@@ -317,14 +317,15 @@ classification or clustering.
317
317
318
318
Similarly to the transformation module, similarity search estimators are either defined
319
319
for single series or for collection of series. The estimators are inheriting from the
320
- [ BaseSimiliaritySearch ] ( similarity_search._base.BaseSimiliaritySearch ) class, which
321
- both [ BaseSeriesSimiliaritySearch ] ( similarity_search.series._base.BaseSeriesSimiliaritySearch )
322
- and [ BaseCollectionSimiliaritySearch ] ( similarity_search.collection._base.BaseCollectionSimiliaritySearch )
320
+ [ BaseSimilaritySearch ] ( similarity_search._base.BaseSimilaritySearch ) class, which
321
+ both [ BaseSeriesSimilaritySearch ] ( similarity_search.series._base.BaseSeriesSimilaritySearch )
322
+ and [ BaseCollectionSimilaritySearch ] ( similarity_search.collection._base.BaseCollectionSimilaritySearch )
323
323
inherit from.
324
324
325
325
All estimators use a ` fit ` ` predict ` interface, where ` predict ` outputs both the
326
326
indexes of the neighbors or motifs and a distance or similarity measure linked to them.
327
327
For example, using ` StompMotif ` to compute the matrix profile between two series :
328
+
328
329
``` {code-block} python
329
330
>>> import numpy as np
330
331
>>> from aeon.similarity_search.series import StompMotif
@@ -333,6 +334,7 @@ For example, using `StompMotif` to compute the matrix profile between two series
333
334
>>> top_k = StompMotif(4).fit(X1) # 4 is length of the motif to search
334
335
>>> distances, indexes = top_k.predict(X2, k=1)
335
336
```
337
+
336
338
Some things to note on this example :
337
339
338
340
- We defined ` 1D ` series of shape ` (n_timepoints) ` , but internally, series estimator
@@ -350,7 +352,7 @@ and those that transform a collection.
350
352
351
353
### Transformers for Single Time Series
352
354
353
- Transformers inheriting from the [ BaseSeriesTransformer] ( transformations.base.BaseSeriesTransformer )
355
+ Transformers inheriting from the [ BaseSeriesTransformer] ( transformations.series. base.BaseSeriesTransformer )
354
356
in the ` aeon.transformations.series ` package transform a single (possibly multivariate)
355
357
time series into a different time series or a feature vector. More info to follow.
356
358
@@ -379,7 +381,7 @@ Most time series classification and regression algorithms are based on some form
379
381
transformation into an alternative feature space. For example, we might extract some
380
382
summary time series features from each series, and fit a traditional classifier or
381
383
regressor on these features. For example, we could use
382
- [ Catch22] ( transformations.collection.feauture_based ) , which calculates 22 summary
384
+ [ Catch22] ( transformations.collection.feature_based.Catch22 ) , which calculates 22 summary
383
385
statistics for each series.
384
386
385
387
``` {code-block} python
@@ -397,7 +399,7 @@ statistics for each series.
397
399
```
398
400
399
401
There are also series-to-series transformations, such as the
400
- [ Padder] ( transformations.collection ) to lengthen
402
+ [ Padder] ( transformations.collection.Padder ) to lengthen
401
403
series and process unequal length collections.
402
404
403
405
``` {code-block} python
@@ -432,7 +434,7 @@ For machine learning tasks such as classification, regression and clustering, th
432
434
` scikit-learn ` ` make_pipeline ` functionality can be used if the transformer outputs
433
435
a valid input type.
434
436
435
- The following example uses the [ Catch22] ( transformations.collection.catch22 .Catch22 )
437
+ The following example uses the [ Catch22] ( transformations.collection.feature_based .Catch22 )
436
438
feature extraction transformer and a random forest classifier to classify.
437
439
438
440
``` {code-block} python
0 commit comments