    .show()
)

[20]:

What if we want to have an idea about the future problems ?

[21]:

predetected = model.predetect(72)

[22]:

(
    on.Plot(test[:72])
    .add(on.marks.line)
    .add(on.marks.mark, data=predetected[:72].rename({"generation solar": "Anomalies"}), type="dot")
    .properties(width=600, height=300)
    .show()
)

[22]:
On this page
Show Source