|
11 | 11 |
|
12 | 12 | from columnflow.util import maybe_import, DotDict |
13 | 13 |
|
14 | | - |
15 | 14 | np = maybe_import("numpy") |
16 | 15 | hist = maybe_import("hist") |
17 | 16 |
|
18 | | - |
19 | 17 | logger = law.logger.get_logger(__name__) |
20 | 18 |
|
21 | 19 |
|
22 | | -# def integrate_hist(h, **kwargs): |
23 | | -# """ |
24 | | -# Given a scikit-hist histogram object return a reduced histogram with specified |
25 | | -# axes integrated out. |
26 | | - |
27 | | -# For scikit-hist histograms, the integration should be formed in 3 steps: |
28 | | -# - slicing the histogram to contain only the range of interest |
29 | | -# - Setting overflow values to 0 (excluding the values from future calculations) |
30 | | -# - Summing over the axes of interest. |
31 | | - |
32 | | -# The latter 2 steps will only be carried out if the var_slice doesn't uniquely |
33 | | -# identify a singular bin in the histogram axis |
34 | | -# """ |
35 | | -# # Reduction in parallel. |
36 | | -# r = h[kwargs] |
37 | | -# for var, var_slice in kwargs.items(): |
38 | | -# # In the case that histogram has been reduced to singular value simple return |
39 | | -# if not isinstance(r, hist.Hist): |
40 | | -# return r |
41 | | -# if var in [x.name for x in r.axes]: |
42 | | -# ax = h.axes[var] |
43 | | -# get_underflow = var_slice.start == None or var_slice.start == -1 |
44 | | -# get_overflow = var_slice.stop == None or var_slice.stop == len(ax) |
45 | | -# if not get_underflow and ax.traits.underflow: |
46 | | -# r[{var: hist.underflow}] = np.zeros_like(r[{var: hist.underflow}]) |
47 | | -# if not get_overflow and ax.traits.overflow: |
48 | | -# r[{var: hist.overflow}] = np.zeros_like(r[{var: hist.overflow}]) |
49 | | - |
50 | | -# # Sum over all remaining elements on axis |
51 | | -# r = r[{var: sum}] |
52 | | -# return r |
53 | | - |
54 | | - |
55 | | -# def rebin_hist(h, **kwargs): |
56 | | -# """ |
57 | | -# Rebinning a scikit-hist histogram. 2 types of values can be accepted as the |
58 | | -# argument values: |
59 | | -# - Derivatives of the `hist.rebin` argument. In this case we directly use the |
60 | | -# UHI facilities to perform the rebinning. |
61 | | -# - A new axis object where all the bin edges lands on the old bin edges of the |
62 | | -# given histogram. In this case a custom intergration loop is performed to |
63 | | -# extract the rebinning. Beware that this methods is very slow, as it requires |
64 | | -# a loop generation of all possible UHI values after the rebinning, so be sure |
65 | | -# that rebinning is performed as the final step of the histogram reduction. See |
66 | | -# `_rebin_single_scikit` for more information regarding this method. |
67 | | -# """ |
68 | | -# h = h.copy() |
69 | | -# for var, var_val in kwargs.items(): |
70 | | -# if isinstance(var_val, hist.rebin): |
71 | | -# h = h[{var: var_val}] |
72 | | -# else: |
73 | | -# h = _rebin_single_scikit(h, var, var_val) |
74 | | -# return h |
75 | | - |
76 | | - |
77 | | -# def __check_scikit_axis_compat(axis1, axis2): |
78 | | -# """ |
79 | | -# Checking that axis 2 is rebin-compatible with axis 1. This checks that: |
80 | | -# 1. The two histogram share the same name. |
81 | | -# 2. The edges of the second axis all land on the edges of the first axis. |
82 | | - |
83 | | -# If the two axis are compatible the function will return an array of the bin |
84 | | -# index of the axis 1 that the bin edges of axis 2 falls on. |
85 | | -# """ |
86 | | -# assert axis1.name == axis2.name, \ |
87 | | -# 'Naming of the axis is required to match' |
88 | | -# # Getting the new bin edges index for the old bin edges |
89 | | -# try: |
90 | | -# return [ |
91 | | -# np.argwhere(axis1.edges == new_edge)[0][0] for new_edge in axis2.edges |
92 | | -# ] |
93 | | -# except IndexError: |
94 | | -# raise ValueError( |
95 | | -# f"Bin edges of the axis {axis2} is incompatible with {axis1}") |
96 | | - |
97 | | - |
98 | | -# def _get_all_indices(axis): |
99 | | -# """ |
100 | | -# Getting all possible (integer) bin index values given a scikit-hep histogram. |
101 | | -# The special indices of hist.underflow and hist.overflow will be included if the |
102 | | -# axis in questions has those traits. |
103 | | -# """ |
104 | | -# idxs = list(range(len(axis))) |
105 | | -# if axis.traits.underflow: # Extension to include the under/overflow bins |
106 | | -# idxs.insert(0, hist.underflow) |
107 | | -# if axis.traits.overflow: |
108 | | -# idxs.append(hist.overflow) |
109 | | -# return idxs |
110 | | - |
111 | | - |
112 | | -# def _rebin_single_scikit(h, old_axis, new_axis): |
113 | | -# """ |
114 | | -# Rebinning a single axis of a scikit-hist histogram. This includes the following |
115 | | -# routines: |
116 | | - |
117 | | -# - Generating a new scikit hep instance that perserves axis ordering with the |
118 | | -# exception of the rebinned axis (in place) replacement. |
119 | | -# - Setting up the integration ranges required to calculate the bin values of the |
120 | | -# new histogram. |
121 | | -# - Looping over the UHI values of the new histogram and performing the a |
122 | | -# summation over the specified range on the old histogram to fill in the new |
123 | | -# values. |
124 | | - |
125 | | -# As here we have variable number of axis each with variable number of bins, this |
126 | | -# method will require the use of more old fashioned python looping, which can be |
127 | | -# very slow for large dimensional histograms with many bins for each axis. So be |
128 | | -# sure to make rebinning be the final step in histogram reduction. |
129 | | -# """ |
130 | | -# # assert isinstance(h, hist.NamedHist), "Can only process named histograms" |
131 | | -# # Additional type casing |
132 | | -# if isinstance(old_axis, str): |
133 | | -# return _rebin_single_scikit(h, h.axes[old_axis], new_axis) |
134 | | -# axis_name = old_axis.name |
135 | | - |
136 | | -# # Creating the new histogram instance with identical axis ordering. |
137 | | -# all_axes = list(h.axes) |
138 | | -# all_axes[all_axes.index(old_axis)] = new_axis |
139 | | -# h_rebinned = hist.Hist(*all_axes, storage=h._storage_type()) |
140 | | - |
141 | | -# # Getting the all possible bin indices for all axes in the old histogram |
142 | | -# bin_idx_dict = {ax.name: _get_all_indices(ax) for ax in h.axes} |
143 | | - |
144 | | -# # Getting the new bin edges index for the old bin edges |
145 | | -# new_bin_edge_idx = __check_scikit_axis_compat(old_axis, new_axis) |
146 | | -# if new_axis.traits.underflow: # Adding additional underflow/overflow |
147 | | -# new_bin_edge_idx.insert(0, bin_idx_dict[axis_name][0]) |
148 | | -# if new_axis.traits.overflow: |
149 | | -# new_bin_edge_idx.append(bin_idx_dict[axis_name][-1]) |
150 | | - |
151 | | -# # Generating a the int range pair. Additional parsing will be required for the |
152 | | -# # under/overflow bins |
153 | | -# def make_slice(index): |
154 | | -# start = new_bin_edge_idx[index] |
155 | | -# stop = new_bin_edge_idx[index + 1] |
156 | | -# if start == hist.underflow: |
157 | | -# start = -1 |
158 | | -# if stop == hist.overflow: |
159 | | -# stop = len(old_axis) |
160 | | -# return slice(int(start), int(stop)) |
161 | | - |
162 | | -# new_axis_idx = _get_all_indices(new_axis) |
163 | | -# new_int_slice = [make_slice(i) for i in range(len(new_axis_idx))] |
164 | | -# assert len(new_axis_idx) == len(new_bin_edge_idx) - 1 |
165 | | - |
166 | | -# new_idx_dict = bin_idx_dict.copy() |
167 | | -# new_idx_dict[axis_name] = new_axis_idx |
168 | | -# bin_idx_dict[axis_name] = new_int_slice |
169 | | - |
170 | | -# name_list = list(bin_idx_dict.keys()) |
171 | | -# new_idx = [x for x in itertools.product(*[x for x in new_idx_dict.values()])] |
172 | | -# old_int = [x for x in itertools.product(*[x for x in bin_idx_dict.values()])] |
173 | | - |
174 | | -# print(new_idx) |
175 | | -# print("Here") |
176 | | -# print(old_int) |
177 | | -# for o, n in zip(old_int, new_idx): |
178 | | -# n_uhi = {name: n[name_idx] for name_idx, name in enumerate(name_list)} |
179 | | -# o_uhi = {name: o[name_idx] for name_idx, name in enumerate(name_list)} |
180 | | -# # Single variable histogram, with just the axis of interest |
181 | | -# h_rebinned[n_uhi] = integrate_hist(h, **o_uhi) |
182 | | - |
183 | | -# return h_rebinned |
184 | | - |
185 | | - |
186 | 20 | def rebin_hist(h, axis_name, edges): |
187 | 21 | if isinstance(edges, int): |
188 | 22 | return h[{axis_name: hist.rebin(edges)}] |
|
0 commit comments