同一套流程,三種工具都跑一遍

# 1. 載入 + 3. 找峰 + 4. 比對(核心)
import numpy as np

def modpoly(x, y, order=5, it=24):        # 2. ModPoly 基線
    yw = y.copy()
    for _ in range(it):
        c = np.polyfit(x, yw, order)
        yw = np.minimum(yw, np.polyval(c, x))
    return np.polyval(np.polyfit(x, yw, order), x)

def pick_peaks(x, y, min_h=0.05, min_dist=9):  # 3. 找峰
    idx = np.where((y[1:-1]>=y[:-2]) & (y[1:-1]>y[2:])
                   & (y[1:-1]>=min_h))[0]+1
    idx = idx[np.argsort(-y[idx])]; keep=[]
    for i in idx:
        if all(abs(x[i]-x[j])>=min_dist for j in keep): keep.append(i)
    return sorted(keep)

# 4. 與參考光譜比對(皮爾森相關)
corr = {name: np.corrcoef(sample, r)[0,1] for name,r in refs.items()}
verdict = max(corr, key=corr.get)          # -> 'CuPc'

# 5. 監督式分類(PCA + kNN,交叉驗證)
from sklearn.pipeline import make_pipeline
from sklearn.decomposition import PCA
from sklearn.neighbors import KNeighborsClassifier
from sklearn.model_selection import cross_val_score
clf = make_pipeline(PCA(5), KNeighborsClassifier(3))
acc = cross_val_score(clf, X, y, cv=5).mean()   # -> 1.00

# 1. 載入
ref  <- read.csv("raman_reference_spectra.csv", check.names=FALSE)
wn   <- ref[["wavenumber_cm-1"]]; cupc <- ref$CuPc; fcf <- ref$FCF

modpoly <- function(x, y, order=5, it=24) {   # 2. ModPoly 基線
  yw <- y; xc <- scale(x)
  for (i in seq_len(it)) yw <- pmin(yw, predict(lm(yw ~ poly(xc, order))))
  predict(lm(yw ~ poly(xc, order)))
}

pick_peaks <- function(x, y, min_h=0.05, min_dist=9) {  # 3. 找峰
  n <- length(y)
  is_max <- c(FALSE, y[2:(n-1)]>=y[1:(n-2)] & y[2:(n-1)]>y[3:n], FALSE)
  cand <- which(is_max & y>=min_h); cand <- cand[order(-y[cand])]
  keep <- integer(0)
  for (i in cand) if (all(abs(x[i]-x[keep])>=min_dist)) keep <- c(keep,i)
  sort(keep)
}

# 4. 比對(相關係數)
scores <- sapply(list(CuPc=cupc, FCF=fcf), function(r) cor(sample, r))
verdict <- names(which.max(scores))    # -> "CuPc"

# 5. PCA + 最近類心(base R;進階可用 class::knn + caret 交叉驗證)
pca <- prcomp(Xlib, center=TRUE); sc <- pca$x[, 1:5]
cent <- aggregate(sc, list(cls), mean)