Merge pull request #23 from comet-toolkit/v0.21.0

allow to not copy data (and hence save memory)

Author: pdevis

comet_maths/_version.py

@@ -1 +1 @@
-__version__ = "0.21.2"
+__version__ = "0.22.3"

comet_maths/generate_sample/generate_sample.py

@@ -59,7 +59,7 @@ def generate_sample(
         corr_x = np.array([corr_x])
         i = 0
 
-    if np.any(u_x[i] < 0):
+    if (not comp_list) and (not u_x[i] is None) and np.any(u_x[i] < 0):
         raise ValueError(
             "comet_maths.generate_sample: u_x cannot have any negative values"
         )
@@ -650,7 +650,13 @@ def generate_sample_cov(
 
 
 def correlate_sample_corr(
-    sample, corr, mean=None, std=None, dtype=None, iterate_sample=False
+    sample,
+    corr,
+    mean=None,
+    std=None,
+    dtype=None,
+    iterate_sample=False,
+    maintain_sample_unmodified=False,
 ):
     """
     Method to correlate independent sample of input quantities using correlation matrix and Cholesky decomposition.
@@ -666,7 +672,9 @@ def correlate_sample_corr(
     :param dtype: dtype of the produced sample
     :type dtype: numpy.dtype, optional
     :param iterate_sample: boolean to indicate if comet_maths should iterate over the different samples when introducing correlation. (This is more time-consuming but might be necessary if the different samples have different error correlations), defaults to False.
-    :type iterate_sample: bool
+    :type iterate_sample: bool, optional
+    :param maintain_sample_unmodified: boolean to indicate if the provided sample must remain unchanged (as opposed to introducing correlation). This requires a full copy of the sample, and thus doubles the memory required. Defaults to False.
+    :type maintain_sample_unmodified: bool, optional
     :return: correlated sample of input quantities
     :rtype: array[array]
     """
@@ -693,7 +701,10 @@ def correlate_sample_corr(
     except:
         L = cm.nearestPD_cholesky(corr, corr=True)
 
-    sample_out = copy.deepcopy(sample)
+    if maintain_sample_unmodified:
+        sample_out = copy.deepcopy(sample)
+    else:
+        sample_out = sample
 
     if iterate_sample:
         sample_out_iter = np.empty(len(sample), dtype=object)

docs/content/random_generator.rst

@@ -54,9 +54,11 @@ generate_sample_random(), generate_sample_systematic() and generate_sample_corre
 There is generate_sample_same() function that trivially generates a sample where u_x is zero and thus consists of repeats of the provided x array.
 Then, there is the generate_sample_cov() function where a covariance matrix is provided instead of the u_x and corr_x (which contain the same information).
 
-A somewhat more useful function is correlate_sample_corr(). This function can be used to add correlation to a previously generated sample.
-This can particularly be useful when multiple MC samples where generated (e.g. for different variables, optionally each with their own internal
+Another useful function is correlate_sample_corr(). This function can be used to add correlation to a previously generated sample.
+This can particularly be useful when multiple MC samples were generated (e.g. for different variables, optionally each with their own internal
 correlation) and these different samples need to be correlated (e.g. because the different variables themselves are correlated).
+Note that by default the provided samples will be modified, though there is a keyword that allows to maintain the samples unchanged (at the cost of higher memory usage).
+
 Finally, there is the generate_error_sample() function, which gives the differences between the sample generated by generate_sample() and x itself.
 
 .. _pdf:

requirements.txt

@@ -1,5 +1,5 @@
 # required modules
-numpy
+numpy>=0.44.2
 matplotlib
 numdifftools
 scikit-learn

setup.py

@@ -30,7 +30,7 @@ def read(filename):
         "scikit-learn",
         "numdifftools",
         "scipy",
-        "punpy",
+        "punpy>=0.44.2",
         "matplotlib",
     ],
     extras_require={