Sintel

Analyze, Interact, Decide

The ecosystem of Sintel allows you to build advanced analytic workflows with human in the loop using the Machine Learning libraries, RESTful APIs, and Visualization tools.

Sintel

Featurization

Featurize time series with domain knowledge encoded for machine learning uses.

Forecasting

Predict future values by analyzing past trends.

Anomaly Detection

Identify anomalous time series segments.

Classification

Classify time series segments into particular categories.

Benchmarking

Benchmark machine learning models for different tasks.

Data Visualization

Use a suite of visualization techniques to support complex analytic workflow with human-in-the-loop

Code

from sigpro import SigPro
from sigpro.demo import (
  get_amplitude_demo,
  load_demo_primitives
)

data = get_amplitude_demo()

trans, aggre = load_demo_primitives()

sigpro = SigPro(
  transformations=trans,
  aggregations=aggre,
  keep_columns=True
)

sigpro.process_signal(data)

Code

from pyteller.data import load_data
from pyteller.core import Pyteller

train_data = load_data(
  'pyteller/data/AL_Weather_current.csv')
test_data = load_data(
  'pyteller/data/AL_Weather_input.csv')

pyteller = Pyteller(
  pipeline=('pyteller/pipelines/sandbox/'
            'LSTM/LSTM_offset.json'),
  pred_length=6, offset=1)

pyteller.fit(
  data=train_data,
  timestamp_col='valid', target_signal='tmpf',
  entity_col='station', entities='8A0')

pyteller.forecast(data=test_data)

Code

from orion.data import load_signal
from orion import Orion

train_data = load_signal('S-1-train')

orion = Orion(pipeline='lstm_dynamic_threshold')

orion.fit(train_data)

new_data = load_signal('S-1-new')

orion.detect(new_data)

Code

from draco.demo import load_demo
from draco.pipeline import DracoPipeline
from sklearn.model_selection import train_test_split

target_times, readings = load_demo()

train, test = train_test_split(
  target_times, test_size=0.25, random_state=0)

test_targets = test.pop('target')

draco = DracoPipeline(
  'classes.normalize_dfs_xgb_classifier')

draco.fit(train, readings)

draco.predict(test, readings)

Code

from orion.benchmark import benchmark

pipelines = [
  'arima',
  'lstm_dynamic_threshold'
]

signals = ['S-1', 'P-1']

metrics = ['f1', 'accuracy', 'recall', 'precision']

benchmark(
  pipelines=pipelines,
  datasets=signals,
  metrics=metrics,
  rank='f1'
)

Use Cases

Sintel have been demonstrated in the following application domains.

Satellite

Monitor satellite telemetry and solve an unsupervised time series anomaly detection problem.

Wind Turbine

Monitor wind energy production systems and solve a time series classification problem.

Water

Monitor water distribution and delivery networks and address time series classification and regression problems.

Open Source

Sintel is an ecosystem for time series processing, labeling, modeling, benchmarking, and user interactions. Explore the ecosystem of open source libraries supporting the Sintel. Each can be used as a standalone package for particular needs.