Differences

This shows you the differences between two versions of the page.

--- other:python:misc_by_jyp [2021/10/27 16:05] – Added sets jypeter
+++ other:python:misc_by_jyp [2022/02/21 17:46] – [numpy related stuff] Added reduceat jypeter
@@ Line 206: / Line 206: @@
 ==== Sorting ====
+  * When dealing with **numerical values**, you should use the [[https://numpy.org/doc/stable/reference/routines.sort.html|numpy sorting, searching, and counting routines]]!
   * [[https://docs.python.org/3/howto/sorting.html|Sorting HOW TO]]
   * Example: sorting the keys and the values of a dictionary, and then using the ''key'' parameter to sort the keys of a dictionary according to the value associated with the key
@@ Line 220: / Line 221: @@
 ['c', 'd', 'b', 'a']</code>
+==== numpy related stuff ====
+=== Finding and counting unique values ===
+Use ''np.unique'', do **not** try to use histogram related functions!
+<code>>>> vals = np.random.randint(2, 5, (10,)) * 0.5 # Get 10 discreet float values
+>>> vals
+array([1. , 2. , 1. , 2. , 2. , 1.5, 1. , 1.5, 2. , 1.5])
+>>> np.unique(vals)
+array([1. , 1.5, 2. ])
+>>> unique_vals, nb_unique = np.unique(vals, return_counts=True)
+>>> unique_vals
+array([1. , 1.5, 2. ])
+>>> nb_unique
+array([3, 3, 4])
+>>> sorted_vals = np.sort(vals) # Sorted copy, in order to check the result
+>>> sorted_vals
+array([1. , 1. , 1. , 1.5, 1.5, 1.5, 2. , 2. , 2. , 2. ])</code>
+=== Applying a ufunc over all the elements of an array ===
+There are all sorts of //ufuncs// (Universal Functions), and we will just use below ''add'' from the [[https://numpy.org/doc/stable/reference/ufuncs.html#math-operations|math operations]], applied on the arrays defined in [[#finding_and_counting_unique_values|Finding and counting unique values]]
+<code># Get the sum of all the elements of 'vals'
+>>> np.add.reduce(vals)
+.5
+>>> np.add.reduce(sorted_vals)
+.5
+>>> vals.sum() # The usual and easy way to do it
+.5
+# Compute the sum of the elements of 'nb_unique'
+# AND keep (accumulate) the intermediate results
+>>> nb_unique
+array([3, 3, 4])
+>>> np.add.accumulate(nb_unique)
+array([ 3,  6, 10])
+# The accumulated values can be used as indices to separate the different groups of sorted values!
+>>> sorted_vals
+array([1. , 1. , 1. , 1.5, 1.5, 1.5, 2. , 2. , 2. , 2. ])
+>>> sorted_vals[0:3]
+array([1., 1., 1.])
+>>> sorted_vals[3:6]
+array([1.5, 1.5, 1.5])
+>>> sorted_vals[6:10]
+array([2., 2., 2., 2.])
+# Compute the sum of each equal-value group
+>>> sorted_vals[0:3].sum(), sorted_vals[3:6].sum(), sorted_vals[6:10].sum()
+(3.0, 4.5, 8.0)</code>
+=== Applying a ufunc over specified sections of an array ===
+The [[https://numpy.org/doc/stable/reference/generated/numpy.ufunc.reduceat.html#numpy.ufunc.reduceat|reduceat]] function can be used to avoid explicit python loops, and improve the speed (but not the readability...) of a script. The example below //improves// what has been shown above
+<code># Define a list with the boundaries of the intervals we want to apply the 'add' function to
+# We need to add the beginning index (0), AND remove the last index
+# (reduceat will automatically go to the end of the input array
+>>> nb_unique
+array([3, 3, 4])
+>>> slices_indices = [0] + list(np.add.accumulate(nb_unique))
+>>> slices_indices.pop() # Remove last element
+>>> slices_indices
+[0, 3, 6]
+# Compute the sums over the selected intervals with just one call
+>>> np.add.reduceat(np.sort(vals), slices_indices)
+array([3. , 4.5, 8. ])</code>
 /*