Quick start

Installation

Install package with pip.

pip install traffic-weaver

Minimal processing example

Importing Traffic Weaver.

import traffic_weaver

To obtain the same results, set numpy seed to 0.

# setting random seed to 0
from numpy import random
random.seed(0)

To load one of the local exemplary datasets, use

# load example dataset containing average measurements over 1 hour
data = traffic_weaver.datasets.load_dataset('sandvine_tiktok')
print(data.T)

which outputs

[[ 0.    1.    2.    3.    4.    5.    6.    7.    8.    9.   10.   11.  12.   13.   14.   15.   16.   17.   18.   19.   20.   21.   22.   23.  ]
 [ 0.45  0.4   0.45  0.2   0.3   0.9   0.9   1.3   1.65  1.15  1.6   1.6  1.5   2.1   1.45  1.6   1.45  2.55  1.4   2.45  1.6   1.    1.35  1.05]]

The traffic_weaver provides Weaver class that serves as an API to other processing capabilities. :code: Weaver.from_2d_array(data) factory takes time series of independent and dependent variables as 2D array.

# create Weaver instance
wv = traffic_weaver.Weaver.from_2d_array(data)

Further signal processing is applied through Weaver methods. Most of the methods return instance to the Weaver itself, allowing for chaining processing commands.

# process it creating samples every minute
wv.oversample(60).integral_match().smooth(1.0).noise(snr=30)

To obtain created new time series call either Weaver.get() or Weaver.to_function(). The former one returns list of created points, and the latter one spline function created based on those points and allow to sample any point.

x, y = wv.get()
print(x)
print(y)

[0.000e+00 1.667e-02 3.333e-02 ... 2.297e+01 2.298e+01 2.300e+01]
[0.504 0.447 0.477 ... 1.047 1.044 1.065]

f = wv.to_function()
print(f(0.01))

0.45616116907018994

To visualize time series, matplotlib library is required.

import matplotlib.pyplot as plt

# plot original signal
fig, axes = plt.subplots(nrows=1, ncols=3, figsize=(14, 4))
axes[0].plot(*wv.get_original(), drawstyle="steps-post")

# plot modified signal
axes[1].plot(*wv.get())

# plot averaged signal
x, y = traffic_weaver.process.average(*wv.get(), 60)
axes[2].plot(x, y, drawstyle="steps-post")

axes[0].set_title("a) Original", loc="left")
axes[1].set_title("b) Processed", loc="left")
axes[2].set_title("c) Averaged", loc="left")
plt.show()

Above code produces the following figures.

Minimal processing example

It is possible to check if integrals actually match between original and processed function.

# compare integrals of original and modified signal
print("original function integral={}".format(sum(traffic_weaver.sorted_array_utils.integral(*(wv.get_original()), method='rectangle'))))
print("modified function integral={}".format(sum(traffic_weaver.sorted_array_utils.integral(wv.x, wv.y, method='trapezoid'))))

Resulting in:

    original function integral=29.35
modified function integral=29.327691841128036

The small difference in integral value is a result of smoothing and adding noise.

Further processing examples

Traffic Weaver provides further methods to process time series.

import traffic_weaver
import matplotlib.pyplot as plt
import math
from numpy import random
random.seed(0)

# load example dataset containing average measurements over 1 hour
x, y = traffic_weaver.datasets.load_dataset('sandvine_tiktok', unpack_dataset_columns=True)

wv = traffic_weaver.Weaver(x, y)
# add one sample to the end to make it first sample of next day
wv.append_one_sample(make_periodic=True)
wv.recreate_from_average(60).integral_match().repeat(14).smooth(1.0).trend(
    lambda x: 2 * x + 1 / 2 * math.sin(math.pi * x * 7), normalized=True).noise(snr=35)

# visualize
fig, axes = plt.subplots(figsize=(12, 4))
axes.plot(*wv.get())

plt.tight_layout()
plt.show()
Further processing example