experiments_ready

src/SoleAudio.jl

@@ -16,7 +16,7 @@ include("utils.jl")
 include("rules.jl")
 include("interface.jl")
 
-export get_interesting_rules
+export get_df_from_rawaudio, get_interesting_rules
 
 export collect_audio_from_folder
 

src/analysis.jl

@@ -84,6 +84,7 @@ function propositional_analisys(
     X_propos = DataFrame([name => Float64[] for name in [match(r_select, v)[1] for v in p_variable_names]])
     push!(X_propos, vcat([vcat([map(func, Array(row)) for func in metaconditions]...) for row in eachrow(X)])...)
 
+    println(rng)
     X_train, y_train, X_test, y_test = partitioning(X_propos, y; train_ratio=train_ratio, rng=rng)
 
     @info("Propositional analysis: train model...")

src/interface.jl

@@ -1,7 +1,7 @@
 # -------------------------------------------------------------------------- #
 #                           get interesting rules                            #
 # -------------------------------------------------------------------------- #
-function get_interesting_rules(;
+function get_df_from_rawaudio(;
     wav_path::String,
     csv_file::Union{String, Nothing}=nothing,
     # labels settings
@@ -17,6 +17,35 @@ function get_interesting_rules(;
     label_df::Symbol=:label,
     label_labels::Symbol=:label,
     # audio settings
+    audioparams::NamedTuple=let sr=8000
+        (
+            sr = sr,
+            norm = true,
+            speech_detect = true,
+            sdetect_thresholds=(0,0), 
+            sdetect_spread_threshold=0.02,
+            nfft = 256,
+            mel_scale = :mel_htk, # :mel_htk, :mel_slaney, :erb, :bark, :semitones, :tuned_semitones
+            mel_nbands = 26,
+            mfcc_ncoeffs = 13,
+            mel_freqrange = (0, round(Int, sr / 2)),
+        )
+    end,
+)
+    df = collect_audio_from_folder(wav_path; audioparams=audioparams, fragmented=fragmented, frag_func=frag_func)
+    labels = isnothing(csv_file) ? 
+        collect_classes(df, classes_dict; classes_func=classes_func) : 
+        collect_classes(csv_file, classes_dict; id_labels=id_labels, label_labels=label_labels, header=header)
+    merge_df_labels!(df, labels; sort_before_merge=sort_before_merge, id_df=id_df, id_labels=id_labels, label_df=label_df, label_labels=label_labels)
+    return df
+end
+
+# -------------------------------------------------------------------------- #
+#                           get interesting rules                            #
+# -------------------------------------------------------------------------- #
+function get_interesting_rules(
+    df::DataFrame;
+    # audio settings
     featset::Tuple=(:mel, :mfcc, :f0, :spectrals),
     audioparams::NamedTuple=let sr=8000
         (
@@ -32,21 +61,20 @@ function get_interesting_rules(;
             mel_freqrange = (0, round(Int, sr / 2)),
         )
     end,
+    # trim length settings
+    splitlabel::Symbol = :label,
+    lengthlabel::Symbol=:length,
+    audiolabel::Symbol=:audio,
     min_length::Int64,
     min_samples::Int64,
     features::Symbol=:catch9,
     nwindows::Int64=20,
     relative_overlap::Float64=0.05,
     train_ratio::Float64=0.8,
-    rng::AbstractRNG=Random.MersenneTwister(11),
+    rng::AbstractRNG=Random.GLOBAL_RNG,
 )
-    df = collect_audio_from_folder(wav_path; audioparams=audioparams, fragmented=fragmented, frag_func=frag_func)
-    labels = isnothing(csv_file) ? 
-        collect_classes(df, classes_dict; classes_func=classes_func) : 
-        collect_classes(csv_file, classes_dict; id_labels=id_labels, label_labels=label_labels, header=header)
-    merge_df_labels!(df, labels; sort_before_merge=sort_before_merge, id_df=id_df, id_labels=id_labels, label_df=label_df, label_labels=label_labels)
     sort_df!(df, :length; rev=true)
-    df = trimlength_df(df, :label, :length, :audio; min_length=min_length, min_samples=min_samples, sr=audioparams.sr)
+    df = trimlength_df(df, splitlabel, lengthlabel, audiolabel; min_length=min_length, min_samples=min_samples, sr=audioparams.sr)
     X, y, variable_names = afe(df, featset, audioparams)
     prop_sole_dt = propositional_analisys(X, y, variable_names=variable_names, features=features, train_ratio=train_ratio, rng=rng)
     modal_sole_dt = modal_analisys(X, y; variable_names=variable_names, features=features, nwindows=nwindows, relative_overlap=relative_overlap, train_ratio=train_ratio, rng=rng)

src/rules.jl

@@ -47,22 +47,21 @@ function interesting_rules(
         map(r->(consequent(r), readmetrics(r)), irules)
         p_irules = sort(irules, by=readmetrics)
 
-        isempty(p_irules) && throw(ArgumentError("No interesting rules found."))
+        isempty(p_irules) && return nothing
 
-        p_X = DataFrame(antecedent=String[], consequent=String[]; [name => Vector{Union{Float64, Int}}() for name in keys(readmetrics(p_irules[1]))]...)
+        p_X = DataFrame(antecedent=String[], consequent=String[]; [name => Vector{Union{Float64, Int}}() for name in keys(readmetrics(p_irules[1]))]..., type=String[])
 #         p_X = DataFrame(
 #     antecedent = String[],
 #     consequent = String[],
 #     [name => Vector{Union{Float64, Int}}() for name in keys(readmetrics(p_irules[1]))]...
 # )
 
-
         for i in eachrow(p_irules)
             a_c = match(r_p_divide, string(i[1]))
             antecedent, consequent = a_c[1], a_c[3]
             antecedent = reduce((s, r) -> replace(s, r => ""), r_p_ant, init=antecedent)
             antecedent = format_float(antecedent)
-            push!(p_X, (antecedent, consequent, readmetrics(i[1])...))
+            push!(p_X, (antecedent, consequent, readmetrics(i[1])..., "propositional"))
         end
     else
         p_X = nothing
@@ -81,7 +80,7 @@ function interesting_rules(
         map(r->(r, readmetrics(r)), irules)
         m_irules = sort(irules, by=readmetrics)
 
-        m_X = DataFrame(antecedent=String[], consequent=String[]; [name => Vector{Union{Float64, Int}}() for name in keys(readmetrics(m_irules[1]))]...)
+        m_X = DataFrame(antecedent=String[], consequent=String[]; [name => Vector{Union{Float64, Int}}() for name in keys(readmetrics(m_irules[1]))]..., type=String[])
         for i in eachrow(m_irules)
             consequent = match(r_cons, string(i[1].consequent))[1]
             # antecedent = foldl((s, r) -> replace(s, r => ""), r_antecedent, init=string(i[1].antecedent))
@@ -98,7 +97,7 @@ function interesting_rules(
                 end)
             end
             antecedent = format_float(antecedent)
-            push!(m_X, (antecedent, consequent, readmetrics(i[1])...))
+            push!(m_X, (antecedent, consequent, readmetrics(i[1])..., "modal"))
         end
     else
         m_X = nothing

test/launcher.sh

@@ -3,34 +3,16 @@
 N_THREADS=30
 
 # ------------------------------------------------------------------------------------ #
-#                                      itadata 2024                                    #
+#                                     itadata2024                                      #
 # ------------------------------------------------------------------------------------ #
-# n=1
-# session="session_"$n
-# tmux new-session -d -s $session
-# tmux send-keys -t $session 'julia -i -t '$N_THREADS' ravdess_emotion_2bins.jl ' Enter
-# [ -n "${TMUX:-}" ]
+n=1
+session="session_"$n
+tmux new-session -d -s $session
+tmux send-keys -t $session 'julia -i -t '$N_THREADS' ravdess_emotion_2bins.jl ' Enter
+[ -n "${TMUX:-}" ]
 
 n=2
 session="session_"$n
 tmux new-session -d -s $session
 tmux send-keys -t $session 'julia -i -t '$N_THREADS' ravdess_emotion_8bins.jl ' Enter
-[ -n "${TMUX:-}" ]
-
-# n=3
-# session="session_"$n
-# tmux new-session -d -s $session
-# tmux send-keys -t $session 'julia -i -t '$N_THREADS' respiratory_pneumonia.jl ' Enter
-# [ -n "${TMUX:-}" ]
-
-# n=5
-# session="session_"$n
-# tmux new-session -d -s $session
-# tmux send-keys -t $session 'julia -i -t '$N_THREADS' respiratory_copd.jl ' Enter
-# [ -n "${TMUX:-}" ]
-
-# n=5
-# session="session_"$n
-# tmux new-session -d -s $session
-# tmux send-keys -t $session 'julia -i -t '$N_THREADS' respiratory_bronchiolitis.jl ' Enter
-# [ -n "${TMUX:-}" ]
+[ -n "${TMUX:-}" ]

test/ravdess_emotion_2bins.jl

@@ -2,11 +2,6 @@ using DataFrames, JLD2
 using SoleAudio, Random
 # using Plots
 
-# TODO
-# scrivi un file text con tutti i settaggi usati
-# output formattato per latex su un file.tex
-# fra gli algoritmi di f0 includi il pagliarini
-
 # -------------------------------------------------------------------------- #
 #                       experiment specific parameters                       #
 # -------------------------------------------------------------------------- #
@@ -49,6 +44,8 @@ elseif classes == :emo8bins
     )
 end
 
+jld2_file = string("ravdess_", classes)
+
 # classes will be taken from audio filename, no csv available
 classes_func(row) = match(r"^(?:[^-]*-){2}([^-]*)", row.filename)[1]
 
@@ -65,19 +62,15 @@ audioparams = let sr = 8000
         sdetect_thresholds=(0,0), 
         sdetect_spread_threshold=0.02,
         nfft = 256,
-        mel_scale = :mel_htk, # :mel_htk, :mel_slaney, :erb, :bark, :semitones, :tuned_semitones
+        mel_scale = :semitones, # :mel_htk, :mel_slaney, :erb, :bark, :semitones, :tuned_semitones
         mel_nbands = 26,
         mfcc_ncoeffs = 13,
-        mel_freqrange = (0, round(Int, sr / 2)),
+        mel_freqrange = (100, round(Int, sr / 2)),
     )
 end
 
-# min_length = 11500
-# min_samples = 400
-
-# only for debugging
-min_length = 11000
-min_samples = 500
+min_length = 14000
+min_samples = 75
 
 features = :catch9
 # features = :minmax
@@ -89,16 +82,22 @@ relative_overlap = 0.05
 
 # partitioning
 train_ratio = 0.8
-train_seed = 11
+train_seed = 1
 rng = Random.MersenneTwister(train_seed)
+Random.seed!(train_seed)
 
 # -------------------------------------------------------------------------- #
 #                                   main                                     #
 # -------------------------------------------------------------------------- #
-irules = get_interesting_rules(
+df = get_df_from_rawaudio(
     wav_path=wav_path,
     classes_dict=classes_dict,
     classes_func=classes_func,
+    audioparams=audioparams,
+)
+
+irules = get_interesting_rules(
+    df;
     featset=featset,
     audioparams=audioparams,
     min_length=min_length,
@@ -110,4 +109,7 @@ irules = get_interesting_rules(
     rng=rng,
 )
 
-jldsave("ravdess_emotion_2bins.jld2", true; irules)
+println(irules)
+
+jldsave(jld2_file * ".jld2", true; irules)
+@info "Done."

test/ravdess_emotion_8bins.jl

@@ -2,11 +2,6 @@ using DataFrames, JLD2
 using SoleAudio, Random
 # using Plots
 
-# TODO
-# scrivi un file text con tutti i settaggi usati
-# output formattato per latex su un file.tex
-# fra gli algoritmi di f0 includi il pagliarini
-
 # -------------------------------------------------------------------------- #
 #                       experiment specific parameters                       #
 # -------------------------------------------------------------------------- #
@@ -49,6 +44,8 @@ elseif classes == :emo8bins
     )
 end
 
+jld2_file = string("ravdess_", classes)
+
 # classes will be taken from audio filename, no csv available
 classes_func(row) = match(r"^(?:[^-]*-){2}([^-]*)", row.filename)[1]
 
@@ -66,18 +63,14 @@ audioparams = let sr = 8000
         sdetect_spread_threshold=0.02,
         nfft = 256,
         mel_scale = :mel_htk, # :mel_htk, :mel_slaney, :erb, :bark, :semitones, :tuned_semitones
-        mel_nbands = 26,
-        mfcc_ncoeffs = 13,
-        mel_freqrange = (0, round(Int, sr / 2)),
+        mel_nbands = 20,
+        mfcc_ncoeffs = 10,
+        mel_freqrange = (100, round(Int, sr / 2)),
     )
 end
 
-# min_length = 11500
-# min_samples = 400
-
-# only for debugging
-min_length = 12000
-min_samples = 46
+min_length = 11500
+min_samples = 50
 
 features = :catch9
 # features = :minmax
@@ -89,16 +82,22 @@ relative_overlap = 0.05
 
 # partitioning
 train_ratio = 0.8
-train_seed = 11
+train_seed = 1
 rng = Random.MersenneTwister(train_seed)
+Random.seed!(train_seed)
 
 # -------------------------------------------------------------------------- #
 #                                   main                                     #
 # -------------------------------------------------------------------------- #
-irules = get_interesting_rules(
+df = get_df_from_rawaudio(
     wav_path=wav_path,
     classes_dict=classes_dict,
     classes_func=classes_func,
+    audioparams=audioparams,
+)
+
+irules = get_interesting_rules(
+    df;
     featset=featset,
     audioparams=audioparams,
     min_length=min_length,
@@ -110,4 +109,7 @@ irules = get_interesting_rules(
     rng=rng,
 )
 
-jldsave("ravdess_emotion_8bins.jld2", true; irules)
+println(irules)
+
+jldsave(jld2_file * ".jld2", true; irules)
+@info "Done."