1
|
|
|
import matplotlib.pyplot as plt |
2
|
|
|
from matplotlib.ticker import MaxNLocator |
3
|
|
|
import pandas as pd |
4
|
|
|
import h5py as h5 |
5
|
|
|
import numpy as np |
6
|
|
|
|
7
|
|
|
class StatsUtils(object): |
8
|
|
|
|
9
|
|
|
_pattern_dict = {"projection": ["SINOGRAM", "PROJECTION", "TANGENTOGRAM", "4D_SCAN", "SINOMOVIE"], |
10
|
|
|
"reconstruction": ["VOLUME_YZ", "VOLUME_XZ", "VOLUME_XY", "VOLUME_3D"]} |
11
|
|
|
_stats_key = ["max", "min", "mean", "mean_std_dev", "median_std_dev", "NRMSD"] |
12
|
|
|
|
13
|
|
|
plt.set_loglevel('WARNING') |
14
|
|
|
|
15
|
|
|
def generate_figures(self, filepath, savepath): |
16
|
|
|
f = h5.File(filepath, 'r') |
17
|
|
|
stats_dict, index_list, times_dict = self._get_dicts_for_graphs(f) |
18
|
|
|
loop_stats, loop_plugins = self._get_dicts_for_loops(f) |
19
|
|
|
f.close() |
20
|
|
|
|
21
|
|
|
#self.make_loop_graphs(loop_stats, loop_plugins, savepath) |
22
|
|
|
|
23
|
|
|
table_index_list = index_list |
24
|
|
|
for i in range(len(loop_plugins)): |
25
|
|
|
for space in list(table_index_list.keys()): |
26
|
|
|
for j, plugin in enumerate(table_index_list[space]): |
27
|
|
|
for loop_plugin in loop_plugins[i]: |
28
|
|
|
if loop_plugin == plugin[3::]: |
29
|
|
|
table_index_list[space][j] = f"{table_index_list[space][j]} (loop{i})" |
30
|
|
|
|
31
|
|
|
self.make_stats_table(stats_dict, table_index_list, f"{savepath}/stats_table.html") |
32
|
|
|
|
33
|
|
|
for p_num in list(times_dict.keys()): |
34
|
|
|
for p_name in index_list["projection"] + index_list["reconstruction"]: |
35
|
|
|
if p_num == p_name[0]: |
36
|
|
|
times_dict[p_name] = times_dict.pop(p_num) |
37
|
|
|
|
38
|
|
|
self.make_times_figure(times_dict, f"{savepath}/times_chart.png") |
39
|
|
|
|
40
|
|
|
if len(stats_dict["projection"]["max"]): |
41
|
|
|
self.make_stats_graphs(stats_dict["projection"], index_list["projection"], "Projection Stats", |
42
|
|
|
f"{savepath}/projection_stats.png") |
43
|
|
|
if len(stats_dict["reconstruction"]["max"]): |
44
|
|
|
self.make_stats_graphs(stats_dict["reconstruction"], index_list["reconstruction"], "Reconstruction Stats", |
45
|
|
|
f"{savepath}/reconstruction_stats.png") |
46
|
|
|
|
47
|
|
|
|
48
|
|
|
|
49
|
|
|
|
50
|
|
|
@staticmethod |
51
|
|
|
def make_stats_table(stats_dict, index_list, savepath): |
52
|
|
|
stats_dict_copy = {} |
53
|
|
|
for space, value in stats_dict.items(): |
54
|
|
|
stats_dict_copy[space] = value.copy() |
55
|
|
|
for stat in list(stats_dict["projection"].keys()): |
56
|
|
|
if all(value is None for value in stats_dict["projection"][stat]) and all(value is None for value in stats_dict["reconstruction"][stat]): |
57
|
|
|
del stats_dict_copy["projection"][stat] |
58
|
|
|
del stats_dict_copy["reconstruction"][stat] |
59
|
|
|
p_stats = pd.DataFrame(stats_dict_copy["projection"], index_list["projection"]) |
60
|
|
|
r_stats = pd.DataFrame(stats_dict_copy["reconstruction"], index_list["reconstruction"]) |
61
|
|
|
all_stats = pd.concat([p_stats, r_stats], keys=["Projection", "Reconstruction"]) |
62
|
|
|
all_stats.to_html(savepath) # create table of stats for all plugins |
63
|
|
|
|
64
|
|
|
def make_loop_graphs(self, loop_stats, loop_plugins, savepath): |
65
|
|
|
for i in range(len(loop_stats)): |
66
|
|
|
y = loop_stats[i]["NRMSD"] |
67
|
|
|
|
68
|
|
|
#x = list(range(1, len(loop_stats[i]["RMSD"]) + 1)) |
69
|
|
|
x = [None]*len(y) |
70
|
|
|
for j in range(len(loop_stats[i]["NRMSD"])): |
71
|
|
|
x[j] = f"{j}-{j+1}" |
72
|
|
|
ax = plt.figure(figsize=(11, 9), dpi=320).gca() |
73
|
|
|
ax.xaxis.set_major_locator(MaxNLocator(integer=True)) |
74
|
|
|
#ax.locator_params(axis='x', nbins=j + 1) |
75
|
|
|
ax.grid(True) |
76
|
|
|
plt.plot(x, y) |
77
|
|
|
maxx = j |
|
|
|
|
78
|
|
|
maxy = max(y) |
79
|
|
|
plt.title("NRMSD over loop 0") |
80
|
|
|
text = f"Loop 0 iterates {maxx + 2} times over:\n" |
81
|
|
|
for plugin in loop_plugins[i]: |
82
|
|
|
text += f"{plugin}\n" |
83
|
|
|
plt.xlabel("Iteration") |
84
|
|
|
plt.ylabel("NRMSD") |
85
|
|
|
|
86
|
|
|
plt.text(maxx, maxy, text, ha="right", va="top", bbox=dict(boxstyle="round", facecolor="red", alpha=0.4)) |
87
|
|
|
plt.savefig(f"{savepath}/loop_stats{i}.png", bbox_inches="tight") |
88
|
|
|
|
89
|
|
|
|
90
|
|
|
def make_stats_graphs(self, stats_dict, index_list, title, savepath): |
91
|
|
|
stats_df = pd.DataFrame(stats_dict, index_list) |
92
|
|
|
stats_dict, array_plugins = self._remove_arrays(stats_dict, index_list) |
93
|
|
|
|
94
|
|
|
stats_df_new = pd.DataFrame(stats_dict, index_list) |
95
|
|
|
|
96
|
|
|
colours = ["red", "blue", "green", "black", "purple", "brown"] #max, min, mean, mean std dev, median std dev, NRMSD |
97
|
|
|
|
98
|
|
|
new_index = [] |
99
|
|
|
legend = "" |
100
|
|
|
for ind in stats_df_new.index: |
101
|
|
|
new_index.append(ind[0]) # change x ticks to only be plugin numbers rather than names (for space) |
102
|
|
|
legend += f"{ind}\n" # This will form a key showing the plugin names corresponding to plugin numbers |
103
|
|
|
stats_df_new.index = new_index |
104
|
|
|
fig, ax = plt.subplots(3, 2, figsize=(11, 9), dpi=320, facecolor="lavender") |
105
|
|
|
i = 0 |
106
|
|
|
for row in ax: |
107
|
|
|
for axis in row: |
108
|
|
|
stat = self._stats_key[i] |
109
|
|
|
axis.plot(stats_df_new[stat], "x-", color=colours[i]) |
110
|
|
|
for plugin in array_plugins: # adding 'error' bars for plugins with multiple values |
111
|
|
|
if stats_df[stat][plugin] is not None: |
112
|
|
|
if not np.isnan(stats_df[stat][plugin]).any(): |
113
|
|
|
my_max = max(stats_df[stat][plugin]) |
114
|
|
|
my_min = min(stats_df[stat][plugin]) |
115
|
|
|
middle = (my_max + my_min) / 2 |
116
|
|
|
my_range = my_max - my_min |
117
|
|
|
axis.errorbar(list(stats_df_new.index).index(plugin[0]), middle, yerr=[my_range / 2], capsize=5) |
118
|
|
|
if i == 1: |
119
|
|
|
maxx = len(stats_df_new[stat]) * 1.08 - 1 |
120
|
|
|
maxy = max(stats_df_new[stat]) |
121
|
|
|
axis.text(maxx, maxy, legend, ha="left", va="top", |
122
|
|
|
bbox=dict(boxstyle="round", facecolor="red", alpha=0.4)) |
123
|
|
|
stat = stat.replace("_", " ") |
124
|
|
|
axis.set_title(stat) |
125
|
|
|
axis.grid(True) |
126
|
|
|
i += 1 |
127
|
|
|
fig.suptitle(title, fontsize="x-large") |
128
|
|
|
plt.savefig(savepath, bbox_inches="tight") |
129
|
|
|
|
130
|
|
|
|
131
|
|
|
def make_times_figure(self, times_dict, savepath): |
132
|
|
|
colors = plt.get_cmap('Blues')(np.linspace(0.2, 0.7, len(list(times_dict.keys())))) |
133
|
|
|
fig, ax = plt.subplots() |
134
|
|
|
total = sum(list(times_dict.values())) |
135
|
|
|
ax.pie(list(times_dict.values()), labels=list(times_dict.keys()), autopct=lambda pct: self._get_times_pct(pct, total), |
136
|
|
|
counterclock=False, startangle=90, colors=colors, radius=3, center=(4, 4), wedgeprops={"linewidth": 1, "edgecolor": "white"}) |
137
|
|
|
fig.suptitle("Plugin Times", x=0.1, y=1.4, horizontalalignment="right", fontsize="x-large") |
138
|
|
|
plt.savefig(savepath, bbox_inches="tight") |
139
|
|
|
|
140
|
|
|
@staticmethod |
141
|
|
|
def _get_times_pct(pct, total): |
142
|
|
|
absolute = (pct/100)*total |
143
|
|
|
return f"{round(pct, 1)}%\n{round(absolute, 1)} (s)" |
144
|
|
|
|
145
|
|
|
@staticmethod |
146
|
|
|
def _get_dicts_for_graphs(file): |
147
|
|
|
stats_dict = {} |
148
|
|
|
stats_dict["projection"] = {"max": [], "min": [], "mean": [], "mean_std_dev": [], "median_std_dev": [], |
149
|
|
|
"NRMSD": [], "zeros": [], "zeros%": [], "time (s)": []} |
150
|
|
|
stats_dict["reconstruction"] = {"max": [], "min": [], "mean": [], "mean_std_dev": [], "median_std_dev": [], |
151
|
|
|
"NRMSD": [], "zeros": [], "zeros%": [], "time (s)": []} |
152
|
|
|
|
153
|
|
|
index_list = {"projection": [], "reconstruction": []} |
154
|
|
|
|
155
|
|
|
times_dict = {} |
156
|
|
|
|
157
|
|
|
group = file["stats"] |
158
|
|
|
for space in ("projection", "reconstruction"): |
159
|
|
|
for index, stat in enumerate(["max", "min", "mean", "mean_std_dev", "median_std_dev", "NRMSD", "zeros", "zeros%"]): |
160
|
|
|
for p_num in list(group.keys()): |
161
|
|
|
if group[p_num].attrs.get("pattern") in StatsUtils._pattern_dict[space]: |
162
|
|
|
if f"{p_num}: {group[p_num].attrs.get('plugin_name')}" not in index_list[space]: |
163
|
|
|
index_list[space].append(f"{p_num}: {group[p_num].attrs.get('plugin_name')}") |
164
|
|
|
if group[p_num].ndim == 1: |
165
|
|
|
stats_key = list(group[p_num].attrs.get("stats_key")) |
166
|
|
|
if stat in stats_key: |
167
|
|
|
stats_dict[space][stat].append(group[p_num][stats_key.index(stat)]) |
168
|
|
|
else: |
169
|
|
|
stats_dict[space][stat].append(None) |
170
|
|
|
elif group[p_num].ndim == 2: |
171
|
|
|
stats_key = list(group[p_num].attrs.get("stats_key")) |
172
|
|
|
if stat in stats_key: |
173
|
|
|
stats_dict[space][stat].append(group[p_num][:, stats_key.index(stat)]) |
174
|
|
|
else: |
175
|
|
|
stats_dict[space][stat].append(None) |
176
|
|
|
for p_num in list(group.keys()): |
177
|
|
|
if group[p_num].attrs.get("pattern") in StatsUtils._pattern_dict[space]: |
178
|
|
|
stats_dict[space]["time (s)"].append(group[p_num].attrs.get("time")) |
179
|
|
|
|
180
|
|
|
for plugin in list(group.keys()): |
181
|
|
|
if group[plugin].attrs.get("time") is not None: |
182
|
|
|
times_dict[plugin] = group[plugin].attrs.get("time") |
183
|
|
|
|
184
|
|
|
return stats_dict, index_list, times_dict |
185
|
|
|
|
186
|
|
|
@staticmethod |
187
|
|
|
def _get_dicts_for_loops(file): |
188
|
|
|
if "iterative" in list(file.keys()): |
189
|
|
|
group = file["iterative"] |
190
|
|
|
loop_stats = [] |
191
|
|
|
loop_plugins = [] |
192
|
|
|
for key in list(group.keys()): |
193
|
|
|
loop_stats.append({"NRMSD": list(group[key])}) |
194
|
|
|
loop_plugins.append(group[key].attrs.get("loop_plugins")) |
195
|
|
|
return loop_stats, loop_plugins |
196
|
|
|
else: |
197
|
|
|
return [], [] |
198
|
|
|
|
199
|
|
|
@staticmethod |
200
|
|
|
def _remove_arrays(stats_dict, index_list): |
201
|
|
|
array_plugins = set(()) |
202
|
|
|
for stat in list(stats_dict.keys()): |
203
|
|
|
for index, value in enumerate(stats_dict[stat]): |
204
|
|
|
if isinstance(value, np.ndarray): |
205
|
|
|
stats_dict[stat][index] = stats_dict[stat][index][0] |
206
|
|
|
array_plugins.add(index_list[index]) |
207
|
|
|
return stats_dict, array_plugins |
208
|
|
|
|
209
|
|
|
|
210
|
|
|
|