-
Notifications
You must be signed in to change notification settings - Fork 0
/
plot_utils.py
287 lines (240 loc) · 11.6 KB
/
plot_utils.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
import numpy as np
from matplotlib import pyplot as plt
from utils import eval_ensemble, multi_roc_sigeffs
def get_sic_curves_multirun(ax, multi_tprs, multi_fprs, y_test,
max_rel_err=0.2, label="", color=None,
linestyle=None):
"""Plot single SIC curve including errror bands for several runs.
Errorbands are defined as inner 68% of SIC value distribution of the
independent runs
Args:
ax (matplotlib.pyplot.axis): Matplotlib axis object to plot on.
multi_tprs (numpy.ndarray): Array of shape
(num_runs, num_tpr_values) containing the TPR values for each run
multi_fprs (numpy.ndarray): Array of shape
(num_runs, num_fpr_values) containing the FPR values for each run
y_test (numpy.ndarray): Truth labels of test set.
max_rel_err (float, optional): Maximum relative error allowed
w.r.t. number of background events up to which we still plot the
SIC curve.
label (str, optional): Label for the plot legend.
Returns:
matplotlib.pyplot.axis: Matplotlib axis object containing plot of
median SIC with error bands.
"""
multi_sics = multi_tprs/np.sqrt(multi_fprs)
median_tprs = np.median(multi_tprs, axis=0)
median_fprs = np.median(multi_fprs, axis=0)
median_sics = np.median(multi_sics, axis=0)
upper_sics = np.percentile(multi_sics, 84, axis=0)
lower_sics = np.percentile(multi_sics, 16, axis=0)
plot_mask = (
(1/np.sqrt(median_fprs*y_test[y_test == 0].shape[0])) < max_rel_err
)
if linestyle is None:
linestyle = 'solid'
else:
assert linestyle in ['solid', 'dashed', 'dashdot', 'dotted'], (
"Error! `linestyle` must be one of 'solid', 'dashed', 'dashdot' "
"or 'dotted'."
)
if color is None:
p = ax.plot(median_tprs[plot_mask], median_sics[plot_mask],
label=label, linestyle=linestyle)
else:
p = ax.plot(median_tprs[plot_mask], median_sics[plot_mask],
label=label, color=color, linestyle=linestyle)
ax.fill_between(median_tprs[plot_mask], upper_sics[plot_mask],
lower_sics[plot_mask], alpha=0.2,
facecolor=p[0].get_color(),
edgecolor=None)
return ax
def plot_sic_curves(tpr_list, fpr_list, y_test_list, max_rel_err=0.2,
color_list=None, title=None, linestyles=None,
xlabel="TPR", ylabel="SIC", out_filename=None, labels=None,
legend_loc="upper right", max_y=None):
"""Plot single SIC curve including errror bands for several runs.
Args:
tpr_list (list of numpy.ndarray): List of arrays of shape
(num_tpr_values) containing the TPR values for each run
fpr_list (list of numpy.ndarray): List of arrays of shape
(num_fpr_values) containing the FPR values for each run
y_test_list (list of numpy.ndarray): List of arrays containing the
truth labels of the test set for each study that should be plotted.
max_rel_err (float, optional): Maximum relative error allowed
w.r.t. number of background events up to which we still plot the
SIC curve.
xlabel (str, optional): String containing x axis label.
Default is 'TPR'.
ylabel (str, optional): String containing y axis label.
Default is 'SIC'.
out_filename (str, optional): String describing the filename under
which the plot should be saved.
labels (NoneType or list of str, optional):
List of labels describing the different training runs that should
be plotted.
legend_loc (str, optional): String describing the location of the
legend. Default is 'upper right'.
max_y (NoneType or float, optional): Maximum value of y axis.
Default is None.
"""
f, ax = plt.subplots()
if (labels is not None) and (len(labels) != len(tpr_list)):
raise ValueError(("Error! `labels` must have same length as "
"`tpr_val_list` and `fpr_val_list`"))
if (color_list is not None) and (len(color_list) != len(tpr_list)):
raise ValueError(("Error! `color_list` must have same length as "
"`tpr_val_list` and `fpr_val_list`"))
if (linestyles is not None) and (len(linestyles) != len(tpr_list)):
raise ValueError(("Error! `color_list` must have same length as "
"`tpr_val_list` and `fpr_val_list`"))
if labels is None:
labels = [None]*len(tpr_list)
if color_list is None:
color_list = [None]*len(tpr_list)
if linestyles is None:
linestyles = [None]*len(tpr_list)
for i in range(len(tpr_list)):
get_sic_curves_multirun(ax, tpr_list[i], fpr_list[i],
y_test_list[i], color=color_list[i],
linestyle=linestyles[i],
max_rel_err=max_rel_err, label=labels[i])
plt.xlabel(xlabel)
plt.xlim(0, 1)
plt.ylim(0, max_y)
plt.ylabel(ylabel)
plt.legend(loc=legend_loc, frameon=False)
if title is not None:
ymin = 0
ymax = max_y
xmin = 0
xmax = 1
a = 0.03
plt.text(xmin + a * (xmax-xmin), ymin + (1-a) * (ymax-ymin),
title, size=plt.rcParams['axes.labelsize'],
color='black', horizontalalignment='left',
verticalalignment='top')
if out_filename is not None:
plt.savefig(out_filename)
plt.show()
plt.close()
def plot_sic_curve_comparison(model_list, data, out_filename=None,
model_types=None,
color_list=None, linestyles=None,
labels=None, xlabel="TPR", ylabel="SIC",
legend_loc="upper right", max_rel_err=0.2,
max_y=None, title=None):
"""A simple wrapper function of `get_sic_curves_multirun`
This function allows to plot multiple SIC curves (e.g. when comparing
several runs with different training settings) including their
respective error bands on one plot.
Args:
model_list (list of list): List of models for which the SIC curves
should be plotted. The structure of the list should be
as follows: The first index refers to a separate "study" (e.g.
using vanilla input settings or 10 gaussian noise features added).
The second index is a single run of this study. The third index
refers to the specific model within the ensemble for that run.
data (dict or list of dict): Dictionary or list of dictionaries
(one for each study) containing the data to be used for evaluation.
Should at least contain the keys "x_test" and "y_test".
out_filename (str, optionaö): String describing the filename under
which the plot should be saved.
model_types (NoneType or list of str, optional): List of strings
describing the model types of the models in `model_list`.
labels (NoneType or list of str, optional):
List of labels describing the different training runs that should
be plotted for comparison. The length of the list *must* be equal
to the number of runs or a `ValueError` is raised.
xlabel (str, optional): String containing x axis label.
Default is 'TPR'.
ylabel (str, optional): String containing y axis label.
Default is 'SIC'.
legend_loc (str, optional): Location parameter of
`matplotlib.pyplot.legend`. Default is "upper right".
max_rel_err (float, optional): Maximum relative error allowed
w.r.t. number of background events up to which we still plot the
SIC curve.
show (bool, optional): Boolean deciding whether plots should
also be directly shown instead of just being stored to disk.
Default is True.
max_y (NoneType or float, optional): Numeric value
defining the maximum y value up to which the SIC is plotted.
Default is None, leading to `matplotlib` figuring out the limit
by itself.
"""
if model_types is not None:
assert len(model_list) == len(model_types), (
"Error! `model_types` must have same length as `model_list`"
)
else:
print(("No model types provided. Assuming all models are scikit-learn "
"HistGradientBoostingClassifier instances."))
model_types = ["HGB"]*len(model_list)
if type(data) is not list:
data = [data]*len(model_list)
else:
assert len(model_list) == len(data), (
"Error! List of `data` must have same length as `model_list`"
)
tpr_val_list = []
fpr_val_list = []
for i in range(len(model_list)):
if model_types is None:
model_types = ["HGB"]*len(model_list[i])
full_preds_tmp = eval_ensemble(
model_list[i], data[i],
model_type=model_types[i],
)
tpr_vals_tmp, fpr_vals_tmp = multi_roc_sigeffs(full_preds_tmp,
data[i]["y_test"])
tpr_val_list.append(tpr_vals_tmp)
fpr_val_list.append(fpr_vals_tmp)
y_test_list = [data[i]["y_test"] for i in range(len(data))]
plot_sic_curves(tpr_val_list, fpr_val_list, y_test_list,
max_rel_err=max_rel_err, xlabel=xlabel, ylabel=ylabel,
out_filename=out_filename, labels=labels,
color_list=color_list, linestyles=linestyles, title=title,
legend_loc=legend_loc, max_y=max_y)
def plot_losses(losses_to_plot, out_file="./losses.pdf", labels=None,
show=True, xlims=None, ylims=None):
"""
Plot the given list of loss values as a line graph and save the figure as
a PDF file.
Args:
losses_to_plot (list or array): A list of arrays or lists, where each
element represents a particular set of loss values (e.g. for
different models or training vs validation losses etc.).
out_file (str): The path to save the output PDF file.
Default is "./losses.pdf".
labels (list, optional): A list of strings, where each element
represents the label for a particular model's loss values.
If None (default), no labels will be shown.
show (bool): A boolean indicating whether to show the plot in a window.
Default is True.
xlims (tuple, optional): A tuple of two values representing the lower
and upper limits of the x-axis.
ylims (tuple, optional): A tuple of two values representing the lower
and upper limits of the y-axis.
Raises:
ValueError: If the lengths of the labels parameter and input
loss list/array do not match.
"""
if labels is not None and len(losses_to_plot) != len(labels):
raise ValueError(
"Lengths of labels parameter and input loss list/array must match!"
)
if labels is None:
labels = [None]*len(losses_to_plot)
for i, loss in enumerate(losses_to_plot):
xvals = np.arange(1, len(loss)+1)
plt.plot(xvals, loss, label=loss[i])
plt.xlabel("Iteration")
plt.ylabel("val loss")
plt.xlim(xlims)
plt.ylim(ylims)
plt.legend(loc="upper right")
plt.savefig(out_file, bbox_inches="tight")
if show:
plt.show()
plt.close()