Topk_matrixprofile Investigator

stumpy/aamp.py

@@ -240,7 +240,8 @@ def _aamp(
     return np.power(P[0, :, :], 1.0 / p), I[0, :, :]
 
 
-def aamp(T_A, m, T_B=None, ignore_trivial=True, p=2.0):
+def aamp(T_A, m, T_B=None, ignore_trivial=True, p=2.0, k=1):
+    # function needs to be changed to return top-k matrix profile
     """
     Compute the non-normalized (i.e., without z-normalization) matrix profile
 
@@ -267,6 +268,11 @@ def aamp(T_A, m, T_B=None, ignore_trivial=True, p=2.0):
     p : float, default 2.0
         The p-norm to apply for computing the Minkowski distance.
 
+    k : int, default 1
+        The number of top `k` smallest distances used to construct the matrix profile.
+        Note that this will increase the total computational time and memory usage
+        when k > 1.
+
     Returns
     -------
     out : numpy.ndarray

stumpy/aamped.py

@@ -12,7 +12,8 @@
 logger = logging.getLogger(__name__)
 
 
-def aamped(dask_client, T_A, m, T_B=None, ignore_trivial=True, p=2.0):
+def aamped(dask_client, T_A, m, T_B=None, ignore_trivial=True, p=2.0, k=1):
+    # function needs to be revised to return top-k matrix profile
     """
     Compute the non-normalized (i.e., without z-normalization) matrix profile
 
@@ -46,6 +47,11 @@ def aamped(dask_client, T_A, m, T_B=None, ignore_trivial=True, p=2.0):
     p : float, default 2.0
         The p-norm to apply for computing the Minkowski distance.
 
+    k : int, default 1
+        The number of top `k` smallest distances used to construct the matrix profile.
+        Note that this will increase the total computational time and memory usage
+        when k > 1.
+
     Returns
     -------
     out : numpy.ndarray

stumpy/core.py

@@ -7,7 +7,7 @@
 import inspect
 
 import numpy as np
-from numba import njit
+from numba import njit, prange
 from scipy.signal import convolve
 from scipy.ndimage import maximum_filter1d, minimum_filter1d
 from scipy import linalg
@@ -2559,3 +2559,48 @@ def _select_P_ABBA_value(P_ABBA, k, custom_func=None):
             MPdist = partition[k]
 
     return MPdist
+
+
+@njit(parallel=True)
+def _merge_topk_PI(PA, PB, IA, IB):
+    """
+    Merge two top-k matrix profiles PA and PB, and update PA (in place) while
+    prioritizing values of PA in ties. Also, update IA accordingly.
+
+    Parameters
+    ----------
+    PA : numpy.ndarray
+        a (top-k) matrix profile, with ndim of 2, where values in each row are
+        sorted in ascending order. Also, it needs to be the same shape as PB.
+
+    PB : numpy.ndarray
+        a (top-k) matrix profile, with ndim of 2, where values in each row are
+        sorted in ascending order. Also, it needs to be the same shape as PA.
+
+    IA : numpy.ndarray
+        a (top-k) matrix profile indices, corresponding to PA
+
+    IB : numpy.ndarray
+        a (top-k) matrix profile indices, corresponding to PB
+
+    Returns
+    -------
+    None
+    """
+    for i in prange(PB.shape[0]):
+        start = 0
+        stop = np.searchsorted(PA[i], PB[i, -1], side="right")
+
+        for j in range(PB.shape[1]):
+            if PB[i, j] < PA[i, -1]:
+                idx = np.searchsorted(PA[i, start:stop], PB[i, j], side="right") + start
+
+                for g in range(PB.shape[1] - 1, idx, -1):
+                    PA[i, g] = PA[i, g - 1]
+                    IA[i, g] = IA[i, g - 1]
+
+                PA[i, idx] = PB[i, j]
+                IA[i, idx] = IB[i, j]
+
+                start = idx
+                stop += 1  # because of shifting elements to the right by one

stumpy/gpu_aamp.py

@@ -339,7 +339,9 @@ def _gpu_aamp(
     return profile_fname, indices_fname
 
 
-def gpu_aamp(T_A, m, T_B=None, ignore_trivial=True, device_id=0, p=2.0):
+def gpu_aamp(T_A, m, T_B=None, ignore_trivial=True, device_id=0, p=2.0, k=1):
+    # function needs to be revised to return (top-k) matrix profile and
+    # matrix profile indices
     """
     Compute the non-normalized (i.e., without z-normalization) matrix profile with one
     or more GPU devices
@@ -375,6 +377,11 @@ def gpu_aamp(T_A, m, T_B=None, ignore_trivial=True, device_id=0, p=2.0):
     p : float, default 2.0
         The p-norm to apply for computing the Minkowski distance.
 
+    k : int, default 1
+        The number of top `k` smallest distances used to construct the matrix profile.
+        Note that this will increase the total computational time and memory usage
+        when k > 1.
+
     Returns
     -------
     out : numpy.ndarray