-
Notifications
You must be signed in to change notification settings - Fork 0
/
propositional_analysis.jl
246 lines (209 loc) · 9.78 KB
/
propositional_analysis.jl
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
using Pkg
Pkg.activate(".")
using MLJ, ModalDecisionTrees
using SoleDecisionTreeInterface, Sole, SoleData
using CategoricalArrays
using DataFrames, JLD2, CSV
using Audio911
using Random
using StatsBase, Catch22
using Test
using Plots
# ---------------------------------------------------------------------------- #
# settings #
# ---------------------------------------------------------------------------- #
experiment = :Pneumonia
# experiment = :Bronchiectasis
# experiment = :COPD
# experiment = :URTI
# experiment = :Bronchiolitis
features = :catch9
# features = :minmax
# features = :custom
# loadset = false
loadset = true
scale = :semitones
# scale = :mel_htk
sr = 8000
featset = ()
# featset = (:mfcc,)
# featset = (:f0,)
# featset = (:mfcc, :f0)
audioparams = (
sr = sr,
nfft = 512,
mel_scale = scale, # :mel_htk, :mel_slaney, :erb, :bark, :semitones, :tuned_semitones
mel_nbands = scale == :semitones ? 14 : 26,
mfcc_ncoeffs = scale == :semitones ? 7 : 13,
mel_freqrange = (300, round(Int, sr / 2)),
mel_dbscale = :mfcc in featset ? false : true,
audio_norm = true,
)
memguard = false;
# memguard = true;
n_elems = 15;
avail_exp = [:Pneumonia, :Bronchiectasis, :COPD, :URTI, :Bronchiolitis]
@assert experiment in avail_exp "Unknown type of experiment: $experiment."
findhealthy = y -> findall(x -> x == "Healthy", y)
ds_path = "/datasets/respiratory_Healthy_" * String(experiment)
findsick = y -> findall(x -> x == String(experiment), y)
filename = "/datasets/itadata2024_" * String(experiment) * "_files"
memguard && begin filename *= string("_memguard") end
destpath = "results/propositional"
jld2file = destpath * "/itadata2024_" * String(experiment) * "_" * String(scale) * ".jld2"
dsfile = destpath * "/ds_test_" * String(experiment) * "_" * String(scale) * ".jld2"
color_code = Dict(:red => 31, :green => 32, :yellow => 33, :blue => 34, :magenta => 35, :cyan => 36);
r_select = r"\e\[\d+m(.*?)\e\[0m";
# ---------------------------------------------------------------------------- #
# prepare dataset for training #
# ---------------------------------------------------------------------------- #
d = jldopen(string((@__DIR__), ds_path, ".jld2"))
x, y = d["dataframe_validated"]
@assert x isa DataFrame
close(d)
memguard && begin
cat2 = round(Int, length(y)/2)
indices = [1:n_elems; cat2:cat2+n_elems-1]
x = x[indices, :]
y = y[indices]
end
freq = round.(Int, afe(x[1, :audio]; featset=(:get_only_freqs), audioparams...))
catch9_f = ["max", "min", "mean", "med", "std", "bsm", "bsd", "qnt", "3ac"]
variable_names = vcat([
vcat(
["\e[$(color_code[:yellow])m$j(mel$i=$(freq[i])Hz)\e[0m" for i in 1:audioparams.mel_nbands],
:mfcc in featset ? ["\e[$(color_code[:red])m$j(mfcc$i)\e[0m" for i in 1:audioparams.mfcc_ncoeffs] : String[],
:f0 in featset ? ["\e[$(color_code[:green])m$j(f0)\e[0m"] : String[],
"\e[$(color_code[:cyan])m$j(cntrd)\e[0m", "\e[$(color_code[:cyan])m$j(crest)\e[0m",
"\e[$(color_code[:cyan])m$j(entrp)\e[0m", "\e[$(color_code[:cyan])m$j(flatn)\e[0m", "\e[$(color_code[:cyan])m$j(flux)\e[0m",
"\e[$(color_code[:cyan])m$j(kurts)\e[0m", "\e[$(color_code[:cyan])m$j(rllff)\e[0m", "\e[$(color_code[:cyan])m$j(skwns)\e[0m",
"\e[$(color_code[:cyan])m$j(decrs)\e[0m", "\e[$(color_code[:cyan])m$j(slope)\e[0m", "\e[$(color_code[:cyan])m$j(sprd)\e[0m"
)
for j in catch9_f
]...)
catch9 = [
maximum,
minimum,
StatsBase.mean,
median,
std,
Catch22.SB_BinaryStats_mean_longstretch1,
Catch22.SB_BinaryStats_diff_longstretch0,
Catch22.SB_MotifThree_quantile_hh,
Catch22.SB_TransitionMatrix_3ac_sumdiagcov,
]
### TODO
# catch9 = [
# maximum, ##
# # minimum,
# # StatsBase.mean,
# # median,
# std, ##
# # Catch22.SB_BinaryStats_mean_longstretch1,
# # Catch22.SB_BinaryStats_diff_longstretch0,
# # Catch22.SB_MotifThree_quantile_hh,
# # Catch22.SB_TransitionMatrix_3ac_sumdiagcov,
# Catch22.DN_HistogramMode_5, ##
# # Catch22.DN_HistogramMode_10,
# # Catch22.CO_Embed2_Dist_tau_d_expfit_meandiff,
# Catch22.CO_f1ecac, ##
# # Catch22.CO_FirstMin_ac,
# Catch22.CO_HistogramAMI_even_2_5, ##
# # Catch22.CO_trev_1_num,
# # Catch22.DN_OutlierInclude_p_001_mdrmd,
# # Catch22.DN_OutlierInclude_n_001_mdrmd,
# # Catch22.FC_LocalSimple_mean1_tauresrat,
# Catch22.FC_LocalSimple_mean3_stderr, #
# # Catch22.IN_AutoMutualInfoStats_40_gaussian_fmmi,
# Catch22.MD_hrv_classic_pnn40, #
# # Catch22.SC_FluctAnal_2_rsrangefit_50_1_logi_prop_r1,
# # Catch22.SC_FluctAnal_2_dfa_50_1_2_logi_prop_r1,
# Catch22.SP_Summaries_welch_rect_area_5_1, #
# Catch22.SP_Summaries_welch_rect_centroid, ##
# # Catch22.PD_PeriodicityWang_th0_01,
# ]
if !loadset
@info("Build dataset...")
X = DataFrame([name => Float64[] for name in [match(r_select, v)[1] for v in variable_names]])
audiofeats = [afe(row[:audio]; featset=featset, audioparams...) for row in eachrow(x)]
push!(X, vcat([vcat([map(func, eachcol(row)) for func in catch9]...) for row in audiofeats])...)
yc = CategoricalArray(y);
train_ratio = 0.8
train, test = partition(eachindex(yc), train_ratio, shuffle=true)
X_train, y_train = X[train, :], yc[train]
X_test, y_test = X[test, :], yc[test]
save(dsfile, Dict("X_test" => X_test, "y_test" => y_test))
println("Training set size: ", size(X_train), " - ", length(y_train))
println("Test set size: ", size(X_test), " - ", length(y_test))
end
# ---------------------------------------------------------------------------- #
# train a model #
# ---------------------------------------------------------------------------- #
if !loadset
learned_dt_tree = begin
Tree = MLJ.@load DecisionTreeClassifier pkg=DecisionTree
model = Tree(max_depth=-1, )
mach = machine(model, X_train, y_train)
fit!(mach)
fitted_params(mach).tree
end
end
# ---------------------------------------------------------------------------- #
# model inspection & rule study #
# ---------------------------------------------------------------------------- #
if !loadset
sole_dt = solemodel(learned_dt_tree)
# Make test instances flow into the model, so that test metrics can, then, be computed.
apply!(sole_dt, X_test, y_test);
# Save solemodel to disk
save(jld2file, Dict("sole_dt" => sole_dt))
else
@info("Load dataset...")
d = jldopen(dsfile)
X_test = d["X_test"]
y_test = d["y_test"]
close(d)
d = jldopen(jld2file)
sole_dt = d["sole_dt"]
close(d)
end
# Print Sole model
printmodel(sole_dt; show_metrics = true, variable_names_map = variable_names);
# ---------------------------------------------------------------------------- #
# extract rules that are at least as good as a random baseline model #
# ---------------------------------------------------------------------------- #
interesting_rules = listrules(sole_dt, min_lift = 1.0, min_ninstances = 0);
printmodel.(interesting_rules; show_metrics = true, variable_names_map = variable_names);
# ---------------------------------------------------------------------------- #
# simplify rules while extracting and prettify result #
# ---------------------------------------------------------------------------- #
interesting_rules = listrules(sole_dt, min_lift = 1.0, min_ninstances = 0, normalize = true);
printmodel.(interesting_rules; show_metrics = true, syntaxstring_kwargs = (; threshold_digits = 2), variable_names_map = variable_names);
# ---------------------------------------------------------------------------- #
# directly access rule metrics #
# ---------------------------------------------------------------------------- #
readmetrics.(listrules(sole_dt; min_lift=1.0, min_ninstances = 0))
# ---------------------------------------------------------------------------- #
# show rules with an additional metric (syntax height of the rule's antecedent)#
# ---------------------------------------------------------------------------- #
printmodel.(sort(interesting_rules, by = readmetrics); show_metrics = (; round_digits = nothing, additional_metrics = (; height = r->SoleLogics.height(antecedent(r)))), variable_names_map = variable_names);
# ---------------------------------------------------------------------------- #
# pretty table of rules and their metrics #
# ---------------------------------------------------------------------------- #
metricstable(interesting_rules; variable_names_map = variable_names, metrics_kwargs = (; round_digits = nothing, additional_metrics = (; height = r->SoleLogics.height(antecedent(r)))))
# ---------------------------------------------------------------------------- #
# inspecting features #
# ---------------------------------------------------------------------------- #
interesting_rules = listrules(sole_dt,
min_lift = 1.0,
# min_lift = 2.0,
min_ninstances = 0,
min_coverage = 0.10,
normalize = true,
);
map(r->(consequent(r), readmetrics(r)), interesting_rules)
printmodel.(interesting_rules; show_metrics = true, syntaxstring_kwargs = (; threshold_digits = 2), variable_names_map=variable_names);
interesting_features = unique(SoleData.feature.(SoleLogics.value.(vcat(SoleLogics.atoms.(i.antecedent for i in interesting_rules)...))))
interesting_variables = sort(SoleData.i_variable.(interesting_features))
## round_digits = 2, min_ncovered = 3, cosi vedi solo quelle che hanno massimo 3 regole, anche in tutti i printmodel
## converti la tabella prettytable in dataframe