Merge pull request #20 from instaclustr/performance

Performance merging

Cargo.toml

@@ -16,6 +16,10 @@ prom-remote-api = { version = "0.3.0", features = ["warp"] }
 async-trait = "0.1.71"
 env_logger = "0.10.0"
 log = "0.4.0"
+clap = {version = "4.3.14", features = ["derive"] }
+regex = "1.9.1"
+median = "0.3.2"
+dtw_rs = "0.9.5"
 
 [profile.dev]
 opt-level = 0

matlab/process_ts.m

@@ -0,0 +1,140 @@
+function [dc, ac, composed, fft_data] = process_ts(ts, w, freq_n, n_hold)
+% ts: timeseries
+% w: window size
+% freq_n: Number of frequencies to find
+tic
+if nargin<4
+  n_hold = 0;
+endif
+
+nanIDX = find(isnan(ts));
+while(~isempty(nanIDX))
+  ts(nanIDX) = ts(nanIDX+1);
+  nanIDX = find(isnan(ts));
+end
+
+% Split the signal in DC and AC parts
+%dc = movmean(ts, w);
+ac = center(ts);
+%ac = ts-dc;
+dc = ts-ac;
+
+%hold on;plot(dc);plot(ac);
+
+window_n = ceil(length(ts)/w);
+data_rebuild = [];
+fft_store = [];
+fft_data = [];
+window_err = [];
+% Process the whole signal
+for i=1:window_n
+  window_s = (i-1)*w + 1;
+  window_e = i*w;
+  if i == window_n
+    data_window = ts(window_s:end);
+    data_dc = dc(window_s:end);
+    data_ac = ac(window_s:end);
+  else
+    data_window = ts(window_s:window_e);
+    %data_dc = movmean(data_window, w/10);
+    data_dc = dc(window_s:window_e);
+    %data_ac = data_window - data_dc;
+    data_ac = ac(window_s:window_e);
+  endif
+  window_size = length(data_dc);
+
+  % Process AC data
+  if isempty(fft_store)
+    f = fft(data_ac);
+    tmp_f = f;
+    out_fft = zeros(1, window_size);
+    window_freqs = [];
+    if freq_n > window_size/2
+      freq_n = floor(window_size/2);
+    endif
+    for i=1:freq_n*2
+      [mx,ix] = max(tmp_f);
+      window_freqs(i,:) = [real(ix) mx];
+      tmp_f(ix) = 0;
+      out_fft(ix) = mx;
+    end
+    fft_data = [fft_data out_fft];
+    %disp("Window Frequencies: ")
+    %disp(sort(window_freqs))
+    out_ift = ifft(out_fft);
+    if n_hold ~= 0
+      fft_store = out_ift;
+    endif
+  elseif n_hold ~= 0
+    out_ift = fft_store;
+    fft_store = [];
+  endif
+
+  % Process DC data
+  yi = polyfit(1:window_size,data_dc,1);
+  %disp("DC points: ")
+  %disp(yi)
+  % Rebuild the sinal for the window
+  yii = polyval(yi,1:window_size);
+  % Build the dataset for the window
+  window_rebuild = real(out_ift)+yii;
+
+  % Calculate the error
+  pererr = abs(data_window-window_rebuild)./data_window*100;
+  mean(pererr)
+  window_err = [window_err pererr];
+  data_rebuild = [data_rebuild window_rebuild];
+
+
+  %plot(abs(out_fft))
+end
+toc
+composed = data_rebuild;
+nnz(fft_data)
+figure;
+plot(window_err);
+
+figure;
+subplot(2,2,1);
+plot(data_rebuild);
+title('Rebuild');
+subplot(2,2,2);
+plot(ts, 'r');
+title('Original');
+subplot(2,2,[3,4]);
+plot(ts,'r',data_rebuild,'b');
+title('Both');
+
+
+%{
+wdw = ac(1:w);
+f = fft(wdw);
+
+% Create a output array
+tmp_f = f;
+out_fft = zeros(1, w);
+
+% Zero out the frequency just found and around it
+for i=1:freq_n*2
+  [mx,ix] = max(tmp_f);
+  tmp_f(ix) = 0;
+  out_fft(ix) = mx;
+end
+
+out_ift = ifft(out_fft);
+ift = ifft(f);
+
+% DC component approximation
+yi = polyfit(1:w,dc(1:w),1);
+% Lets see the aproximattion
+yii = polyval(yi,1:w);
+rebuilt = real(out_ift)+yii;
+x = rebuilt;
+
+hold on;
+%plot(abs(out_fft))
+%plot(wdw)
+%plot(real(ift))
+plot(rebuilt)
+plot(ts(1:w))
+%}