Multiclass classification¶

---

This example shows how to compare the performance of three models on a multiclass classification task.

Import the wine dataset from sklearn.datasets. This is a small and easy to train dataset whose goal is to predict wines into three groups (which cultivator it's from) using features based on the results of chemical analysis.

Load the data¶

In [8]:

# Import packages
from sklearn.datasets import load_wine
from atom import ATOMClassifier

# Import packages from sklearn.datasets import load_wine from atom import ATOMClassifier

In [9]:

# Load data
X, y = load_wine(return_X_y=True, as_frame=True)

# Let's have a look
X.head()

# Load data X, y = load_wine(return_X_y=True, as_frame=True) # Let's have a look X.head()

Out[9]:

	alcohol	malic_acid	ash	alcalinity_of_ash	magnesium	total_phenols	flavanoids	nonflavanoid_phenols	proanthocyanins	color_intensity	hue	od280/od315_of_diluted_wines	proline
0	14.23	1.71	2.43	15.6	127.0	2.80	3.06	0.28	2.29	5.64	1.04	3.92	1065.0
1	13.20	1.78	2.14	11.2	100.0	2.65	2.76	0.26	1.28	4.38	1.05	3.40	1050.0
2	13.16	2.36	2.67	18.6	101.0	2.80	3.24	0.30	2.81	5.68	1.03	3.17	1185.0
3	14.37	1.95	2.50	16.8	113.0	3.85	3.49	0.24	2.18	7.80	0.86	3.45	1480.0
4	13.24	2.59	2.87	21.0	118.0	2.80	2.69	0.39	1.82	4.32	1.04	2.93	735.0

Run the pipeline¶

In [10]:

atom = ATOMClassifier(X, y, n_jobs=-1, warnings=False, verbose=2, random_state=1)

# Fit the pipeline with the selected models
atom.run(
    models=["LR","LDA", "RF"],
    metric="roc_auc_ovr",
    n_calls=4,
    n_initial_points=3,
    bo_params={"base_estimator": "rf", "max_time": 100},
    n_bootstrap=5,
)

atom = ATOMClassifier(X, y, n_jobs=-1, warnings=False, verbose=2, random_state=1) # Fit the pipeline with the selected models atom.run( models=["LR","LDA", "RF"], metric="roc_auc_ovr", n_calls=4, n_initial_points=3, bo_params={"base_estimator": "rf", "max_time": 100}, n_bootstrap=5, )

<< ================== ATOM ================== >>
Algorithm task: multiclass classification.
Parallel processing with 16 cores.

Dataset stats ==================== >>
Shape: (178, 14)
Scaled: False
Outlier values: 10 (0.5%)
-------------------------------------
Train set size: 143
Test set size: 35
-------------------------------------
|    |    dataset |      train |       test |
| -- | ---------- | ---------- | ---------- |
| 0  |   59 (1.2) |   50 (1.4) |    9 (1.0) |
| 1  |   71 (1.5) |   58 (1.7) |   13 (1.4) |
| 2  |   48 (1.0) |   35 (1.0) |   13 (1.4) |


Training ========================= >>
Models: LR, LDA, RF
Metric: roc_auc_ovr


Running BO for Logistic Regression...
| call             | penalty |       C |  solver | max_iter | l1_ratio | roc_auc_ovr | best_roc_auc_ovr |    time | total_time |
| ---------------- | ------- | ------- | ------- | -------- | -------- | ----------- | ---------------- | ------- | ---------- |
| Initial point 1  |    none |     --- |   lbfgs |      745 |      --- |      0.9242 |           0.9242 |  0.766s |     0.777s |
| Initial point 2  |      l2 |  0.0085 | newto.. |      490 |      --- |      0.9897 |           0.9897 |  0.344s |     1.262s |
| Initial point 3  |      l2 |  0.0368 | libli.. |      352 |      --- |         1.0 |              1.0 |  0.016s |     1.341s |
| Iteration 4      |      l2 |  5.6426 | newto.. |      378 |      --- |         1.0 |              1.0 |  0.422s |     2.060s |
Bayesian Optimization ---------------------------
Best call --> Iteration 4
Best parameters --> {'penalty': 'l2', 'C': 5.6426, 'solver': 'newton-cg', 'max_iter': 378}
Best evaluation --> roc_auc_ovr: 1.0
Time elapsed: 2.325s
Fit ---------------------------------------------
Train evaluation --> roc_auc_ovr: 1.0
Test evaluation --> roc_auc_ovr: 0.9965
Time elapsed: 0.297s
Bootstrap ---------------------------------------
Evaluation --> roc_auc_ovr: 0.996 ± 0.0009
Time elapsed: 1.531s
-------------------------------------------------
Total time: 4.154s


Running BO for Linear Discriminant Analysis...
| call             |  solver | shrinkage | roc_auc_ovr | best_roc_auc_ovr |    time | total_time |
| ---------------- | ------- | --------- | ----------- | ---------------- | ------- | ---------- |
| Initial point 1  |   eigen |       1.0 |      0.9125 |           0.9125 |  0.016s |     0.016s |
| Initial point 2  |     svd |       --- |         1.0 |              1.0 |  0.000s |     0.094s |
| Initial point 3  |     svd |       --- |         1.0 |              1.0 |  0.000s |     0.172s |
| Iteration 4      |    lsqr |       0.8 |      0.8783 |              1.0 |  0.016s |     0.422s |
Bayesian Optimization ---------------------------
Best call --> Initial point 2
Best parameters --> {'solver': 'svd'}
Best evaluation --> roc_auc_ovr: 1.0
Time elapsed: 0.656s
Fit ---------------------------------------------
Train evaluation --> roc_auc_ovr: 1.0
Test evaluation --> roc_auc_ovr: 1.0
Time elapsed: 0.000s
Bootstrap ---------------------------------------
Evaluation --> roc_auc_ovr: 0.9966 ± 0.0024
Time elapsed: 0.031s
-------------------------------------------------
Total time: 0.688s


Running BO for Random Forest...
| call             | n_estimators | criterion | max_depth | min_samples_split | min_samples_leaf | max_features | bootstrap | ccp_alpha | max_samples | roc_auc_ovr | best_roc_auc_ovr |    time | total_time |
| ---------------- | ------------ | --------- | --------- | ----------------- | ---------------- | ------------ | --------- | --------- | ----------- | ----------- | ---------------- | ------- | ---------- |
| Initial point 1  |          245 |   entropy |      None |                13 |                6 |         sqrt |      True |    0.0065 |         0.6 |      0.9966 |           0.9966 |  0.266s |     0.281s |
| Initial point 2  |          400 |   entropy |         8 |                 7 |               19 |          0.6 |      True |     0.008 |         0.7 |      0.9886 |           0.9966 |  0.422s |     0.781s |
| Initial point 3  |           78 |      gini |         5 |                 2 |               14 |          0.8 |     False |    0.0032 |         --- |      0.9916 |           0.9966 |  0.078s |     0.938s |
| Iteration 4      |          394 |   entropy |         3 |                19 |               14 |          0.8 |     False |    0.0148 |         --- |      0.9841 |           0.9966 |  0.344s |     1.672s |
Bayesian Optimization ---------------------------
Best call --> Initial point 1
Best parameters --> {'n_estimators': 245, 'criterion': 'entropy', 'max_depth': None, 'min_samples_split': 13, 'min_samples_leaf': 6, 'max_features': 'sqrt', 'bootstrap': True, 'ccp_alpha': 0.0065, 'max_samples': 0.6}
Best evaluation --> roc_auc_ovr: 0.9966
Time elapsed: 2.235s
Fit ---------------------------------------------
Train evaluation --> roc_auc_ovr: 1.0
Test evaluation --> roc_auc_ovr: 0.9883
Time elapsed: 0.319s
Bootstrap ---------------------------------------
Evaluation --> roc_auc_ovr: 0.98 ± 0.0095
Time elapsed: 1.379s
-------------------------------------------------
Total time: 3.933s


Final results ==================== >>
Duration: 8.774s
-------------------------------------
Logistic Regression          --> roc_auc_ovr: 0.996 ± 0.0009
Linear Discriminant Analysis --> roc_auc_ovr: 0.9966 ± 0.0024 !
Random Forest                --> roc_auc_ovr: 0.98 ± 0.0095

Analyze the results¶

In [11]:

atom.results

atom.results

Out[11]:

	metric_bo	time_bo	metric_train	metric_test	time_fit	mean_bootstrap	std_bootstrap	time_bootstrap	time
LR	1.000000	2.325s	1.0	0.996503	0.297s	0.996037	0.000932	1.531s	4.154s
LDA	1.000000	0.656s	1.0	1.000000	0.000s	0.996633	0.002441	0.031s	0.688s
RF	0.996612	2.235s	1.0	0.988345	0.319s	0.979953	0.009537	1.379s	3.933s

In [12]:

# Show the score for some different metrics
atom.evaluate(["precision_macro", "recall_macro", "jaccard_weighted"])

# Show the score for some different metrics atom.evaluate(["precision_macro", "recall_macro", "jaccard_weighted"])

Out[12]:

	precision_macro	recall_macro	jaccard_weighted
LR	0.976190	0.974359	0.944898
LDA	1.000000	1.000000	1.000000
RF	0.955556	0.948718	0.893333

In [13]:

# Some plots allow you to choose the target class to look at
atom.rf.plot_probabilities(dataset="train", target=2)

# Some plots allow you to choose the target class to look at atom.rf.plot_probabilities(dataset="train", target=2)

In [14]:

atom.lda.heatmap_plot(target=2, show=10, figsize=(8, 5))

atom.lda.heatmap_plot(target=2, show=10, figsize=(8, 5))