In [1]:

%load_ext autoreload
%autoreload 2

import numpy as np
import pandas as pd
import matplotlib

import matplotlib.pyplot as plt
%matplotlib inline

import adtk.detector as detector
import adtk.transformer as transformer
import adtk.aggregator as aggregator
from adtk.visualization import plot

Single Series, no anomaly¶

In [2]:

rand = np.random.RandomState(123)
s = pd.Series(np.cumsum(rand.normal(size=300)), index=pd.date_range(start="2017-1-1", periods=300, freq="D"))

plot(s);

Single Series, single anomaly¶

In [3]:

anomaly = detector.QuantileAD(low=0.05, high=0.95).fit_detect(s, return_list=False)
plot(s, anomaly);

In [4]:

anomaly = detector.QuantileAD(low=0.05, high=0.95).fit_detect(s, return_list=False)
plot(s, anomaly, anomaly_tag="marker");

In [5]:

anomaly = detector.QuantileAD(low=0.05, high=0.95).fit_detect(s, return_list=True)
plot(s, anomaly);

In [6]:

anomaly = detector.QuantileAD(low=0.05, high=0.95).fit_detect(s, return_list=True)
plot(s, anomaly, anomaly_tag="marker");

In [7]:

plot(None, anomaly);

Single Series, multiple anomalies¶

In [8]:

low = detector.QuantileAD(low=0.05).fit_detect(s, return_list=True)
high = detector.QuantileAD(high=0.95).fit_detect(s, return_list=True)
anomaly = {"low": low, "high": high}
plot(s, anomaly);

In [9]:

plot(s, anomaly, anomaly_tag="marker");

In [10]:

low = detector.QuantileAD(low=0.05).fit_detect(s, return_list=False)
high = detector.QuantileAD(high=0.95).fit_detect(s, return_list=False)
anomaly = {"low": low, "high": high}
plot(s, anomaly);

In [11]:

plot(s, anomaly, anomaly_tag="marker");

In [12]:

plot(None, anomaly);

Multiple Series, no anomaly¶

In [13]:

df = pd.concat([s.rename('A'), (s**2/10).rename("B"), (-s*2).rename("C")], axis=1)
plot(df);

In [14]:

plot(df, curve_group='all');

In [15]:

plot(df, curve_group=[["A", "B"], ["B", "C"], "C"]);

Multiple Series, single anomaly¶

In [16]:

anomaly = detector.PcaAD(k=1, c=1).fit_detect(df)
plot(df, anomaly);

In [17]:

plot(df, anomaly, anomaly_tag="marker");

Multiple Series, respective anomaly¶

In [18]:

anomaly = detector.ThresholdAD(low=-10, high=15).detect(df)
plot(df, anomaly);

In [19]:

plot(df, anomaly, anomaly_tag="marker");

In [20]:

plot(df, anomaly, curve_group="all", anomaly_tag="marker");

In [21]:

anomaly = {
    "low": detector.QuantileAD(low=0.1).fit_detect(df),
    "high": detector.QuantileAD(high=0.9).fit_detect(df)
}
plot(df, anomaly);

In [22]:

plot(df, anomaly, anomaly_tag="marker");

In [23]:

plot(None, anomaly);

Multiple Series, multiple anomaly¶

In [24]:

anomaly = {
    "low": detector.QuantileAD(low=0.1).fit_detect(df),
    "high": detector.QuantileAD(high=0.9).fit_detect(df),
    "pca": detector.PcaAD(k=1).fit_detect(df)
}

In [25]:

plot(df, anomaly);

In [26]:

plot(df, anomaly, anomaly_tag="marker");

In [27]:

plot(df, anomaly, anomaly_tag={"low":"marker", "high":"marker"});

In [28]:

plot(df, anomaly, curve_group=[("A", "B"), ("B", "C"), ("C", "A")], anomaly_tag={"low":"marker", "high":"marker"});

In [29]:

plot(None, anomaly);

Play with properties¶

In [30]:

plot(df, anomaly, curve_group=[("A", "B"), ("B", "C"), ("C", "A")],
     ts_linewidth=1,
     ts_marker='v',
     ts_markersize=3,
     ts_alpha={"C": 0.2},
     ts_color={"A":"k"},
     anomaly_tag={"low":"marker", "high":"marker"},
     anomaly_color={"low":{"B":"m"}},
     anomaly_alpha=0.9,
     anomaly_marker={"high":"+"},
     anomaly_markersize={"low":{"A": 10}},
     freq_as_period=False,
     figsize=(16, 20)
    );

In [31]:

f, axes = plt.subplots(nrows=3, figsize=(12,10))
plot(df, anomaly, curve_group=[("A", "B"), ("B", "C"), ("C", "A")],
     ts_linewidth=1,
     ts_marker='v',
     ts_markersize=3,
     ts_alpha={"C": 0.2},
     ts_color={"A":"k"},
     anomaly_tag={"low":"marker", "high":"marker"},
     anomaly_color={"low":{"B":"m"}},
     anomaly_alpha=0.9,
     anomaly_marker={"high":"+"},
     anomaly_markersize={"low":{"A": 10}},
     freq_as_period=False,
     axes=axes,
     legend=False
    );

In [32]:

plot();

In [ ]: