-
Notifications
You must be signed in to change notification settings - Fork 0
/
sim_data_load.py
278 lines (218 loc) · 12.1 KB
/
sim_data_load.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
import os
from shutil import copyfile
from pickle_util import PickleUtil
from matplotlib import pyplot as plt
import seaborn as sns
import pandas as pd
from scipy import stats
import numpy as np
class SimDataUtil:
def __init__(self, data_dir):
self.data_directory = data_dir
self.data_by_user = self.load_data()
self.user_numbers = set(self.data_by_user.keys())
self.make_data_frame()
self.plot_colors = ["#0000ff", "#00aa00", "#aa0000", "#ff7700", "#aa00aa"]
def load_data(self):
data_by_user = dict()
for path, dir, files in os.walk(self.data_directory):
user_dir = path[(len(self.data_directory)+1):]
if user_dir is not "":
if len(files) > 0:
user_data = dict()
for file in files:
file_data = PickleUtil(os.path.join(path, file)).safe_load()
if "dist_id" in file:
user_data["click_dist"] = file_data
continue
else:
data_value_names = {'errors', 'selections', 'characters', 'presses_sel', 'presses_char',
'presses_word', 'kde_mses', 'kde'}
self.param_names = set(file_data.keys()) - data_value_names
params = tuple(file_data[name] for name in self.param_names)
# print(self.param_names)
# print(params)
user_data[params] = file_data
data_by_user[int(user_dir)] = user_data
return data_by_user
def plot_across_user(self, metric, params=None, trends=False, log=False, legend=None):
if isinstance(metric, str):
metric = [metric]
for m in metric:
dep_vars = []
ind_vars = []
for user in self.user_numbers:
user_data = self.data_by_user[user]
if params is None:
params = list(user_data.keys())[1]
else:
if params not in user_data.keys():
raise KeyError("Parameters are not in saved data")
if m not in user_data[params].keys():
raise KeyError("Metric is not in saved data")
dep_vars += [user_data[params][m]]
if "attribute" in user_data[params]:
if "supercloud_results_20" in self.data_directory:
ind_vars += [user_data[params]["attribute"]*4]
else:
ind_vars += [user_data[params]["attribute"]]
else:
ind_vars += [user]
data_points = list(zip(ind_vars, dep_vars))
data_points.sort()
ind_vars, dep_vars = zip(*data_points)
if np.array(dep_vars[0]).size == 1:
if log:
dep_vars = np.log(dep_vars)
plt.plot(ind_vars, dep_vars)
else:
colors = np.arange(len(ind_vars))
colors = colors/(len(ind_vars))
if trends:
avg_grads = []
for i, line in enumerate(dep_vars):
label = ind_vars[i]
x_values = np.arange(line.size)+1
line_norm = line-np.min(line)
smoothing = 20
smooth_x = x_values[smoothing:-smoothing]
smooth_line = np.convolve(line_norm, np.ones((smoothing,))/smoothing)[smoothing:len(x_values)-smoothing]
# plt.plot(smooth_x, np.gradient(smooth_line), color=(min(1, (1-colors[i])*2),0,min(1, colors[i]*2)))
avg_grads += [-np.average(np.gradient(smooth_line))]
plt.plot(ind_vars[1:], avg_grads[1:] / np.max(avg_grads))
else:
plt.figure(figsize=(10, 12))
x_pos = np.log(max([s.size for s in dep_vars])*1.05)
plt.xlim(np.log(4), x_pos*1.1)
max_y = -float("inf")
for i, line in enumerate(dep_vars[1:]):
label = ind_vars[i]
x_values = np.arange(line.size) + 1
if log:
line = np.log(line)
x_values = np.log(x_values)
max_y = max(max_y, max(line))
plt.plot(x_values[5:], line[5:],
color=(min(1, (1 - colors[i]) * 2), 0, min(1, colors[i] * 2)))
y_pos = line[-1] - 0.0005
plt.text(x_pos, y_pos, str(label), fontsize=12, color=(min(1, (1 - colors[i]) * 2), 0, min(1, colors[i] * 2)))
if legend is not None:
plt.text(x_pos/1.075, max_y+abs(max_y*0.0075), legend["multi"], fontsize=11)
if legend is not None:
plt.title(legend["title"])
plt.xlabel(legend["x"])
plt.ylabel(legend["y"])
plt.show()
def make_data_frame(self):
average_data = {}
num_users = len(self.user_numbers)
for user in self.user_numbers:
user_data = self.data_by_user[user]
for param in user_data:
if param != "click_dist":
if param not in average_data:
average_data[param] = {'errors': [], 'selections': [], 'characters': [], 'presses_word': [],
'presses_char': []}
print(user_data[param].keys())
for data_label in ['errors', 'selections', 'characters', 'presses_char', 'presses_word']:
average_data[param][data_label] += user_data[param][data_label]
data_labels = {'errors', 'selections', 'characters', 'presses_word', 'presses_char'}
param_name_dict = {'num_words': "Word Predictions Max Count", 'order': 'Keyboard Layout',
'words_first': "Words First", 'attribute': 'Attribute', "false_positive": "False Positive Rate",
"delay": "Delay", "scan_delay": "Scanning Delay", "easy_corpus": "Corpus"}
var_name_dict = {'selections': "Selections/Min", 'characters': "Characters/Min",
'presses_char': "Clicks/Character",
'presses_word': "Clicks/Word", 'errors': "Error Rate"}
long_form_data = []
for params in average_data.keys():
param_names = [param_name_dict[name] for name in self.param_names]
observation = dict(zip(param_names, params))
param_data = average_data[params]
num_observations = len(param_data['errors'])
for obs in range(num_observations):
for data_label in data_labels:
observation[var_name_dict[data_label]] = param_data[data_label][obs]
long_form_data += [observation.copy()]
df = pd.DataFrame(long_form_data)
self.DF = df
# self.DF["Adjusted Scanning Delay"] = self.DF["Scanning Delay"] != 0.5
self.DF["Words First | Alpha Sorted"] = self.DF["Words First"]
self.DF["Words First | Freq Sorted"] = self.DF["Words First"]
def plot_across_params(self):
ind_var_name = "Scanning Delay"
for dep_var_name in ['Error Rate', 'Selections/Min', 'Characters/Min', 'Clicks/Word',
'Clicks/Character', 'Error Rate']:
DF = self.DF
pd.set_option('display.max_columns', 500)
fig, ax = plt.subplots()
fig.set_size_inches(10, 8)
sns.set(font_scale=1.5, rc={"lines.linewidth": 3})
sns.set_style({'font.serif': 'Helvetica'})
if ind_var_name == "Word Predictions Max Count":
sns.lineplot(x=ind_var_name, y=dep_var_name, hue="Words First | Freq Sorted",
palette=sns.cubehelix_palette(2, start=2, rot=0.2, dark=.2, light=.7, reverse=True),
data=DF[DF["Keyboard Layout"] == "sorted"], ci="sd", ax=ax)
sns.lineplot(x=ind_var_name, y=dep_var_name, hue="Words First | Alpha Sorted",
palette=sns.cubehelix_palette(2, start=3, rot=0.2, dark=.2, light=.7, reverse=True),
data=DF[DF["Keyboard Layout"] == 'default'], ci="sd", ax=ax)
elif ind_var_name == "Left Context":
sns.violinplot(x=ind_var_name, y=dep_var_name, hue="Left Context", data=DF, inner="points", figsize=(10, 8))
lc_false = self.DF[self.DF[ind_var_name] == True][dep_var_name]
lc_false_mean = np.mean(lc_false.values)
plt.axhline(lc_false_mean, linestyle='--', color=(0.4, 0.4, 0.9))
lc_true = self.DF[self.DF[ind_var_name] == False][dep_var_name]
lc_true_mean = np.mean(lc_true.values)
plt.axhline(lc_true_mean, linestyle='--', color=(0.9, 0.9, 0.4))
t_value, p_value = stats.ttest_ind(lc_false, lc_true, equal_var=False)
plt.text(0.9, -.1, 'p-value: '+str(round(p_value, 2)), ha='center', va='center', transform=ax.transAxes)
elif ind_var_name == "Easy Corpus":
sns.violinplot(x=ind_var_name, y=dep_var_name,
data=DF, ci="sd")
else:
sns.lineplot(x=ind_var_name, y=dep_var_name, color="cadetblue",
data=DF, ci="sd", ax=ax)
sns.lineplot(x=ind_var_name, y=dep_var_name, color="darkslategrey",
data=DF, ax=ax)
plt.title("Row Col: "+ dep_var_name+" vs. "+ind_var_name)
plt.show()
# break
def order_data(dir):
click_dists = []
if not os.path.exists(os.path.join(dir, "ordered_data")):
os.mkdir(os.path.join(dir, "ordered_data"))
for path, __, files in os.walk(dir):
for file in files:
if "dist_id" in file:
click_dist = PickleUtil(os.path.join(path, file)).safe_load()
if click_dist not in click_dists:
click_dists += [click_dist]
# os.mkdir(os.path.join(dir, os.path.join("ordered_data", str(click_dists.index(click_dist)))))
print(path)
plt.plot(click_dist)
plt.show()
# if "npred" in file:
# new_dir = os.path.join(dir, os.path.join("ordered_data", str(click_dists.index(click_dist))))
# if not os.path.exists(os.path.join(new_dir, file)):
# copyfile(os.path.join(path, file), os.path.join(new_dir, file))
# else:
# new_dir = os.path.join(dir, os.path.join("ordered_data", str(len(click_dists))))
# if not os.path.exists(new_dir):
# os.mkdir(new_dir)
# copyfile(os.path.join(path, file), os.path.join(new_dir, file))
def main():
# sdu = SimDataUtil("simulations/increasing_variance/supercloud_results")
# plot_legend = {"title": "MSE Improvement of Nomon KDE vs Click Distribution Variance", "x": "Standard Deviation (# hist bins)",
# "y": "Average (-) Gradient of MSE Over Presses"}
# sdu.plot_across_user("kde_mses", (3, 0.008), trends=True, log=False, legend=plot_legend)
sdu = SimDataUtil("simulations/scan_delay/supercloud_results")
sdu.plot_across_params()
# plot_legend = {"title": "MSE of Nomon KDE vs Bimodal Distance",
# "x": "log Number of Presses ( log(presses) )",
# "y": "MSE of KDE", "multi": "Modal Separation\n (# hist bins)"}
#
# sdu.plot_across_user("kde_mses", (3, 0.008), trends=False, log=False, legend=plot_legend)
# sdu.plot_across_user(["selections", "presses"], (3, 0.008))
# sdu.plot_across_user("errors", (3, 0.008))
# sdu.plot_across_user(["kde_mses", "errors"], (3, 0.008), trends=True)
if __name__ == '__main__':
main()