-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathplot-accuracy.py
295 lines (277 loc) · 12.9 KB
/
plot-accuracy.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
288
289
290
291
292
293
294
295
#!/usr/bin/env python
# -*- coding: utf-8 -*-
"""
===============================================================================
Script 'plot-behavior'
===============================================================================
This script plots behavioral data for the CAPD attention switching
pupillometry experiment.
"""
# @author: Dan McCloy (drmccloy@uw.edu)
# Created on Fri Aug 10 13:43:29 PDT 2018
# License: BSD (3-clause)
import yaml
import os.path as op
import numpy as np
import pandas as pd
from pandas.api.types import CategoricalDtype
from expyfun.analyze import dprime
import matplotlib.pyplot as plt
def probit(x):
from scipy.stats import norm
return norm.ppf(x)
pd.set_option('display.max_columns', 50)
pd.set_option('display.width', None)
savefig = True
do_probit = True
# file I/O
indir = 'posthocs'
datadir = 'processed-data'
outdir = op.join('figures', 'manuscript')
styledir = 'styles'
# figure params
with open(op.join(styledir, 'figure-params.yaml'), 'r') as fp:
fig_params = yaml.load(fp)
signifcol = fig_params['signif_color']
prp = fig_params['prp'] # purple
pch = fig_params['pch'] # peach
cyn = fig_params['cyn'] # cyan
blu = fig_params['blu'] # blue
mgn = fig_params['mgn'] # magenta
tll = fig_params['tll'] # teal
orn = fig_params['orn'] # orange
ylo = fig_params['ylo'] # yellow
red = fig_params['red'] # red
grn = fig_params['grn'] # green
gry = fig_params['gry'] # gray
dgy = fig_params['dgy'] # dark gray
colors = {'spatial': orn, 'non-spatial': tll, 'mixed': mgn, 'ctrl': prp,
'ldiff': blu, 'maintain': cyn, 'switch': blu, 'main effect': gry,
'control': prp, 'listening difficulty': blu}
model_summary = pd.read_csv(op.join(indir, 'model-summary.csv'))
main_effects = pd.read_csv(op.join(indir, 'main-effects-estimates.csv'))
interactions = pd.read_csv(op.join(indir, 'interaction-estimates.csv'))
slot_df = pd.read_csv(op.join(datadir, 'behavioral-data-by-slot.tsv'),
sep='\t')
# maineff_posthocs = pd.read_csv('main-effects-posthocs.csv')
# interact_posthocs = pd.read_csv('interaction-posthocs.csv')
# parse model summary
model_summary['terms'] = (model_summary[['ldiff', 'attn', 'space']]
.applymap(lambda x: isinstance(x, str))
.apply('sum', axis=1))
model_summary['effect'] = model_summary['terms'].map({0: 'intercepts',
1: 'main effect',
2: 'interaction'})
# merge dataframes
main_effects['effect'] = 'main effect'
interactions['effect'] = 'interaction'
estimates = pd.concat([main_effects, interactions])
col_order = ['effect', 'grouping', 'group', 'variable', 'level', 'token',
'prob', 'asymp.LCL', 'asymp.UCL', 'SE', 'df']
estimates = estimates[col_order]
# factor order
categories = ['control', 'ctrl', 'ldiff', 'listening\ndifficulty', 'mixed',
'non-\nspatial', 'non-spatial', 'spatial', 'maintain', 'switch']
dt = CategoricalDtype(categories=categories, ordered=True)
estimates['level'] = estimates['level'].astype(dt)
estimates['group'] = estimates['group'].astype(dt)
# relative error
estimates['probit'] = probit(estimates['prob'])
if do_probit:
estimates['lerr'] = np.abs(estimates['probit'] -
probit(estimates['asymp.LCL']))
estimates['uerr'] = np.abs(estimates['probit'] -
probit(estimates['asymp.UCL']))
else:
estimates['lerr'] = np.abs(estimates['prob'] - estimates['asymp.LCL'])
estimates['uerr'] = np.abs(estimates['prob'] - estimates['asymp.UCL'])
# style setup
style_files = ('font-libertine.yaml', 'font-latex.yaml', 'garnish.yaml')
plt.style.use([op.join(styledir, sf) for sf in style_files])
# undo setting from garnish.yaml
plt.style.use({'axes.spines.left': True, 'xtick.labelsize': 8,
'ytick.labelsize': 8})
# renamers
x_axis_labeller = {'attn': 'Attention condition', 'space': 'Spatial condition',
'ldiff': 'Participant group', 'maintain': 'maint.',
'listening\ndifficulty': 'lis.\ndiffic.'}
legend_labeller = {'ldiff': 'listening difficulty',
'ctrl': 'age-matched control'}
# kwargs
signif_kwargs = dict(xytext=(0, 4), textcoords='offset points', ha='center',
va='bottom', fontsize=9)
# axis lims & dodge
x_dodge = (-0.04, 0, 0.04)
x_gap = 1.2
x_pad = 0.15
zero = 2e-2 if do_probit else 0.
fzero = 1.5e-2 if do_probit else 0.
ylims = dict(target=(0.75, 0.95), foil=(fzero, 0.12), neither=(zero, 0.1))
yticks = dict(target=(0.75, 0.8, 0.85, 0.9, 0.95), foil=(zero, 0.05, 0.1),
neither=(zero, 0.05, 0.1), dprime=range(5))
yls = probit(list(ylims.values())) if do_probit else list(ylims.values())
ylims.update(dict(dprime=(0, 4))) # add this after the probit in yls
width_ratios = np.array([2, 2, 3]) * 2 + x_gap + 1.5 * x_pad
height_ratios = [np.diff(yl)[0] for yl in yls]
height_ratios += [np.mean(height_ratios)] # height of dprime row
# init figure
fig, axs = plt.subplots(4, 3, figsize=(6.5, 4.5), sharey='row', sharex='col',
gridspec_kw=dict(width_ratios=width_ratios,
height_ratios=height_ratios))
fig.subplots_adjust(left=0.1, right=0.95, bottom=0.15, top=0.95, hspace=0.3,
wspace=0.2)
# plot
for ax_row, token in zip(axs, ('target', 'foil', 'neither', 'dprime')):
for ax, variable in zip(ax_row, ('attn', 'ldiff', 'space')):
if token == 'dprime':
column_dict = dict(attn='attn', ldiff='group', space='s_cond')
var = column_dict[variable]
# main effect
cols = ['subj', var] + list('hmfc')
hmfc = slot_df[cols].groupby(cols[:-4]).aggregate('sum')
hmfc['dprime'] = hmfc.apply(dprime, axis=1)
hmfc_main = hmfc.reset_index().sort_values([var, 'subj'])
# interaction
ivar = estimates.loc[(estimates['variable'] == variable) &
(estimates['effect'] == 'interaction'),
'grouping'].iat[0]
inter_var = column_dict[ivar]
cols = ['subj', var, inter_var] + list('hmfc')
hmfc = slot_df[cols].groupby(cols[:-4]).aggregate('sum')
hmfc['dprime'] = hmfc.apply(dprime, axis=1)
hmfc_int = hmfc.reset_index().sort_values([var, inter_var, 'subj'])
# combine
gb_main = hmfc_main[[var, 'dprime']].groupby(var)
gb_int = hmfc_int[[var, inter_var, 'dprime']
].groupby([var, inter_var])
means = pd.concat([gb_main.aggregate('mean'),
gb_int.aggregate('mean')])
stds = pd.concat([gb_main.aggregate('std'),
gb_int.aggregate('std')])
# plot
n = len(gb_main)
m = len(gb_int) // n # integer divide OK; guaranteed no remainder
w = 0.25
x = (np.repeat(np.arange(n * 2), np.repeat((1, m), n)) +
np.array([0] * n + [x_gap - 1] * n * m))
offsets = np.concatenate((np.zeros(n),
np.tile(np.arange(m)*w - (m-1)*w/2, n)))
x += offsets
keys = (np.repeat(['main effect'], n),
np.tile(hmfc_int[inter_var].unique().astype(str), n))
this_colors = [colors[k] for k in np.concatenate(keys)]
ax.bar(x, means['dprime'], yerr=stds['dprime'], width=w,
color=this_colors, error_kw=dict(size=0.1, capsize=1.5,
alpha=0.3))
else:
df = estimates.loc[(estimates['token'] == token) &
(estimates['variable'] == variable)]
xticks = []
xticklabels = []
# plot main effect
this_df = df.loc[df['effect'] == 'main effect'
].sort_values(by='level')
x = np.arange(this_df.shape[0])
y = 'probit' if do_probit else 'prob'
ax.errorbar(x=x, y=this_df[y],
yerr=this_df[['lerr', 'uerr']].values.T,
color=colors['main effect'], label=None)
xticks.extend(x)
xticklabels.extend(this_df['level'].unique())
# signif
locator = ((model_summary['truth'] == token) &
(model_summary['effect'] == 'main effect') &
(model_summary[this_df['variable'].iloc[0]
].map(lambda x: isinstance(x, str))))
signif = model_summary.loc[locator]
for ss, s in enumerate(signif['signif']):
if isinstance(s, str):
ax.annotate(s, xy=(x[ss:(ss+2)].mean(), this_df[y].mean()),
color=colors['main effect'], **signif_kwargs)
# plot interaction
this_df = df.loc[df['effect'] == 'interaction'
].sort_values(by='group')
for ig, group in enumerate(this_df['group'].unique()):
this_line = this_df.loc[this_df['group'] == group
].sort_values(by='level')
xx = (x.max() + x_gap + np.arange(this_line.shape[0]) +
x_dodge[ig])
label = this_line['group'].iloc[0]
ax.errorbar(x=xx, y=this_line[y],
yerr=this_line[['lerr', 'uerr']].values.T,
color=colors[group],
label=legend_labeller.get(label, label))
xticks.extend(xx - x_dodge[ig])
xticklabels.extend(this_line['level'].unique())
# signif
locator = ((model_summary['truth'] == token) &
(model_summary['effect'] == 'interaction') &
(model_summary[this_line['variable'].iloc[0]
].map(lambda x: isinstance(x, str))) &
(model_summary[this_line['grouping'].iloc[0]
].map(lambda x: isinstance(x, str))))
signif = model_summary.loc[locator]
for ss, s in enumerate(signif['signif']):
this_x = xx[ss:(ss+2)] if this_line.shape[0] > 2 else xx
if isinstance(s, str):
ax.annotate(s, xy=(this_x.mean(), this_line[y].mean()),
color=colors[group], **signif_kwargs)
# garnish
ax.set_xticks(xticks)
ax.set_xticklabels([x_axis_labeller.get(xtl, xtl) for xtl in
xticklabels])
if do_probit:
if token == 'dprime':
ax.set_ylim(ylims[token])
else:
ax.set_ylim(probit(ylims[token]))
ax.set_yticks(probit(yticks[token]))
ax.set_yticklabels([f'{y:.2f}' for y in yticks[token]])
else:
ax.set_ylim(ylims[token])
ax.set_yticks(yticks[token])
# x labels
if token == 'dprime':
ax.set_xlabel(x_axis_labeller[variable])
else:
# remove x tickmarks
ax.tick_params(axis='x', bottom=False)
# make room for subplot tags
xlim = np.array(ax.get_xlim()) - np.array((x_pad, -0.5*x_pad))
ax.set_xlim(xlim)
# subplot tags
tag = list('ABCDEFGHIJKL')[np.where(axs.ravel() == ax)[0][0]]
ax.annotate(f'({tag})', xy=(0, 0), xytext=(8, 2),
xycoords='axes fraction', textcoords='offset points',
fontsize=9, fontweight='bold', va='bottom', ha='center', color=dgy)
# legend
if token == 'target':
# the h[0] hack is to suppress the errorbars in the legend
hh, ll = ax.get_legend_handles_labels()
legend = ax.legend(handles=[h[0] for h in hh], labels=ll,
loc='upper right', bbox_to_anchor=(1, 1.3),
fontsize=8, frameon=False, markerfirst=False,
labelspacing=0.3)
# row labels
row_labels = dict(dprime='d-prime')
ax.annotate(row_labels.get(token, token).upper(), xy=(1, 0.5),
xytext=(8, 0), xycoords='axes fraction',
textcoords='offset points', rotation=270, ha='left',
va='center', fontsize=10)
# y axis label
if token == 'dprime':
ax_row[0].set_ylabel('d′', labelpad=10)
elif token == 'foil':
ps = ', probit-scaled' if do_probit else ''
label = r'$P\thinspace(\mathdefault{{button\ press}})${}'.format(ps)
# label = r'$\Pr(\mathdefault{{button\ press}})${}'.format(ps)
# label = 'ℙ(button press){}'.format(ps)
ax_row[0].set_ylabel(label, labelpad=10)
fig.align_labels()
if savefig:
p = '-probit' if do_probit else ''
fname = f'accuracy-figure{p}.pdf'
fig.savefig(op.join(outdir, fname))
else:
plt.ion()
plt.show()