samples

3,720

wavelengths

sources

targets

metadata

mixed

family

Dataset property explorer

Mean profile risk0.42

Highest axisBaseline / forme · 0.75

Diagnostics10

Sources profiled6

Profile axes

Intégrité	0.00
Artefacts locaux	0.73
Bruit	0.01
Outliers PCA	0.59
Distance à la référence	0.60
Répétabilité	0.00
Baseline / forme	0.75
Structure multi-régimes	0.70

Diagnostic	Score	Force	Signaux	Interprétation probable
Splice / raccord détecteursX5	0.72	moyenne	SNR non dégradé 1.00, RMS/SAM référence 0.97, Spike rate 0.85	Rupture aux jonctions de détecteurs, calibration locale ou sonde différente.
Erreur calibration / référence blancheX5	0.68	moyenne	Baseline/mean/area 1.00, RMS/SAM référence 0.97, artefacts locaux 0.85	Décalage systématique entre campagnes, instruments ou référence blanche.
Erreur interpolation / rééchantillonnageX1	0.64	moyenne	PCA Q 1.00, Spike rate 1.00, SNR normal/élevé 1.00	Artefacts numériques ou traitement spectral incorrect.
Fond différentX5	0.64	moyenne	Baseline/mean/area 1.00, RMS/SAM référence 0.97, Mahalanobis / T2 0.43	Effet systématique du support, blanc/noir, transflectance ou environnement de mesure.
Spectre hors domaine valideX4	0.61	moyenne	artefacts faibles 0.96, Structure PCA 0.89, Mahalanobis / T2 0.49	Variété, espèce, lot ou condition différente mais physiquement plausible.
Dataset multi-régimesX1	0.60	moyenne	Structure PCA 1.00, PCA Q 1.00, RMS/SAM référence 0.78	Mélange de campagnes, opérateurs, lots, setups ou sous-populations spectrales.
Signature VERA25-likeX1	0.60	moyenne	PCA Q 1.00, Spike rate 1.00, RMS/SAM référence 0.78	Combinaison possible changement de sonde + splice, amplifiée par géométrie, fond ou calibration.
Différence de sonde / géométrieX1	0.54	moyenne	PCA Q 1.00, RMS/SAM référence 0.78, Mahalanobis / T2 0.55	Modification de l'illumination, collecte, angle ou distance sonde-échantillon.

Spectral sources

Infratec 1255 NIT

X1 · NIR · Infratec 1255 NIT

Sampling

Wavelengths	100
Axis range	850–1,048 nm
Mean spacing	2 nm
Grid	uniform
Observations	32

Signal & quality

Value range	1.39 – 2.35
Mean range	1.52 – 1.82
Mean level	1.661
Area	328.9
PTP	0.3041
Noise RMS	0.00017995
SNR	9.2e+03
SNR dB	8e+01 dB
Dynamic range	0.304
Smoothness	0.001911
Saturated	0.0%
X-outliers	10

Integrity & artefacts

NaN ratio	0.00%
Inf count	0
Zero ratio	0.00%
Spike count	51
Spike rate	1.63%
Jump count	0
Jump rate	0.00%
Clip fraction	0.06%

Shape & reference

Baseline slope	0.30285
Curvature RMS	0.00082456
D1 RMS	0.0048986
RMS to mean	0.15475
RMS p95	0.32426
SAM to mean	0.019896
SAM p95	0.029254
Affine offset p95	0.92909
Affine gain p95 Δ	0.74937
Affine residual p95	0.020287
Xcorr lag p95	0

Outliers & repeatability

PCA Q p95/median	11
Hotelling T2 p95/median	4.4
Mahalanobis H p95/median	2.1
Repeat groups	0

Dimensionality (PCA)

Effective rank	1
PCs → 95% var	1
PCs → 99% var	1
Top-10 cum. var	100.0%

Computed metric scores 29worst 1.00

Famille	Métrique calculée	Valeur	Score	Niveau	Interprétation dataset	Causes typiques	Calcul / scoring
Intégrité des données	`NaN ratio`integrity.nan_ratio	0%	0.00	faible	Spectre complet	Erreur acquisition/export	count(isnan(X)) / X.sizealert = min(1, nan_ratio / 0.05)
Intégrité des données	`Inf count`integrity.inf_count	0	0.00	faible	Normal	Calculs invalides	count(isinf(X))alert = min(1, inf_count / 1)
Intégrité des données	`Zero ratio`integrity.zero_ratio	0%	0.00	faible	Normal	Export, saturation	count(X == 0) / count(finite X)alert = min(1, zero_ratio / 0.05)
Amplitude globale	`Mean reflectance`amplitude.mean_reflectance	1.6606	0.36	faible	Trop sombre	Fond, géométrie	mean(X finite)alert reuses baseline/shape drift because absolute reflectance ranges are technology-dependent
Amplitude globale	`Area under curve`amplitude.area_under_curve	328.89	0.36	faible	Normal	Distance sonde	trapezoid(mean_spectrum, spectral_axis)alert reuses baseline/shape drift because area scale depends on axis and units
Amplitude globale	`Peak-to-peak (PTP)`amplitude.peak_to_peak	0.30408	0.00	faible	Variabilité forte	Saturation	max(mean_spectrum) - min(mean_spectrum)alert increases when dynamic range is abnormally flat
Amplitude globale	`Variance`amplitude.variance	0.043156	0.00	faible	Normal ou hétérogène	Mauvais contact	var(X finite)alert increases when variance/dynamic range is abnormally flat
Bruit	`Noise RMS`noise.noise_rms	0.00017995	0.00	faible	Stable	Lampe, détecteur	median MAD(second derivative) * 1.4826 / sqrt(6)alert = noise_rms / signal_scale, saturated at 5%
Bruit	`SNR`noise.snr	9228	0.00	faible	Bon signal	Acquisition	mean(abs(X)) / noise_rmsalert decreases with SNR dB; >=40 dB is treated as low alert
Bruit	`Bandwise SNR`noise.bandwise_snr_min	956.47	0.00	faible	Zone fiable	Détecteur	min(abs(mean_spectrum) / local second-derivative noise)alert decreases with worst-band SNR dB; >=35 dB is treated as low alert
Artefacts locaux	`Spike count`artefacts.spike_count	51	1.00	fort	Artefacts	Cosmic rays, splice	count robust outliers in second derivativealert follows spike_rate, saturated at 1%
Artefacts locaux	`Spike rate`artefacts.spike_rate	1.63%	1.00	fort	Spectre suspect	Interpolation	spike_count / (n_samples * (n_features - 2))alert = min(1, spike_rate / 0.01)
Artefacts locaux	`Jump count`artefacts.jump_count	0	0.00	faible	Continu	Splice	count robust outliers in first derivativealert follows jump_rate, saturated at 1%
Artefacts locaux	`Jump rate`artefacts.jump_rate	0%	0.00	faible	Normal	Calibration	jump_count / (n_samples * (n_features - 1))alert = min(1, jump_rate / 0.01)
Artefacts locaux	`Clip fraction`artefacts.clip_fraction	0.0625%	0.06	faible	Normal	Détecteur saturé	fraction of finite cells equal to repeated min/max extremaalert = min(1, clip_fraction / 0.01)
Forme spectrale	`Baseline slope`shape.baseline_slope	0.30285	0.36	faible	Stable	Éclairement	linear slope of mean_spectrum over normalized axisalert = abs(slope / signal_scale), saturated at 0.5
Forme spectrale	`Curvature RMS`shape.curvature_rms	0.00082456	0.05	faible	Lisse	Fond, splice	median RMS(second derivative per spectrum)alert = curvature_rms / signal_scale, saturated at 1%
Forme spectrale	`D1 RMS`shape.d1_rms	0.0048986	0.06	faible	Plat	Biologie ou artefact	median RMS(first derivative per spectrum)alert = d1_rms / signal_scale, saturated at 5%
Outliers multivariés	`PCA Q (SPE)`outliers.pca_q_ratio	10.794	1.00	fort	Spectre atypique	Artefact, mélange	p95(Q/SPE residual) / median(Q/SPE residual)alert = min(1, pca_q_ratio / 8)
Outliers multivariés	`Hotelling T²`outliers.hotelling_t2_ratio	4.396	0.55	moyen	Extrême mais cohérent	Variabilité naturelle	p95(Hotelling T2) / median(Hotelling T2)alert = min(1, hotelling_t2_ratio / 8)
Outliers multivariés	`Mahalanobis H`outliers.mahalanobis_h_ratio	2.0965	0.52	moyen	Outlier global	Domaine différent	p95(sqrt(T2)) / median(sqrt(T2))alert = min(1, mahalanobis_h_ratio / 4)
Comparaison à référence	`RMS to mean spectrum`reference.rms_to_mean_spectrum_p95	0.32426	0.78	fort	Spectre différent	Domain shift	p95 RMS distance to dataset mean spectrumalert = RMS_p95 / signal_scale, saturated at 25%
Comparaison à référence	`Spectral Angle Mapper (SAM)`reference.sam_to_mean_spectrum_p95	0.029254	0.08	faible	Similaire	Fond, géométrie	p95 spectral angle to dataset mean spectrumalert = min(1, SAM_p95 / 0.35 rad)
Répétabilité	`RMS intra-ID`repeatability.rms_intra_id	—	0.00	faible	Stable	Positionnement	median RMS distance to repeated-sample centroidalert = RMS_intra_ID / signal_scale, saturated at 10%
Répétabilité	`SAM intra-ID`repeatability.sam_intra_id	—	0.00	faible	Stable	Acquisition	median SAM to repeated-sample centroidalert = min(1, SAM_intra_ID / 0.15 rad)
Répétabilité	`CV intra-ID`repeatability.cv_intra_id	—	0.00	faible	Stable	Opérateur	median within-ID band CValert = min(1, CV_intra_ID / 0.25)
Structure du dataset	`PCA score density`structure.pca_score_density	3.3677	1.00	fort	Sous-populations	Lots différents	1 / median kNN distance in PCA score spacealert follows density_cv/profile structure complexity, not raw density alone
Structure du dataset	`Local Outlier Factor (LOF)`structure.local_outlier_factor_p95	3.5583	1.00	fort	Spectre isolé	Cas rares	p95 approximate LOF from PCA-score kNN distancesalert = min(1, max(0, LOF_p95 - 1) / 2)
Structure du dataset	`Isolation Forest score`structure.isolation_forest_score_p95	0.61031	1.00	fort	Spectre atypique	Diverses causes	p95 IsolationForest anomaly score on PCA scoresalert follows structure complexity; raw score is implementation-dependent

X-Y spectral correlation 2

Target	max \|r\|	axis @ max	mean \|r\|	\|r\| ≥ .5
ref_sugar	0.979	982	0.969	100.0%
ref_moisture	0.907	926	0.879	100.0%

NIRSystems 6500 sample cup reflection

X2 · NIR · NIRSystems 6500 sample cup reflection

Sampling

Wavelengths	1,000
Axis range	450–2,448 nm
Mean spacing	2 nm
Grid	uniform
Observations	32

Signal & quality

Value range	0.223 – 2.02
Mean range	0.351 – 1.92
Mean level	1.162
Area	2322
PTP	1.565
Noise RMS	0.00017747
SNR	6.5e+03
SNR dB	8e+01 dB
Dynamic range	1.57
Smoothness	0.001224
Saturated	0.0%
X-outliers	10

Integrity & artefacts

NaN ratio	0.00%
Inf count	0
Zero ratio	0.00%
Spike count	181
Spike rate	0.57%
Jump count	620
Jump rate	1.94%
Clip fraction	0.01%

Shape & reference

Baseline slope	1.9165
Curvature RMS	0.0011971
D1 RMS	0.0057058
RMS to mean	0.10729
RMS p95	0.19257
SAM to mean	0.04689
SAM p95	0.10669
Affine offset p95	0.20207
Affine gain p95 Δ	0.10668
Affine residual p95	0.093652
Xcorr lag p95	0

Outliers & repeatability

PCA Q p95/median	5.1
Hotelling T2 p95/median	3.2
Mahalanobis H p95/median	1.8
Repeat groups	0

Dimensionality (PCA)

Effective rank	2
PCs → 95% var	3
PCs → 99% var	4
Top-10 cum. var	100.0%

Computed metric scores 29worst 1.00

Famille	Métrique calculée	Valeur	Score	Niveau	Interprétation dataset	Causes typiques	Calcul / scoring
Intégrité des données	`NaN ratio`integrity.nan_ratio	0%	0.00	faible	Spectre complet	Erreur acquisition/export	count(isnan(X)) / X.sizealert = min(1, nan_ratio / 0.05)
Intégrité des données	`Inf count`integrity.inf_count	0	0.00	faible	Normal	Calculs invalides	count(isinf(X))alert = min(1, inf_count / 1)
Intégrité des données	`Zero ratio`integrity.zero_ratio	0%	0.00	faible	Normal	Export, saturation	count(X == 0) / count(finite X)alert = min(1, zero_ratio / 0.05)
Amplitude globale	`Mean reflectance`amplitude.mean_reflectance	1.1622	1.00	fort	Valeur atypique: Trop clair / fond visible ou Trop sombre	Fond, géométrie	mean(X finite)alert reuses baseline/shape drift because absolute reflectance ranges are technology-dependent
Amplitude globale	`Area under curve`amplitude.area_under_curve	2321.6	1.00	fort	Valeur atypique: Différence d'éclairement ou Normal	Distance sonde	trapezoid(mean_spectrum, spectral_axis)alert reuses baseline/shape drift because area scale depends on axis and units
Amplitude globale	`Peak-to-peak (PTP)`amplitude.peak_to_peak	1.565	0.00	faible	Variabilité forte	Saturation	max(mean_spectrum) - min(mean_spectrum)alert increases when dynamic range is abnormally flat
Amplitude globale	`Variance`amplitude.variance	0.37165	0.00	faible	Normal ou hétérogène	Mauvais contact	var(X finite)alert increases when variance/dynamic range is abnormally flat
Bruit	`Noise RMS`noise.noise_rms	0.00017747	0.00	faible	Stable	Lampe, détecteur	median MAD(second derivative) * 1.4826 / sqrt(6)alert = noise_rms / signal_scale, saturated at 5%
Bruit	`SNR`noise.snr	6548.7	0.00	faible	Bon signal	Acquisition	mean(abs(X)) / noise_rmsalert decreases with SNR dB; >=40 dB is treated as low alert
Bruit	`Bandwise SNR`noise.bandwise_snr_min	500.69	0.00	faible	Zone fiable	Détecteur	min(abs(mean_spectrum) / local second-derivative noise)alert decreases with worst-band SNR dB; >=35 dB is treated as low alert
Artefacts locaux	`Spike count`artefacts.spike_count	181	0.57	moyen	Artefacts	Cosmic rays, splice	count robust outliers in second derivativealert follows spike_rate, saturated at 1%
Artefacts locaux	`Spike rate`artefacts.spike_rate	0.567%	0.57	moyen	Spectre suspect	Interpolation	spike_count / (n_samples * (n_features - 2))alert = min(1, spike_rate / 0.01)
Artefacts locaux	`Jump count`artefacts.jump_count	620	1.00	fort	Raccord détecteur	Splice	count robust outliers in first derivativealert follows jump_rate, saturated at 1%
Artefacts locaux	`Jump rate`artefacts.jump_rate	1.94%	1.00	fort	Problème spectral	Calibration	jump_count / (n_samples * (n_features - 1))alert = min(1, jump_rate / 0.01)
Artefacts locaux	`Clip fraction`artefacts.clip_fraction	0.00625%	0.01	faible	Normal	Détecteur saturé	fraction of finite cells equal to repeated min/max extremaalert = min(1, clip_fraction / 0.01)
Forme spectrale	`Baseline slope`shape.baseline_slope	1.9165	1.00	fort	Dérive	Éclairement	linear slope of mean_spectrum over normalized axisalert = abs(slope / signal_scale), saturated at 0.5
Forme spectrale	`Curvature RMS`shape.curvature_rms	0.0011971	0.08	faible	Lisse	Fond, splice	median RMS(second derivative per spectrum)alert = curvature_rms / signal_scale, saturated at 1%
Forme spectrale	`D1 RMS`shape.d1_rms	0.0057058	0.07	faible	Plat	Biologie ou artefact	median RMS(first derivative per spectrum)alert = d1_rms / signal_scale, saturated at 5%
Outliers multivariés	`PCA Q (SPE)`outliers.pca_q_ratio	5.0785	0.63	moyen	Spectre atypique	Artefact, mélange	p95(Q/SPE residual) / median(Q/SPE residual)alert = min(1, pca_q_ratio / 8)
Outliers multivariés	`Hotelling T²`outliers.hotelling_t2_ratio	3.1984	0.40	faible	Central	Variabilité naturelle	p95(Hotelling T2) / median(Hotelling T2)alert = min(1, hotelling_t2_ratio / 8)
Outliers multivariés	`Mahalanobis H`outliers.mahalanobis_h_ratio	1.7883	0.45	moyen	Outlier global	Domaine différent	p95(sqrt(T2)) / median(sqrt(T2))alert = min(1, mahalanobis_h_ratio / 4)
Comparaison à référence	`RMS to mean spectrum`reference.rms_to_mean_spectrum_p95	0.19257	0.49	moyen	Spectre différent	Domain shift	p95 RMS distance to dataset mean spectrumalert = RMS_p95 / signal_scale, saturated at 25%
Comparaison à référence	`Spectral Angle Mapper (SAM)`reference.sam_to_mean_spectrum_p95	0.10669	0.30	faible	Similaire	Fond, géométrie	p95 spectral angle to dataset mean spectrumalert = min(1, SAM_p95 / 0.35 rad)
Répétabilité	`RMS intra-ID`repeatability.rms_intra_id	—	0.00	faible	Stable	Positionnement	median RMS distance to repeated-sample centroidalert = RMS_intra_ID / signal_scale, saturated at 10%
Répétabilité	`SAM intra-ID`repeatability.sam_intra_id	—	0.00	faible	Stable	Acquisition	median SAM to repeated-sample centroidalert = min(1, SAM_intra_ID / 0.15 rad)
Répétabilité	`CV intra-ID`repeatability.cv_intra_id	—	0.00	faible	Stable	Opérateur	median within-ID band CValert = min(1, CV_intra_ID / 0.25)
Structure du dataset	`PCA score density`structure.pca_score_density	0.60277	0.59	moyen	Sous-populations	Lots différents	1 / median kNN distance in PCA score spacealert follows density_cv/profile structure complexity, not raw density alone
Structure du dataset	`Local Outlier Factor (LOF)`structure.local_outlier_factor_p95	1.773	0.39	faible	Population normale	Cas rares	p95 approximate LOF from PCA-score kNN distancesalert = min(1, max(0, LOF_p95 - 1) / 2)
Structure du dataset	`Isolation Forest score`structure.isolation_forest_score_p95	0.57903	0.59	moyen	Spectre atypique	Diverses causes	p95 IsolationForest anomaly score on PCA scoresalert follows structure complexity; raw score is implementation-dependent

X-Y spectral correlation 2

Target	max \|r\|	axis @ max	mean \|r\|	\|r\| ≥ .5
ref_sugar	0.952	1,854	0.734	77.0%
ref_moisture	0.922	800	0.726	82.5%

NIRSystems 6500 fibre probe reflection

X3 · NIR · NIRSystems 6500 fibre probe reflection

Sampling

Wavelengths	600
Axis range	850–2,048 nm
Mean spacing	2 nm
Grid	uniform
Observations	32

Signal & quality

Value range	0.285 – 1.27
Mean range	0.475 – 1.14
Mean level	0.8632
Area	1034
PTP	0.6688
Noise RMS	0.00011614
SNR	7.4e+03
SNR dB	8e+01 dB
Dynamic range	0.669
Smoothness	0.0005434
Saturated	0.0%
X-outliers	10

Integrity & artefacts

NaN ratio	0.00%
Inf count	0
Zero ratio	0.00%
Spike count	91
Spike rate	0.48%
Jump count	13
Jump rate	0.07%
Clip fraction	0.01%

Shape & reference

Baseline slope	0.73586
Curvature RMS	0.00050351
D1 RMS	0.0030987
RMS to mean	0.077231
RMS p95	0.16142
SAM to mean	0.045676
SAM p95	0.11439
Affine offset p95	0.31286
Affine gain p95 Δ	0.25134
Affine residual p95	0.038022
Xcorr lag p95	0

Outliers & repeatability

PCA Q p95/median	2.2
Hotelling T2 p95/median	3.4
Mahalanobis H p95/median	1.8
Repeat groups	0

Dimensionality (PCA)

Effective rank	1.3
PCs → 95% var	2
PCs → 99% var	2
Top-10 cum. var	100.0%

Computed metric scores 29worst 1.00

Famille	Métrique calculée	Valeur	Score	Niveau	Interprétation dataset	Causes typiques	Calcul / scoring
Intégrité des données	`NaN ratio`integrity.nan_ratio	0%	0.00	faible	Spectre complet	Erreur acquisition/export	count(isnan(X)) / X.sizealert = min(1, nan_ratio / 0.05)
Intégrité des données	`Inf count`integrity.inf_count	0	0.00	faible	Normal	Calculs invalides	count(isinf(X))alert = min(1, inf_count / 1)
Intégrité des données	`Zero ratio`integrity.zero_ratio	0%	0.00	faible	Normal	Export, saturation	count(X == 0) / count(finite X)alert = min(1, zero_ratio / 0.05)
Amplitude globale	`Mean reflectance`amplitude.mean_reflectance	0.86315	1.00	fort	Valeur atypique: Trop clair / fond visible ou Trop sombre	Fond, géométrie	mean(X finite)alert reuses baseline/shape drift because absolute reflectance ranges are technology-dependent
Amplitude globale	`Area under curve`amplitude.area_under_curve	1034.2	1.00	fort	Valeur atypique: Différence d'éclairement ou Normal	Distance sonde	trapezoid(mean_spectrum, spectral_axis)alert reuses baseline/shape drift because area scale depends on axis and units
Amplitude globale	`Peak-to-peak (PTP)`amplitude.peak_to_peak	0.66882	0.00	faible	Variabilité forte	Saturation	max(mean_spectrum) - min(mean_spectrum)alert increases when dynamic range is abnormally flat
Amplitude globale	`Variance`amplitude.variance	0.062471	0.00	faible	Normal ou hétérogène	Mauvais contact	var(X finite)alert increases when variance/dynamic range is abnormally flat
Bruit	`Noise RMS`noise.noise_rms	0.00011614	0.00	faible	Stable	Lampe, détecteur	median MAD(second derivative) * 1.4826 / sqrt(6)alert = noise_rms / signal_scale, saturated at 5%
Bruit	`SNR`noise.snr	7432	0.00	faible	Bon signal	Acquisition	mean(abs(X)) / noise_rmsalert decreases with SNR dB; >=40 dB is treated as low alert
Bruit	`Bandwise SNR`noise.bandwise_snr_min	785.23	0.00	faible	Zone fiable	Détecteur	min(abs(mean_spectrum) / local second-derivative noise)alert decreases with worst-band SNR dB; >=35 dB is treated as low alert
Artefacts locaux	`Spike count`artefacts.spike_count	91	0.48	moyen	Artefacts	Cosmic rays, splice	count robust outliers in second derivativealert follows spike_rate, saturated at 1%
Artefacts locaux	`Spike rate`artefacts.spike_rate	0.476%	0.48	moyen	Spectre suspect	Interpolation	spike_count / (n_samples * (n_features - 2))alert = min(1, spike_rate / 0.01)
Artefacts locaux	`Jump count`artefacts.jump_count	13	0.07	faible	Continu	Splice	count robust outliers in first derivativealert follows jump_rate, saturated at 1%
Artefacts locaux	`Jump rate`artefacts.jump_rate	0.0678%	0.07	faible	Normal	Calibration	jump_count / (n_samples * (n_features - 1))alert = min(1, jump_rate / 0.01)
Artefacts locaux	`Clip fraction`artefacts.clip_fraction	0.0104%	0.01	faible	Normal	Détecteur saturé	fraction of finite cells equal to repeated min/max extremaalert = min(1, clip_fraction / 0.01)
Forme spectrale	`Baseline slope`shape.baseline_slope	0.73586	1.00	fort	Dérive	Éclairement	linear slope of mean_spectrum over normalized axisalert = abs(slope / signal_scale), saturated at 0.5
Forme spectrale	`Curvature RMS`shape.curvature_rms	0.00050351	0.06	faible	Lisse	Fond, splice	median RMS(second derivative per spectrum)alert = curvature_rms / signal_scale, saturated at 1%
Forme spectrale	`D1 RMS`shape.d1_rms	0.0030987	0.07	faible	Plat	Biologie ou artefact	median RMS(first derivative per spectrum)alert = d1_rms / signal_scale, saturated at 5%
Outliers multivariés	`PCA Q (SPE)`outliers.pca_q_ratio	2.1685	0.27	faible	Conforme	Artefact, mélange	p95(Q/SPE residual) / median(Q/SPE residual)alert = min(1, pca_q_ratio / 8)
Outliers multivariés	`Hotelling T²`outliers.hotelling_t2_ratio	3.4201	0.43	moyen	Extrême mais cohérent	Variabilité naturelle	p95(Hotelling T2) / median(Hotelling T2)alert = min(1, hotelling_t2_ratio / 8)
Outliers multivariés	`Mahalanobis H`outliers.mahalanobis_h_ratio	1.8493	0.46	moyen	Outlier global	Domaine différent	p95(sqrt(T2)) / median(sqrt(T2))alert = min(1, mahalanobis_h_ratio / 4)
Comparaison à référence	`RMS to mean spectrum`reference.rms_to_mean_spectrum_p95	0.16142	0.75	fort	Spectre différent	Domain shift	p95 RMS distance to dataset mean spectrumalert = RMS_p95 / signal_scale, saturated at 25%
Comparaison à référence	`Spectral Angle Mapper (SAM)`reference.sam_to_mean_spectrum_p95	0.11439	0.33	faible	Similaire	Fond, géométrie	p95 spectral angle to dataset mean spectrumalert = min(1, SAM_p95 / 0.35 rad)
Répétabilité	`RMS intra-ID`repeatability.rms_intra_id	—	0.00	faible	Stable	Positionnement	median RMS distance to repeated-sample centroidalert = RMS_intra_ID / signal_scale, saturated at 10%
Répétabilité	`SAM intra-ID`repeatability.sam_intra_id	—	0.00	faible	Stable	Acquisition	median SAM to repeated-sample centroidalert = min(1, SAM_intra_ID / 0.15 rad)
Répétabilité	`CV intra-ID`repeatability.cv_intra_id	—	0.00	faible	Stable	Opérateur	median within-ID band CValert = min(1, CV_intra_ID / 0.25)
Structure du dataset	`PCA score density`structure.pca_score_density	1.2874	0.64	moyen	Sous-populations	Lots différents	1 / median kNN distance in PCA score spacealert follows density_cv/profile structure complexity, not raw density alone
Structure du dataset	`Local Outlier Factor (LOF)`structure.local_outlier_factor_p95	1.9288	0.46	moyen	Spectre isolé	Cas rares	p95 approximate LOF from PCA-score kNN distancesalert = min(1, max(0, LOF_p95 - 1) / 2)
Structure du dataset	`Isolation Forest score`structure.isolation_forest_score_p95	0.61634	0.64	moyen	Spectre atypique	Diverses causes	p95 IsolationForest anomaly score on PCA scoresalert follows structure complexity; raw score is implementation-dependent

X-Y spectral correlation 2

Target	max \|r\|	axis @ max	mean \|r\|	\|r\| ≥ .5
ref_sugar	0.931	1,210	0.789	88.3%
ref_moisture	0.956	1,408	0.849	94.7%

Infraprover II FT-NIR

X4 · NIR · Infraprover II FT-NIR

Sampling

Wavelengths	406
Axis range	1051–2148 nm
Mean spacing	2.71 nm
Grid	irregular
Observations	32

Signal & quality

Value range	0.524 – 0.966
Mean range	0.696 – 0.94
Mean level	0.8307
Area	944.8
PTP	0.2433
Noise RMS	0.00021694
SNR	3.8e+03
SNR dB	7e+01 dB
Dynamic range	0.243
Smoothness	0.0006818
Saturated	0.0%
X-outliers	10

Integrity & artefacts

NaN ratio	0.00%
Inf count	0
Zero ratio	0.00%
Spike count	5
Spike rate	0.04%
Jump count	0
Jump rate	0.00%
Clip fraction	0.02%

Shape & reference

Baseline slope	0.26991
Curvature RMS	0.00058439
D1 RMS	0.001255
RMS to mean	0.045331
RMS p95	0.10081
SAM to mean	0.033626
SAM p95	0.080962
Affine offset p95	0.60416
Affine gain p95 Δ	0.63667
Affine residual p95	0.011017
Xcorr lag p95	0

Outliers & repeatability

PCA Q p95/median	3.1
Hotelling T2 p95/median	3.9
Mahalanobis H p95/median	2
Repeat groups	0

Dimensionality (PCA)

Effective rank	1.2
PCs → 95% var	2
PCs → 99% var	2
Top-10 cum. var	100.0%

Computed metric scores 29worst 0.89

Famille	Métrique calculée	Valeur	Score	Niveau	Interprétation dataset	Causes typiques	Calcul / scoring
Intégrité des données	`NaN ratio`integrity.nan_ratio	0%	0.00	faible	Spectre complet	Erreur acquisition/export	count(isnan(X)) / X.sizealert = min(1, nan_ratio / 0.05)
Intégrité des données	`Inf count`integrity.inf_count	0	0.00	faible	Normal	Calculs invalides	count(isinf(X))alert = min(1, inf_count / 1)
Intégrité des données	`Zero ratio`integrity.zero_ratio	0%	0.00	faible	Normal	Export, saturation	count(X == 0) / count(finite X)alert = min(1, zero_ratio / 0.05)
Amplitude globale	`Mean reflectance`amplitude.mean_reflectance	0.83068	0.65	moyen	Valeur atypique: Trop clair / fond visible ou Trop sombre	Fond, géométrie	mean(X finite)alert reuses baseline/shape drift because absolute reflectance ranges are technology-dependent
Amplitude globale	`Area under curve`amplitude.area_under_curve	944.79	0.65	moyen	Valeur atypique: Différence d'éclairement ou Normal	Distance sonde	trapezoid(mean_spectrum, spectral_axis)alert reuses baseline/shape drift because area scale depends on axis and units
Amplitude globale	`Peak-to-peak (PTP)`amplitude.peak_to_peak	0.24334	0.00	faible	Variabilité forte	Saturation	max(mean_spectrum) - min(mean_spectrum)alert increases when dynamic range is abnormally flat
Amplitude globale	`Variance`amplitude.variance	0.0099667	0.00	faible	Normal ou hétérogène	Mauvais contact	var(X finite)alert increases when variance/dynamic range is abnormally flat
Bruit	`Noise RMS`noise.noise_rms	0.00021694	0.01	faible	Stable	Lampe, détecteur	median MAD(second derivative) * 1.4826 / sqrt(6)alert = noise_rms / signal_scale, saturated at 5%
Bruit	`SNR`noise.snr	3829.1	0.00	faible	Bon signal	Acquisition	mean(abs(X)) / noise_rmsalert decreases with SNR dB; >=40 dB is treated as low alert
Bruit	`Bandwise SNR`noise.bandwise_snr_min	1553.5	0.00	faible	Zone fiable	Détecteur	min(abs(mean_spectrum) / local second-derivative noise)alert decreases with worst-band SNR dB; >=35 dB is treated as low alert
Artefacts locaux	`Spike count`artefacts.spike_count	5	0.04	faible	Spectre propre	Cosmic rays, splice	count robust outliers in second derivativealert follows spike_rate, saturated at 1%
Artefacts locaux	`Spike rate`artefacts.spike_rate	0.0387%	0.04	faible	Normal	Interpolation	spike_count / (n_samples * (n_features - 2))alert = min(1, spike_rate / 0.01)
Artefacts locaux	`Jump count`artefacts.jump_count	0	0.00	faible	Continu	Splice	count robust outliers in first derivativealert follows jump_rate, saturated at 1%
Artefacts locaux	`Jump rate`artefacts.jump_rate	0%	0.00	faible	Normal	Calibration	jump_count / (n_samples * (n_features - 1))alert = min(1, jump_rate / 0.01)
Artefacts locaux	`Clip fraction`artefacts.clip_fraction	0.0154%	0.02	faible	Normal	Détecteur saturé	fraction of finite cells equal to repeated min/max extremaalert = min(1, clip_fraction / 0.01)
Forme spectrale	`Baseline slope`shape.baseline_slope	0.26991	0.65	moyen	Dérive	Éclairement	linear slope of mean_spectrum over normalized axisalert = abs(slope / signal_scale), saturated at 0.5
Forme spectrale	`Curvature RMS`shape.curvature_rms	0.00058439	0.07	faible	Lisse	Fond, splice	median RMS(second derivative per spectrum)alert = curvature_rms / signal_scale, saturated at 1%
Forme spectrale	`D1 RMS`shape.d1_rms	0.001255	0.03	faible	Plat	Biologie ou artefact	median RMS(first derivative per spectrum)alert = d1_rms / signal_scale, saturated at 5%
Outliers multivariés	`PCA Q (SPE)`outliers.pca_q_ratio	3.053	0.38	faible	Conforme	Artefact, mélange	p95(Q/SPE residual) / median(Q/SPE residual)alert = min(1, pca_q_ratio / 8)
Outliers multivariés	`Hotelling T²`outliers.hotelling_t2_ratio	3.878	0.48	moyen	Extrême mais cohérent	Variabilité naturelle	p95(Hotelling T2) / median(Hotelling T2)alert = min(1, hotelling_t2_ratio / 8)
Outliers multivariés	`Mahalanobis H`outliers.mahalanobis_h_ratio	1.9701	0.49	moyen	Outlier global	Domaine différent	p95(sqrt(T2)) / median(sqrt(T2))alert = min(1, mahalanobis_h_ratio / 4)
Comparaison à référence	`RMS to mean spectrum`reference.rms_to_mean_spectrum_p95	0.10081	0.49	moyen	Spectre différent	Domain shift	p95 RMS distance to dataset mean spectrumalert = RMS_p95 / signal_scale, saturated at 25%
Comparaison à référence	`Spectral Angle Mapper (SAM)`reference.sam_to_mean_spectrum_p95	0.080962	0.23	faible	Similaire	Fond, géométrie	p95 spectral angle to dataset mean spectrumalert = min(1, SAM_p95 / 0.35 rad)
Répétabilité	`RMS intra-ID`repeatability.rms_intra_id	—	0.00	faible	Stable	Positionnement	median RMS distance to repeated-sample centroidalert = RMS_intra_ID / signal_scale, saturated at 10%
Répétabilité	`SAM intra-ID`repeatability.sam_intra_id	—	0.00	faible	Stable	Acquisition	median SAM to repeated-sample centroidalert = min(1, SAM_intra_ID / 0.15 rad)
Répétabilité	`CV intra-ID`repeatability.cv_intra_id	—	0.00	faible	Stable	Opérateur	median within-ID band CValert = min(1, CV_intra_ID / 0.25)
Structure du dataset	`PCA score density`structure.pca_score_density	3.2265	0.89	fort	Sous-populations	Lots différents	1 / median kNN distance in PCA score spacealert follows density_cv/profile structure complexity, not raw density alone
Structure du dataset	`Local Outlier Factor (LOF)`structure.local_outlier_factor_p95	2.682	0.84	fort	Spectre isolé	Cas rares	p95 approximate LOF from PCA-score kNN distancesalert = min(1, max(0, LOF_p95 - 1) / 2)
Structure du dataset	`Isolation Forest score`structure.isolation_forest_score_p95	0.57702	0.89	fort	Spectre atypique	Diverses causes	p95 IsolationForest anomaly score on PCA scoresalert follows structure complexity; raw score is implementation-dependent

X-Y spectral correlation 2

Target	max \|r\|	axis @ max	mean \|r\|	\|r\| ≥ .5
ref_sugar	0.897	1.12e+03	0.597	61.1%
ref_moisture	0.914	1.33e+03	0.668	70.7%

Bomem MB160 Diffusir FT-NIR

X5 · NIR · Bomem MB160 Diffusir FT-NIR

Sampling

Wavelengths	664
Axis range	850–2443 nm
Mean spacing	2.4 nm
Grid	irregular
Observations	32

Signal & quality

Value range	-0.295 – 1.25
Mean range	-0.173 – 1.05
Mean level	0.3831
Area	927.4
PTP	1.22
Noise RMS	0.0012629
SNR	3.3e+02
SNR dB	5e+01 dB
Dynamic range	1.22
Smoothness	0.005749
Saturated	0.0%
X-outliers	10

Integrity & artefacts

NaN ratio	0.00%
Inf count	0
Zero ratio	0.00%
Spike count	180
Spike rate	0.85%
Jump count	158
Jump rate	0.74%
Clip fraction	0.01%

Shape & reference

Baseline slope	1.2411
Curvature RMS	0.0050735
D1 RMS	0.0089656
RMS to mean	0.09958
RMS p95	0.2052
SAM to mean	0.10555
SAM p95	0.33872
Affine offset p95	0.15348
Affine gain p95 Δ	0.12015
Affine residual p95	0.088313
Xcorr lag p95	0

Outliers & repeatability

PCA Q p95/median	2.7
Hotelling T2 p95/median	3
Mahalanobis H p95/median	1.7
Repeat groups	0

Dimensionality (PCA)

Effective rank	1.3
PCs → 95% var	2
PCs → 99% var	3
Top-10 cum. var	100.0%

Computed metric scores 29worst 1.00

Famille	Métrique calculée	Valeur	Score	Niveau	Interprétation dataset	Causes typiques	Calcul / scoring
Intégrité des données	`NaN ratio`integrity.nan_ratio	0%	0.00	faible	Spectre complet	Erreur acquisition/export	count(isnan(X)) / X.sizealert = min(1, nan_ratio / 0.05)
Intégrité des données	`Inf count`integrity.inf_count	0	0.00	faible	Normal	Calculs invalides	count(isinf(X))alert = min(1, inf_count / 1)
Intégrité des données	`Zero ratio`integrity.zero_ratio	0%	0.00	faible	Normal	Export, saturation	count(X == 0) / count(finite X)alert = min(1, zero_ratio / 0.05)
Amplitude globale	`Mean reflectance`amplitude.mean_reflectance	0.38312	1.00	fort	Valeur atypique: Trop clair / fond visible ou Trop sombre	Fond, géométrie	mean(X finite)alert reuses baseline/shape drift because absolute reflectance ranges are technology-dependent
Amplitude globale	`Area under curve`amplitude.area_under_curve	927.41	1.00	fort	Valeur atypique: Différence d'éclairement ou Normal	Distance sonde	trapezoid(mean_spectrum, spectral_axis)alert reuses baseline/shape drift because area scale depends on axis and units
Amplitude globale	`Peak-to-peak (PTP)`amplitude.peak_to_peak	1.2198	0.00	faible	Variabilité forte	Saturation	max(mean_spectrum) - min(mean_spectrum)alert increases when dynamic range is abnormally flat
Amplitude globale	`Variance`amplitude.variance	0.17618	0.00	faible	Normal ou hétérogène	Mauvais contact	var(X finite)alert increases when variance/dynamic range is abnormally flat
Bruit	`Noise RMS`noise.noise_rms	0.0012629	0.02	faible	Stable	Lampe, détecteur	median MAD(second derivative) * 1.4826 / sqrt(6)alert = noise_rms / signal_scale, saturated at 5%
Bruit	`SNR`noise.snr	334.06	0.00	faible	Bon signal	Acquisition	mean(abs(X)) / noise_rmsalert decreases with SNR dB; >=40 dB is treated as low alert
Bruit	`Bandwise SNR`noise.bandwise_snr_min	0.079264	1.00	fort	Zone problématique	Détecteur	min(abs(mean_spectrum) / local second-derivative noise)alert decreases with worst-band SNR dB; >=35 dB is treated as low alert
Artefacts locaux	`Spike count`artefacts.spike_count	180	0.85	fort	Artefacts	Cosmic rays, splice	count robust outliers in second derivativealert follows spike_rate, saturated at 1%
Artefacts locaux	`Spike rate`artefacts.spike_rate	0.85%	0.85	fort	Spectre suspect	Interpolation	spike_count / (n_samples * (n_features - 2))alert = min(1, spike_rate / 0.01)
Artefacts locaux	`Jump count`artefacts.jump_count	158	0.74	fort	Raccord détecteur	Splice	count robust outliers in first derivativealert follows jump_rate, saturated at 1%
Artefacts locaux	`Jump rate`artefacts.jump_rate	0.745%	0.74	fort	Problème spectral	Calibration	jump_count / (n_samples * (n_features - 1))alert = min(1, jump_rate / 0.01)
Artefacts locaux	`Clip fraction`artefacts.clip_fraction	0.00941%	0.01	faible	Normal	Détecteur saturé	fraction of finite cells equal to repeated min/max extremaalert = min(1, clip_fraction / 0.01)
Forme spectrale	`Baseline slope`shape.baseline_slope	1.2411	1.00	fort	Dérive	Éclairement	linear slope of mean_spectrum over normalized axisalert = abs(slope / signal_scale), saturated at 0.5
Forme spectrale	`Curvature RMS`shape.curvature_rms	0.0050735	0.42	faible	Lisse	Fond, splice	median RMS(second derivative per spectrum)alert = curvature_rms / signal_scale, saturated at 1%
Forme spectrale	`D1 RMS`shape.d1_rms	0.0089656	0.15	faible	Plat	Biologie ou artefact	median RMS(first derivative per spectrum)alert = d1_rms / signal_scale, saturated at 5%
Outliers multivariés	`PCA Q (SPE)`outliers.pca_q_ratio	2.7459	0.34	faible	Conforme	Artefact, mélange	p95(Q/SPE residual) / median(Q/SPE residual)alert = min(1, pca_q_ratio / 8)
Outliers multivariés	`Hotelling T²`outliers.hotelling_t2_ratio	3.0067	0.38	faible	Central	Variabilité naturelle	p95(Hotelling T2) / median(Hotelling T2)alert = min(1, hotelling_t2_ratio / 8)
Outliers multivariés	`Mahalanobis H`outliers.mahalanobis_h_ratio	1.7329	0.43	moyen	Outlier global	Domaine différent	p95(sqrt(T2)) / median(sqrt(T2))alert = min(1, mahalanobis_h_ratio / 4)
Comparaison à référence	`RMS to mean spectrum`reference.rms_to_mean_spectrum_p95	0.2052	0.67	moyen	Spectre différent	Domain shift	p95 RMS distance to dataset mean spectrumalert = RMS_p95 / signal_scale, saturated at 25%
Comparaison à référence	`Spectral Angle Mapper (SAM)`reference.sam_to_mean_spectrum_p95	0.33872	0.97	fort	Forme différente	Fond, géométrie	p95 spectral angle to dataset mean spectrumalert = min(1, SAM_p95 / 0.35 rad)
Répétabilité	`RMS intra-ID`repeatability.rms_intra_id	—	0.00	faible	Stable	Positionnement	median RMS distance to repeated-sample centroidalert = RMS_intra_ID / signal_scale, saturated at 10%
Répétabilité	`SAM intra-ID`repeatability.sam_intra_id	—	0.00	faible	Stable	Acquisition	median SAM to repeated-sample centroidalert = min(1, SAM_intra_ID / 0.15 rad)
Répétabilité	`CV intra-ID`repeatability.cv_intra_id	—	0.00	faible	Stable	Opérateur	median within-ID band CValert = min(1, CV_intra_ID / 0.25)
Structure du dataset	`PCA score density`structure.pca_score_density	0.87479	0.62	moyen	Sous-populations	Lots différents	1 / median kNN distance in PCA score spacealert follows density_cv/profile structure complexity, not raw density alone
Structure du dataset	`Local Outlier Factor (LOF)`structure.local_outlier_factor_p95	1.8622	0.43	moyen	Spectre isolé	Cas rares	p95 approximate LOF from PCA-score kNN distancesalert = min(1, max(0, LOF_p95 - 1) / 2)
Structure du dataset	`Isolation Forest score`structure.isolation_forest_score_p95	0.58465	0.62	moyen	Spectre atypique	Diverses causes	p95 IsolationForest anomaly score on PCA scoresalert follows structure complexity; raw score is implementation-dependent

X-Y spectral correlation 2

Target	max \|r\|	axis @ max	mean \|r\|	\|r\| ≥ .5
ref_sugar	0.954	1.86e+03	0.851	93.1%
ref_moisture	0.921	851	0.809	91.4%

IR spectroscopy

X6 · MIR · IR spectroscopy

Sampling

Wavelengths	950
Axis range	651–3,799 cm-1
Mean spacing	3.32 cm-1
Grid	irregular
Observations	32

Signal & quality

Value range	0.0251 – 2.53
Mean range	0.0296 – 2.29
Mean level	0.7209
Area	2079
PTP	2.256
Noise RMS	0.001611
SNR	4.5e+02
SNR dB	5e+01 dB
Dynamic range	2.26
Smoothness	0.01152
Saturated	0.0%
X-outliers	10

Integrity & artefacts

NaN ratio	0.00%
Inf count	0
Zero ratio	0.00%
Spike count	895
Spike rate	2.95%
Jump count	204
Jump rate	0.67%
Clip fraction	0.01%

Shape & reference

Baseline slope	-0.55222
Curvature RMS	0.01154
D1 RMS	0.027924
RMS to mean	0.030789
RMS p95	0.052762
SAM to mean	0.029553
SAM p95	0.039962
Affine offset p95	0.016317
Affine gain p95 Δ	0.054719
Affine residual p95	0.037258
Xcorr lag p95	0

Outliers & repeatability

PCA Q p95/median	2.8
Hotelling T2 p95/median	3.8
Mahalanobis H p95/median	2
Repeat groups	0

Dimensionality (PCA)

Effective rank	3.3
PCs → 95% var	4
PCs → 99% var	7
Top-10 cum. var	99.8%

Computed metric scores 29worst 1.00

Famille	Métrique calculée	Valeur	Score	Niveau	Interprétation dataset	Causes typiques	Calcul / scoring
Intégrité des données	`NaN ratio`integrity.nan_ratio	0%	0.00	faible	Spectre complet	Erreur acquisition/export	count(isnan(X)) / X.sizealert = min(1, nan_ratio / 0.05)
Intégrité des données	`Inf count`integrity.inf_count	0	0.00	faible	Normal	Calculs invalides	count(isinf(X))alert = min(1, inf_count / 1)
Intégrité des données	`Zero ratio`integrity.zero_ratio	0%	0.00	faible	Normal	Export, saturation	count(X == 0) / count(finite X)alert = min(1, zero_ratio / 0.05)
Amplitude globale	`Mean reflectance`amplitude.mean_reflectance	0.72093	0.49	moyen	Valeur atypique: Trop clair / fond visible ou Trop sombre	Fond, géométrie	mean(X finite)alert reuses baseline/shape drift because absolute reflectance ranges are technology-dependent
Amplitude globale	`Area under curve`amplitude.area_under_curve	2079	0.49	moyen	Valeur atypique: Différence d'éclairement ou Normal	Distance sonde	trapezoid(mean_spectrum, spectral_axis)alert reuses baseline/shape drift because area scale depends on axis and units
Amplitude globale	`Peak-to-peak (PTP)`amplitude.peak_to_peak	2.2556	0.00	faible	Variabilité forte	Saturation	max(mean_spectrum) - min(mean_spectrum)alert increases when dynamic range is abnormally flat
Amplitude globale	`Variance`amplitude.variance	0.34179	0.00	faible	Normal ou hétérogène	Mauvais contact	var(X finite)alert increases when variance/dynamic range is abnormally flat
Bruit	`Noise RMS`noise.noise_rms	0.001611	0.01	faible	Stable	Lampe, détecteur	median MAD(second derivative) * 1.4826 / sqrt(6)alert = noise_rms / signal_scale, saturated at 5%
Bruit	`SNR`noise.snr	447.51	0.00	faible	Bon signal	Acquisition	mean(abs(X)) / noise_rmsalert decreases with SNR dB; >=40 dB is treated as low alert
Bruit	`Bandwise SNR`noise.bandwise_snr_min	87.489	0.00	faible	Zone fiable	Détecteur	min(abs(mean_spectrum) / local second-derivative noise)alert decreases with worst-band SNR dB; >=35 dB is treated as low alert
Artefacts locaux	`Spike count`artefacts.spike_count	895	1.00	fort	Artefacts	Cosmic rays, splice	count robust outliers in second derivativealert follows spike_rate, saturated at 1%
Artefacts locaux	`Spike rate`artefacts.spike_rate	2.95%	1.00	fort	Spectre suspect	Interpolation	spike_count / (n_samples * (n_features - 2))alert = min(1, spike_rate / 0.01)
Artefacts locaux	`Jump count`artefacts.jump_count	204	0.67	moyen	Raccord détecteur	Splice	count robust outliers in first derivativealert follows jump_rate, saturated at 1%
Artefacts locaux	`Jump rate`artefacts.jump_rate	0.672%	0.67	moyen	Problème spectral	Calibration	jump_count / (n_samples * (n_features - 1))alert = min(1, jump_rate / 0.01)
Artefacts locaux	`Clip fraction`artefacts.clip_fraction	0.00658%	0.01	faible	Normal	Détecteur saturé	fraction of finite cells equal to repeated min/max extremaalert = min(1, clip_fraction / 0.01)
Forme spectrale	`Baseline slope`shape.baseline_slope	-0.55222	0.49	moyen	Dérive	Éclairement	linear slope of mean_spectrum over normalized axisalert = abs(slope / signal_scale), saturated at 0.5
Forme spectrale	`Curvature RMS`shape.curvature_rms	0.01154	0.51	moyen	Forme inhabituelle	Fond, splice	median RMS(second derivative per spectrum)alert = curvature_rms / signal_scale, saturated at 1%
Forme spectrale	`D1 RMS`shape.d1_rms	0.027924	0.25	faible	Plat	Biologie ou artefact	median RMS(first derivative per spectrum)alert = d1_rms / signal_scale, saturated at 5%
Outliers multivariés	`PCA Q (SPE)`outliers.pca_q_ratio	2.7951	0.35	faible	Conforme	Artefact, mélange	p95(Q/SPE residual) / median(Q/SPE residual)alert = min(1, pca_q_ratio / 8)
Outliers multivariés	`Hotelling T²`outliers.hotelling_t2_ratio	3.8275	0.48	moyen	Extrême mais cohérent	Variabilité naturelle	p95(Hotelling T2) / median(Hotelling T2)alert = min(1, hotelling_t2_ratio / 8)
Outliers multivariés	`Mahalanobis H`outliers.mahalanobis_h_ratio	1.9586	0.49	moyen	Outlier global	Domaine différent	p95(sqrt(T2)) / median(sqrt(T2))alert = min(1, mahalanobis_h_ratio / 4)
Comparaison à référence	`RMS to mean spectrum`reference.rms_to_mean_spectrum_p95	0.052762	0.09	faible	Typique	Domain shift	p95 RMS distance to dataset mean spectrumalert = RMS_p95 / signal_scale, saturated at 25%
Comparaison à référence	`Spectral Angle Mapper (SAM)`reference.sam_to_mean_spectrum_p95	0.039962	0.11	faible	Similaire	Fond, géométrie	p95 spectral angle to dataset mean spectrumalert = min(1, SAM_p95 / 0.35 rad)
Répétabilité	`RMS intra-ID`repeatability.rms_intra_id	—	0.00	faible	Stable	Positionnement	median RMS distance to repeated-sample centroidalert = RMS_intra_ID / signal_scale, saturated at 10%
Répétabilité	`SAM intra-ID`repeatability.sam_intra_id	—	0.00	faible	Stable	Acquisition	median SAM to repeated-sample centroidalert = min(1, SAM_intra_ID / 0.15 rad)
Répétabilité	`CV intra-ID`repeatability.cv_intra_id	—	0.00	faible	Stable	Opérateur	median within-ID band CValert = min(1, CV_intra_ID / 0.25)
Structure du dataset	`PCA score density`structure.pca_score_density	1.2889	0.48	moyen	Sous-populations	Lots différents	1 / median kNN distance in PCA score spacealert follows density_cv/profile structure complexity, not raw density alone
Structure du dataset	`Local Outlier Factor (LOF)`structure.local_outlier_factor_p95	1.427	0.21	faible	Population normale	Cas rares	p95 approximate LOF from PCA-score kNN distancesalert = min(1, max(0, LOF_p95 - 1) / 2)
Structure du dataset	`Isolation Forest score`structure.isolation_forest_score_p95	0.59057	0.48	moyen	Spectre atypique	Diverses causes	p95 IsolationForest anomaly score on PCA scoresalert follows structure complexity; raw score is implementation-dependent

X-Y spectral correlation 2

Target	max \|r\|	axis @ max	mean \|r\|	\|r\| ≥ .5
ref_sugar	0.965	903	0.346	34.7%
ref_moisture	0.893	1,236	0.418	45.1%

Metric interpretation reference

Metric catalog 29

Famille	Métrique	Ce qu’elle détecte	Forte valeur =	Faible valeur =	Causes typiques	Calcul / score
Intégrité des données	`NaN ratio`	Données manquantes	Spectre corrompu	Spectre complet	Erreur acquisition/export	count(isnan(X)) / X.sizealert = min(1, nan_ratio / 0.05)
Intégrité des données	`Inf count`	Valeurs infinies	Corruption	Normal	Calculs invalides	count(isinf(X))alert = min(1, inf_count / 1)
Intégrité des données	`Zero ratio`	Colonnes ou cellules nulles	Spectre tronqué	Normal	Export, saturation	count(X == 0) / count(finite X)alert = min(1, zero_ratio / 0.05)
Amplitude globale	`Mean reflectance`	Niveau moyen	Trop clair / fond visible	Trop sombre	Fond, géométrie	mean(X finite)alert reuses baseline/shape drift because absolute reflectance ranges are technology-dependent
Amplitude globale	`Area under curve`	Intensité globale	Différence d'éclairement	Normal	Distance sonde	trapezoid(mean_spectrum, spectral_axis)alert reuses baseline/shape drift because area scale depends on axis and units
Amplitude globale	`Peak-to-peak (PTP)`	Dynamique	Variabilité forte	Spectre plat	Saturation	max(mean_spectrum) - min(mean_spectrum)alert increases when dynamic range is abnormally flat
Amplitude globale	`Variance`	Variabilité spectrale	Normal ou hétérogène	Spectre plat	Mauvais contact	var(X finite)alert increases when variance/dynamic range is abnormally flat
Bruit	`Noise RMS`	Bruit haute fréquence	Bruité	Stable	Lampe, détecteur	median MAD(second derivative) * 1.4826 / sqrt(6)alert = noise_rms / signal_scale, saturated at 5%
Bruit	`SNR`	Qualité signal	Bon signal	Mauvais signal	Acquisition	mean(abs(X)) / noise_rmsalert decreases with SNR dB; >=40 dB is treated as low alert
Bruit	`Bandwise SNR`	Bruit localisé	Zone fiable	Zone problématique	Détecteur	min(abs(mean_spectrum) / local second-derivative noise)alert decreases with worst-band SNR dB; >=35 dB is treated as low alert
Artefacts locaux	`Spike count`	Pics étroits	Artefacts	Spectre propre	Cosmic rays, splice	count robust outliers in second derivativealert follows spike_rate, saturated at 1%
Artefacts locaux	`Spike rate`	Densité de pics	Spectre suspect	Normal	Interpolation	spike_count / (n_samples * (n_features - 2))alert = min(1, spike_rate / 0.01)
Artefacts locaux	`Jump count`	Discontinuités	Raccord détecteur	Continu	Splice	count robust outliers in first derivativealert follows jump_rate, saturated at 1%
Artefacts locaux	`Jump rate`	Fréquence de sauts	Problème spectral	Normal	Calibration	jump_count / (n_samples * (n_features - 1))alert = min(1, jump_rate / 0.01)
Artefacts locaux	`Clip fraction`	Saturation	Clipping	Normal	Détecteur saturé	fraction of finite cells equal to repeated min/max extremaalert = min(1, clip_fraction / 0.01)
Forme spectrale	`Baseline slope`	Pente globale	Dérive	Stable	Éclairement	linear slope of mean_spectrum over normalized axisalert = abs(slope / signal_scale), saturated at 0.5
Forme spectrale	`Curvature RMS`	Courbure	Forme inhabituelle	Lisse	Fond, splice	median RMS(second derivative per spectrum)alert = curvature_rms / signal_scale, saturated at 1%
Forme spectrale	`D1 RMS`	Variabilité locale	Spectre structuré	Plat	Biologie ou artefact	median RMS(first derivative per spectrum)alert = d1_rms / signal_scale, saturated at 5%
Outliers multivariés	`PCA Q (SPE)`	Non expliqué par PCA	Spectre atypique	Conforme	Artefact, mélange	p95(Q/SPE residual) / median(Q/SPE residual)alert = min(1, pca_q_ratio / 8)
Outliers multivariés	`Hotelling T²`	Extrême dans PCA	Extrême mais cohérent	Central	Variabilité naturelle	p95(Hotelling T2) / median(Hotelling T2)alert = min(1, hotelling_t2_ratio / 8)
Outliers multivariés	`Mahalanobis H`	Distance au nuage	Outlier global	Population normale	Domaine différent	p95(sqrt(T2)) / median(sqrt(T2))alert = min(1, mahalanobis_h_ratio / 4)
Comparaison à référence	`RMS to mean spectrum`	Distance moyenne	Spectre différent	Typique	Domain shift	p95 RMS distance to dataset mean spectrumalert = RMS_p95 / signal_scale, saturated at 25%
Comparaison à référence	`Spectral Angle Mapper (SAM)`	Différence de forme	Forme différente	Similaire	Fond, géométrie	p95 spectral angle to dataset mean spectrumalert = min(1, SAM_p95 / 0.35 rad)
Répétabilité	`RMS intra-ID`	Reproductibilité	Mauvaise répétabilité	Stable	Positionnement	median RMS distance to repeated-sample centroidalert = RMS_intra_ID / signal_scale, saturated at 10%
Répétabilité	`SAM intra-ID`	Variation de forme	Instable	Stable	Acquisition	median SAM to repeated-sample centroidalert = min(1, SAM_intra_ID / 0.15 rad)
Répétabilité	`CV intra-ID`	Variabilité interne	Mauvais contrôle	Stable	Opérateur	median within-ID band CValert = min(1, CV_intra_ID / 0.25)
Structure du dataset	`PCA score density`	Clusters	Sous-populations	Homogène	Lots différents	1 / median kNN distance in PCA score spacealert follows density_cv/profile structure complexity, not raw density alone
Structure du dataset	`Local Outlier Factor (LOF)`	Anomalie locale	Spectre isolé	Population normale	Cas rares	p95 approximate LOF from PCA-score kNN distancesalert = min(1, max(0, LOF_p95 - 1) / 2)
Structure du dataset	`Isolation Forest score`	Anomalie globale	Spectre atypique	Normal	Diverses causes	p95 IsolationForest anomaly score on PCA scoresalert follows structure complexity; raw score is implementation-dependent

Technology-specific extensions

Technologie	Adaptations / métriques	Anomalies ciblées	Commentaire pratique
UV-Vis 300-1000 nm	Baseline, pente globale, dérive aux bords 300-350 et 900-1000; métriques par zones	Lumière parasite, mauvais blanc, saturation, faible signal aux extrémités	Les bords sont souvent instables; calculer aussi des scores edge/middle.
UV-Vis 300-1000 nm	Saturation / clipping proche absorbance max ou réflectance max	Signal écrêté	Très important si absorption forte.
UV-Vis 300-1000 nm	Red-edge, position de maximum, ratios de bandes si végétal	Décalage biologique ou artefact optique	Aide à distinguer changement réel et problème d'acquisition.
UV-Vis 300-1000 nm	Smoothness / roughness index	Bruit haute fréquence	Souvent plus informatif que le SNR seul.
MIR / ATR-FTIR	ATR contact quality index: intensité globale, aire totale, profondeur des bandes clés	Mauvais contact cristal-échantillon	Crucial: beaucoup d'anomalies viennent du contact ATR.
MIR / ATR-FTIR	CO2 / H2O atmospheric bands	Mauvaise correction atmosphérique	Pics parasites fréquents.
MIR / ATR-FTIR	Baseline curvature / rubber-band residual	Diffusion, contact, dérive baseline	Très utile avant PCA.
MIR / ATR-FTIR	Peak position shift	Mauvais alignement spectral / calibration	Important en FTIR car de petits shifts comptent.
MIR / ATR-FTIR	Band area ratios sur bandes connues	Spectre chimiquement incohérent	À adapter par matrice: polysaccharides, protéines, lipides, etc.
HS-MS	Total Ion Current (TIC), Base Peak Intensity (BPI)	Injection faible, ionisation instable	Équivalent MS du niveau global spectral.
HS-MS	Nombre de pics détectés	Spectre pauvre ou trop bruité	Trop peu = mauvais signal; trop = bruit/contamination.
HS-MS	Mass accuracy / m/z drift	Problème calibration masse	Fondamental en HRMS.
HS-MS	Retention time drift si LC/GC-MS	Dérive chromatographique	À suivre sur standards/QC pools.
HS-MS	Blank contamination score	Contaminants / carry-over	Comparer échantillons vs blancs.
HS-MS	Internal standard CV	Variabilité instrumentale	Très robuste si standards disponibles.
HS-MS	Missingness par feature	Instabilité de détection	Crucial pour filtrer les variables.
Avec répétitions	RMS intra-échantillon	Répétabilité globale	Applicable à toutes les technologies.
Avec répétitions	SAM / corrélation intra-échantillon	Répétabilité de forme	Très utile pour spectres.
Avec répétitions	CV intra-échantillon par bande / feature	Répétabilité locale	Détecte les zones instables.
Avec répétitions	ICC ou variance components	Part variance échantillon vs technique	Très utile si plusieurs répétitions par sample.
Avec répétitions	Distance au centroïde intra-ID	Répétition aberrante	Permet de flagger la mauvaise répétition plutôt que le sample entier.

Bug-hunting / supervised audits

Famille de bug potentiel	Méthodes à ajouter	Ce que ça détecte	État dans l’explorateur
Shift spectral global	Corrélation spectre moyen inter-dataset, DTW, cross-correlation, comparaison positions de pics	Décalage en longueur d'onde, mauvais alignement, interpolation différente	Partiellement calculé: cross-correlation lag et dispersion des positions de pics vs spectre moyen.
Baseline / offset / gain	Régression chaque spectre vs spectre moyen: x = a + b ref + residual; suivi de a, b, RMS résiduel	Offset additif, effet multiplicatif, dérive de baseline	Calculé dans reference.affine_*.
Mélange de lignes / mauvais appariement X-M-Y	Vérification index, hash des lignes, duplication ID, distance spectrale intra-ID, labels incohérents	Lignes mélangées, metadata mal alignées, Y attribué au mauvais spectre	Partiellement couvert par répétabilité intra-ID; checks index/hash à ajouter au pipeline canonical.
Fuite d'information / répétitions mal splitées	GroupKFold par sample_id vs StratifiedKFold random; audit des partitions par sample_id	Performance artificiellement bonne due aux répétitions	Nécessite splits et benchmark modèle; non calculé par la carte descriptive.
Label bugs	Échantillons proches en X mais Y différents, confident learning, erreurs systématiques FP/FN	Y inversés, erreurs de saisie, classes ambiguës	Nécessite Y et/ou modèle; recommandé pour l'explorateur supervisé.
Sous-domaines cachés	PCA/UMAP/t-SNE + clustering non supervisé + association avec dataset/Y/date/operator	Lots, campagnes, sondes, backgrounds non renseignés	Partiellement calculé par structure PCA/LOF; UMAP/t-SNE hors carte statique.
Artefacts localisés inconnus	Carte wavelength x dataset: différence moyenne, différence variance, KS par longueur d'onde	Régions spectrales anormales non anticipées	À calculer au niveau banque quand plusieurs datasets partagent un axe spectral.
Ruptures instrumentales	Discontinuités dans dérivées, changepoint detection	Splice, raccord détecteur, saut local non prévu	Calculé par jump/spike rates; changepoint plus avancé à ajouter.
Mélange / contamination spectrale	NMF / unmixing / reconstruction par convex hull	Composante externe: fond, plastique, sol	Non calculé automatiquement; nécessite hypothèses de composants ou grande bibliothèque.
Features instables mais prédictives	Importance modèle vs instabilité QC par variable	Modèle qui apprend un artefact plutôt qu'un signal biologique	Nécessite modèle supervisé; recommandé pour rapports de benchmark.

Variables

Targets 2

ref_sugar

target · numeric

n / missing	32 / 0
Mean ± SD	46.25 ± 12.9
Median	42.65
Range	32.7 – 68.7
CV	0.279
Skew / kurtosis	0.66 / -1.1
Normal?	no

ref_moisture

target · numeric

n / missing	32 / 0
Mean ± SD	13.57 ± 3.66
Median	14.9
Range	6.8 – 18.6
CV	0.27
Skew / kurtosis	-0.86 / -0.66
Normal?	yes

Constant metadata 6

licenseNOASSERTION
rights_statusmanual_review_needed
usage_scopeinternal_benchmark_only_until_license_review
public_release_allowedFalse
spectroscopy_typemulti_instrument_NIR_IR
notesPublic UCPH/ERDA dataset, local readme does not state explicit redistribution license.

Alignment

Alignment level	sample
Sample id available	yes
Samples	32
Observations (total)	192
Reps per sample	min 1 · mean 1 · max 1

Provenance & citation

Contributor	Data sets for multi-variate data analysis - Marzipan
Origin · url [open]	https://sid.erda.dk/share_redirect/AY8wS3D8Kn/Marzipan.zip
Origin · url [manual]	https://ucphchemometrics.com/datasets/
Origin · script [manual]	source_to_standard.py — standardization script (maintainer-only)

Governance & integrity

Tier	private
License	LicenseRef-not-cleared
Permitted use	Research and benchmarking; private use only.
Access policy	Manual download / private-use-only per source.
Redistribution	No explicit redistribution license found on checked UCPH/ERDA source.
Content version	1.0.0
Schema / protocol	2.0
Content hash	`bc87d194f274c7a3…`
Processing hash	`41b845dcc03f5249…`
Metadata hash	`ec2a55c7e3289c24…`

Load this dataset

# pip install nirs4all-datasets
from nirs4all_datasets import get

# private dataset — export requires a Dataverse token
ds = get("ucph_marzipan_multi_instrument_spectroscopy", token="…")
X, y = ds.x(), ds.y()
print(X.shape, y.shape)

Metadata downloads are available for public datasets only. The dataset bytes are never served here — fetch them from the origin / DOI above.

UCPH marzipan multi-instrument spectroscopy

Dataset property explorer

Profile axes

Spectral sources

Infratec 1255 NIT

Sampling

Signal & quality

Integrity & artefacts

Shape & reference

Outliers & repeatability

Dimensionality (PCA)

NIRSystems 6500 sample cup reflection

Sampling

Signal & quality

Integrity & artefacts

Shape & reference

Outliers & repeatability

Dimensionality (PCA)

NIRSystems 6500 fibre probe reflection

Sampling

Signal & quality

Integrity & artefacts

Shape & reference

Outliers & repeatability

Dimensionality (PCA)

Infraprover II FT-NIR

Sampling

Signal & quality

Integrity & artefacts

Shape & reference

Outliers & repeatability

Dimensionality (PCA)

Bomem MB160 Diffusir FT-NIR

Sampling

Signal & quality

Integrity & artefacts

Shape & reference

Outliers & repeatability

Dimensionality (PCA)

IR spectroscopy

Sampling

Signal & quality

Integrity & artefacts

Shape & reference

Outliers & repeatability

Dimensionality (PCA)

Metric interpretation reference

Variables

Targets 2

ref_sugar

ref_moisture

Alignment

Provenance & citation

Governance & integrity

Load this dataset