diff --git a/docs/evaluation/Ranking der Modelle im Vergleich zu den Autoren.md b/docs/evaluation/Ranking der Modelle im Vergleich zu den Autoren.md
index 3fd162808f3f63a86e890fed0bed37224c68400e..54a132d4b63ebaeecd6884e528f1ee75eb30c23d 100644
--- a/docs/evaluation/Ranking der Modelle im Vergleich zu den Autoren.md
+++ b/docs/evaluation/Ranking der Modelle im Vergleich zu den Autoren.md
@@ -2,18 +2,18 @@
| Platz nach Autoren | Model | VUS-PR nach Autoren | AUC-PR | VUS-PR | Platz nach AUC-PR | Platz nach VUS-PR |
| ------------------ | ----------- | ------------------- | ------ | ------ | ----------------- | ----------------- |
| 1 | Sub-PCA | 0.42 | 0.407 | 0.444 | 2 | 4 |
-| 2 | KShapeAD | 0.40 | 0.383 | 0.423 | | |
+| 2 | KShapeAD | 0.40 | 0.383 | 0.423 | 4 | 6 |
| 3 | POLY | 0.39 | 0.394 | 0.478 | 3 | 2 |
| 4 | KMeansAD | 0.37 | 0.444 | 0.489 | 1 | 1 |
-| 5 | USAD | 0.36 | 0.352 | 0.375 | | |
-| 6 | Sub-KNN | 0.35 | 0.292 | 0.384 | | |
-| 7 | CNN | 0.34 | 0.366 | 0.367 | | |
-| 8 | LSTMAD | 0.33 | 0.313 | 0.320 | | |
-| 9 | IForest | 0.30 | 0.304 | 0.341 | | |
-| 10 | OmniAnomaly | 0.29 | 0.354 | 0.377 | | |
-| 11 | Sub-LOF | 0.25 | 0.374 | 0.473 | | 3 |
-| 12 | Sub-OCSVM | 0.23 | 0.275 | 0.339 | | |
-| 13 | Sub-iForest | 0.22 | | | | |
+| 5 | USAD | 0.36 | 0.352 | 0.375 | 8 | 8 |
+| 6 | Sub-KNN | 0.35 | 0.351 | 0.431 | 9 | 5 |
+| 7 | CNN | 0.34 | 0.366 | 0.367 | 6 | 9 |
+| 8 | LSTMAD | 0.33 | 0.313 | 0.320 | 11 | 13 |
+| 9 | IForest | 0.30 | 0.304 | 0.341 | 12 | 11 |
+| 10 | OmniAnomaly | 0.29 | 0.354 | 0.377 | 7 | 7 |
+| 11 | Sub-LOF | 0.25 | 0.374 | 0.473 | 5 | 3 |
+| 12 | Sub-OCSVM | 0.23 | 0.275 | 0.339 | 13 | 12 |
+| 13 | Sub-iForest | 0.22 | 0.326 | 0.362 | 10 | 10 |
| 14 | LOF | 0.17 | 0.156 | 0.185 | 14 | 14 |
# Verbesserungen der Modelle
@@ -25,15 +25,16 @@
| POLY | 0.31 | 0.394 | 27.1 | 0.39 | 0.478 | 22.6 |
| KMeansAD | 0.32 | 0.444 | 38.8 | 0.37 | 0.489 | 32.2 |
| USAD | 0.32 | 0.352 | 9.4 | 0.36 | 0.375 | 4.2 |
-| Sub-KNN | 0.27 | 0.292 | 8.15 | 0.35 | 0.384 | 9.7 |
-| Sub-KNN V2 | 0.27 | | | 0.35 | | |
+| Sub-KNN V2 | 0.27 | 0.351 | 30 | 0.35 | 0.431 | 23.1 |
| CNN | 0.33 | 0.366 | 10.9 | 0.34 | 0.367 | 7.9 |
| LSTMAD | 0.31 | 0.313 | - | 0.33 | 0.320 | 3.0 |
| IForest | 0.29 | 0.304 | 4.8 | 0.30 | 0.341 | 13.7 |
-| Sub-iForest | 0.16 | | | 0.22 | | |
+| Sub-iForest | 0.16 | 0.326 | 103.8 | 0.22 | 0.362 | 64.5 |
| OmniAnomaly | 0.27 | 0.354 | 31.1 | 0.29 | 0.377 | 30 |
| Sub-LOF | 0.16 | 0.374 | 133.8 | 0.25 | 0.473 | 89.2 |
| Sub-OCSVM | 0.16 | 0.275 | 71.9 | 0.23 | 0.339 | 47.4 |
| LOF | 0.14 | 0.156 | 11.4 | 0.17 | 0.185 | 8.8 |
-| **Durchschnitt** | | | $\approx$ 28.21 | | | $\approx$ 21.55 |
+| **Durchschnitt** | | | $\approx$ 35.17 | | | $\approx$ 25.58 |
+| **Range** | | 0.288 | 133.8 | | 0.304 | 86.2 |
+| *Alt*: Sub-KNN | 0.27 | 0.292 | 8.15 | 0.35 | 0.384 | 9.7 |
diff --git a/docs/evaluation/Reproducibility.md b/docs/evaluation/Reproducibility.md
index 027b9b8c5839094ef5a01cd016eb7d2367ec120f..53ab3f4c5263a97ba87c886263d2542f709ff52e 100644
--- a/docs/evaluation/Reproducibility.md
+++ b/docs/evaluation/Reproducibility.md
@@ -76,9 +76,8 @@
**Allgemein:**
* Da eine Grid-Suche pro Zeitreihe durchgeführt wurde und nicht pauschal die optimalen Hyperparameter der Autoren für alle Zeitserien verwendet wurden, ist eine Verbesserung der Modelle zu erwarten
+ In einigen Datensätzen bei einigen Modellen sind die Ergebnisse jedoch nicht reproduzierbar, selbst mit den optimalen Hyperparametern der Autoren (meist noch schlechtere Werte). Hier nur Vergleich mit VUS-PR möglich, da zu AUC-PR keine Daten pro Datensatz zu Model.
- *Missing: Sub-IForest und KNN*
*Betreffende Datensätze:*
- + **TAO** (Modelle: **PCA**: -0.09; **KShapeAD**: - 0.19; **Poly**: - 0.11; **KMeans**: - 0.17; **IForest**: - 0.26; **Sub-LOF**: - 0.11; **Sub-OCSVM**: - 0.2; **LOF** - 0.2, **Sub-IForest**: -0.17; )
+ + **TAO** (Modelle: **PCA**: -0.09; **KShapeAD**: - 0.19; **Poly**: - 0.11; **KMeans**: - 0.17; **IForest**: - 0.26; **Sub-LOF**: - 0.11; **Sub-OCSVM**: - 0.2; **LOF** - 0.2, **Sub-IForest**: -0.17; **Sub-KNN**: - 0.17; )
+ **Stock** (Modelle: **IForest**: - 0.25; )
+ **MSL** (Modelle: **KShapeAD**: -0.09; )
+ **IOPS**: (Modelle: **IForest**: - 0.15; )
@@ -89,18 +88,18 @@
+ **SMAP** (Modelle: **Sub-IForest**: - 0.11; )
+ **TODS** (Modelle: **Sub-IForest**: - 0.16; )
*Stark Verbesserte Datensätze bei einzelnen Modellen:*
-+ **Power** (Modelle: **KShapeAD**: + 0.32; **KMeans**: + 0.13; )
++ **Power** (Modelle: **KShapeAD**: + 0.32; **KMeans**: + 0.13; **Sub-KNN**: + 0.12; )
+ **MSL** (Modelle: **POLY**: + 0.19; **IForest**: + 0.16; **OA**: + 0.18; **Sub-LOF**: + 0.19; **Sub-IForest**: + 0.12; )
+ **OPPORTUNITY** (Modelle: **Poly**: + 0.33; **CNN**: +0.22; **IForest**: + 0.31; **Sub-IForest**: + 0.69; )
-+ **Exathlon:** (Modelle: **Poly**: + 0.11; **KMeans**: + 0.21; **IForest**:+ 0.29; **Sub-OCSVM**: + 0.27; **Sub-IForest**: + 0.47; )
-+ **NAB** (Modelle: **Poly**: + 0.1; **IForest**: + 0.12; **OA**: + 0.15; **Sub-LOF**: + 0.21; **Sub-OCSVM**: + 0.12; **Sub-IForest**: + 0.1; )
++ **Exathlon:** (Modelle: **Poly**: + 0.11; **KMeans**: + 0.21; **IForest**:+ 0.29; **Sub-OCSVM**: + 0.27; **Sub-IForest**: + 0.47; **Sub-KNN**: + 0.13; )
++ **NAB** (Modelle: **Poly**: + 0.1; **IForest**: + 0.12; **OA**: + 0.15; **Sub-LOF**: + 0.21; **Sub-OCSVM**: + 0.12; **Sub-IForest**: + 0.1; **Sub-KNN**: + 0.12; )
+ **NEK** (Modelle: **Poly**: + 0.15; **KMeans:** + 0.12; **USAD**: + 0.1; **Sub-LOF**: + 0.44; **Sub-OCSVM**: + 0.11; **Sub-IForest**: + 0.54; )
+ **WSD** (Modelle: **Poly**: + 0.12; **USAD**: + 0.12; **Sub-LOF**: + 0.52; )
-+ **LTDB** (Modelle: **KMeans**: + 0.35; **IForest**: + 0.2; **Sub-OCSVM**: + 0.36; )
-+ **MITDB** (Modelle: **KMeans**: + 0.32; **Sub-LOF**: + 0.11; **Sub-OCSVM**: + 0.21; )
++ **LTDB** (Modelle: **KMeans**: + 0.35; **IForest**: + 0.2; **Sub-OCSVM**: + 0.36; **Sub-KNN**: + 0.13; )
++ **MITDB** (Modelle: **KMeans**: + 0.32; **Sub-LOF**: + 0.11; **Sub-OCSVM**: + 0.21; **Sub-KNN**: + 0.19; )
+ **SVDB** (Modelle: **KMeans**: + 0.37; **IForest**: + 0.22; **OA**: + 0.36; **Sub-LOF**: + 0.2; **Sub-OCSVM**: + 0.39; )
-+ **UCR** (Modelle: **KMeans**: + 0.11; **Sub-LOF**: + 0.22; **Sub-OCSVM**: + 0.16; )
-+ **Yahoo** (Modelle: **KMeans**: + 0.23; **Sub-LOF**: + 0.15; **Sub-IForest**: + 0.1; )
++ **UCR** (Modelle: **KMeans**: + 0.11; **Sub-LOF**: + 0.22; **Sub-OCSVM**: + 0.16; **Sub-KNN**: + 0.11; )
++ **Yahoo** (Modelle: **KMeans**: + 0.23; **Sub-LOF**: + 0.15; **Sub-IForest**: + 0.1; **Sub-KNN**: + 0.21; )
+ **Daphnet** (Modelle: **IForest**: + 0.16; **Sub-LOF**: + 0.21; **Sub-IForest**: +0.46; )
+ **SED** (Modelle: **IForest**: + 0.46; **Sub-OCSVM**: + 0.1; )
+ **Catsv2** (Modelle: **OA**: + 0.16; )
@@ -212,30 +211,30 @@
*Anpassen nach hinzufügen von Sub-IForest und ggf. neues KNN*
-| Datensatz | AUC-PR | VUS-PR | Differenz AUC zu VUS-PR | Differenz prozentual |
-| ------------ | ------ | ------ | ----------------------- | -------------------- |
-| **Exathlon** | 0.658 | 0.659 | 0.001 | |
-| **NAB** | 0.341 | 0.367 | 0.026 | |
-| **NEK** | 0.566 | 0.592 | 0.026 | |
-| **LTDB** | 0.485 | 0.545 | *0.06* | |
-| **Dapthnet** | 0.227 | 0.217 | -0.01 | |
-| **SWaT** | 0.434 | 0.302 | *-0.132* | |
-| **Power** | 0.174 | 0.172 | -0.002 | |
-| **TAO** | 0.297 | 0.817 | *0.52* | |
-| **TODS** | 0.252 | 0.614 | *0.362* | |
-| **Stock** | 0.214 | 0.769 | *0.555* | |
-| MITDB | 0.313 | 0.300 | | |
-| MSL | 0.357 | 0.412 | | |
-| YAHOO | 0.268 | 0.374 | | |
-| OPPORTUNITY | 0.464 | 0.464 | | |
-| SMAP | 0.426 | 0.470 | | |
-| MGAB | 0.095 | 0.081 | | |
-| UCR | 0.210 | 0.219 | | |
-| SVDB | 0.441 | 0.440 | | |
-| SMD | 0.396 | 0.381 | | |
-| WSD | 0.206 | 0.188 | | |
-| SED | 0.260 | 0.332 | | |
-| IOPS | 0.226 | 0.187 | | |
-| CATSv2 | 0.394 | 0.266 | | |
-| **Gesamt:** | 0.335 | 0.399 | | |
+| Datensatz | AUC-PR | VUS-PR | Differenz AUC zu VUS-PR |
+| ------------ | ------ | ------ | ----------------------- |
+| **Exathlon** | 0.682 | 0.683 | 0.001 |
+| **NAB** | 0.346 | 0.371 | 0.025 |
+| **NEK** | 0.590 | 0.614 | 0.024 |
+| **LTDB** | 0.490 | 0.549 | *0.069* |
+| **Dapthnet** | 0.247 | 0.240 | - 0.007 |
+| **SWaT** | 0.453 | 0.309 | *- 0.144* |
+| **Power** | 0.179 | 0.178 | - 0.001 |
+| **TAO** | 0.284 | 0.815 | *0.531* |
+| **TODS** | 0.241 | 0.603 | *0.362* |
+| **Stock** | 0.210 | 0.769 | *0.559* |
+| MITDB | 0.305 | 0.293 | - 0.012 |
+| MSL | 0.361 | 0.413 | 0.052 |
+| YAHOO | 0.263 | 0.360 | 0.103 |
+| OPPORTUNITY | 0.488 | 0.488 | 0.00 |
+| SMAP | 0.423 | 0.463 | 0.040 |
+| MGAB | 0.088 | 0.080 | 0.00 |
+| UCR | 0.213 | 0.219 | 0.006 |
+| SVDB | 0.436 | 0.436 | 0.00 |
+| SMD | 0.398 | 0.381 | - 0.017 |
+| WSD | 0.198 | 0.180 | - 0.018 |
+| SED | 0.275 | 0.358 | 0.088 |
+| IOPS | 0.225 | 0.184 | - 0.041 |
+| CATSv2 | 0.378 | 0.257 | - 0.121 |
+| **Gesamt:** | 0.335 | 0.399 | 0.064 |
diff --git a/src/group_evaluation/create_heatmap.ipynb b/src/group_evaluation/create_heatmap.ipynb
index 07d16e8e566fdcfb434acf85ec35d57b76a30619..cc245bef517dfef99a77e18d72ba7d926f640589 100644
--- a/src/group_evaluation/create_heatmap.ipynb
+++ b/src/group_evaluation/create_heatmap.ipynb
@@ -12,7 +12,7 @@
},
{
"cell_type": "code",
- "execution_count": 7,
+ "execution_count": 10,
"metadata": {},
"outputs": [
{
@@ -245,7 +245,7 @@
"[5 rows x 23 columns]"
]
},
- "execution_count": 7,
+ "execution_count": 10,
"metadata": {},
"output_type": "execute_result"
}
@@ -1932,7 +1932,7 @@
},
{
"cell_type": "code",
- "execution_count": 9,
+ "execution_count": 12,
"metadata": {},
"outputs": [
{
@@ -1978,7 +1978,7 @@
},
{
"cell_type": "code",
- "execution_count": 10,
+ "execution_count": 13,
"metadata": {},
"outputs": [
{
@@ -1990,7 +1990,7 @@
" dtype='object', name='Model')"
]
},
- "execution_count": 10,
+ "execution_count": 13,
"metadata": {},
"output_type": "execute_result"
}
diff --git a/src/models/desi/desi_evaluate_groups.ipynb b/src/models/desi/desi_evaluate_groups.ipynb
index 14a616c7b029798574639238e1da6898edd8110f..0a2f81a6b8fc1191b87c8ddf782ecc503ca9d980 100644
--- a/src/models/desi/desi_evaluate_groups.ipynb
+++ b/src/models/desi/desi_evaluate_groups.ipynb
@@ -5976,47 +5976,47 @@
},
{
"cell_type": "code",
- "execution_count": 31,
+ "execution_count": 32,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"group\n",
- "CATSv2 0.361837\n",
"Daphnet 0.037377\n",
- "Exathlon 0.597445\n",
"IOPS 0.105166\n",
- "LTDB 0.709125\n",
+ "WSD 0.106099\n",
+ "SWaT 0.110668\n",
+ "SMD 0.193423\n",
+ "NEK 0.273441\n",
"MGAB 0.301167\n",
- "MITDB 0.548677\n",
"MSL 0.321555\n",
- "NAB 0.373190\n",
- "NEK 0.273441\n",
- "OPPORTUNITY 0.333758\n",
"Power 0.333374\n",
- "SED 0.882987\n",
+ "OPPORTUNITY 0.333758\n",
+ "CATSv2 0.361837\n",
+ "NAB 0.373190\n",
+ "UCR 0.475202\n",
+ "YAHOO 0.521357\n",
+ "MITDB 0.548677\n",
"SMAP 0.573129\n",
- "SMD 0.193423\n",
+ "Exathlon 0.597445\n",
"SVDB 0.617909\n",
- "SWaT 0.110668\n",
+ "LTDB 0.709125\n",
"Stock 0.719049\n",
- "TAO 0.752695\n",
"TODS 0.742028\n",
- "UCR 0.475202\n",
- "WSD 0.106099\n",
- "YAHOO 0.521357\n",
+ "TAO 0.752695\n",
+ "SED 0.882987\n",
"Name: VUS-PR, dtype: float64"
]
},
- "execution_count": 31,
+ "execution_count": 32,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"#Gruppen aufsteigend sortiert nach durchschnittlichem VUS-PR wert\n",
- "df_groups['VUS-PR'].mean().sort_index()"
+ "df_groups['VUS-PR'].mean().sort_values()"
]
},
{