Images about time-series-segmentation-benchmark HOT 14 CLOSED

Hi,

I just added the code to generate the visualizations of the time series as well as predicted segmentations. Check out the updated README.

Cheers, Arik

from time-series-segmentation-benchmark.

How please to generate all the images for all the dataset, i try this but something was wrong.

for _, (ts, ts_name, cps, found_cps) in tssb.iterrows():
import matplotlib.pyplot as plt
from tssb.utils import visualize_time_series
fig, ax = visualize_time_series(ts, ts_name, cps, found_cps)
plt.show()

from time-series-segmentation-benchmark.

The row

for _, (ts, ts_name, cps, found_cps) in tssb.iterrows():

is false. TSSB consists of tuples (ts_name, window_size, cps, ts). Try:

for _, (ts_name, window_size, cps, ts) in tssb.iterrows():
    fig, ax = visualize_time_series(ts, ts_name, cps)
    plt.savefig(f"images/{ts_name}.png", bbox_inches="tight")
    plt.clf()

Cheers, Arik

from time-series-segmentation-benchmark.

for _, (ts_name, window_size, cps, ts) in tssb.iterrows():
fig, ax = visualize_time_series(ts, ts_name, cps)
plt.savefig(f"images/{ts_name}.png", bbox_inches="tight")
plt.clf()

Thank u, but this is before segmentation prediction, i want after the prediction (found changes points)

from time-series-segmentation-benchmark.

In that case, simply add your predicted change points to the method

fig, ax = visualize_time_series(ts, ts_name, cps, found_cps)

from time-series-segmentation-benchmark.

In that case, simply add your predicted change points to the method fig, ax = visualize_time_series(ts, ts_name, cps, found_cps)

i did it, but the result was wrong... Like this..
-----for _, (ts_name, window_size, cps, ts) in tssb.iterrows():
fig, ax = visualize_time_series(ts, ts_name, cps, found_cps)
plt.savefig(f"tssb/results/{ts_name}.png", bbox_inches="tight")
plt.clf()-----

and this is an example of the result...

from time-series-segmentation-benchmark.

Well, is your prediction

found_cps = your_algorithm_predict(ts)

in the for loop? And also, can you print your prediction (found_cps)? It seems like it is way too large.

from time-series-segmentation-benchmark.

The following code

import numpy as np
import matplotlib.pyplot as plt

from tssb.utils import load_time_series_segmentation_datasets, visualize_time_series
tssb = load_time_series_segmentation_datasets(names=["Adiac"])

for _, (ts_name, window_size, cps, ts) in tssb.iterrows():
    found_cps = np.array([1000])
    fig, ax = visualize_time_series(ts, ts_name, cps, found_cps)
    plt.show()
    plt.clf()

leads to the following image:

from time-series-segmentation-benchmark.

I'm sorry, but i tried that, and it's correct juste for Adiac.
For example, for beetlefly, i got that

from time-series-segmentation-benchmark.

But this is exactly what you would expect to get for BeetleFly, if your predicted Change Point is at position 1000.

from time-series-segmentation-benchmark.

But this is exactly what you would expect to get for BeetleFly, if your predicted Change Point is at position 10
oh yes, i'm stupide.. but how to print for all our predictions, for example like this (but for all images) ...

from time-series-segmentation-benchmark.

Well, you simply predict the CPs in the for loop and save the predicted segmentations.

import numpy as np
import matplotlib.pyplot as plt

from tssb.utils import load_time_series_segmentation_datasets, visualize_time_series
tssb = load_time_series_segmentation_datasets(names=["Adiac"])

for _, (ts_name, window_size, cps, ts) in tssb.iterrows():
    found_cps = your_segmentation_algorithm_predict(ts, len(cps))
    fig, ax = visualize_time_series(ts, ts_name, cps, found_cps)
    plt.savefig(f"images/{ts_name}.png", bbox_inches="tight")
    plt.clf()

from time-series-segmentation-benchmark.

Well, you simply predict the CPs in the for loop and save the predicted segmentations.

import numpy as np
import matplotlib.pyplot as plt

from tssb.utils import load_time_series_segmentation_datasets, visualize_time_series
tssb = load_time_series_segmentation_datasets(names=["Adiac"])

for _, (ts_name, window_size, cps, ts) in tssb.iterrows():
    found_cps = your_segmentation_algorithm_predict(ts, len(cps))
    fig, ax = visualize_time_series(ts, ts_name, cps, found_cps)
    plt.savefig(f"images/{ts_name}.png", bbox_inches="tight")
    plt.clf()

Yes, now it's okay, thank you very much. This is the final code for your implementation

for _, (ts_name, window_size, cps, ts) in tssb.iterrows():
found_cps = ClaSPSegmentation(window_size, n_cps=len(cps)).fit_predict(ts)
fig, ax = visualize_time_series(ts, ts_name, cps, found_cps)
plt.savefig(f"tssb/results/{ts_name}.png", bbox_inches="tight")
plt.clf()

from time-series-segmentation-benchmark.

Yes! You are welcome!

from time-series-segmentation-benchmark.