Regression¶

---

This example shows how to use ATOM to apply PCA on the data and run a regression pipeline.

Download the abalone dataset from https://archive.ics.uci.edu/ml/datasets/Abalone. The goal of this dataset is to predict the rings (age) of abalone shells from physical measurements.

Load the data¶

In [1]:

# Import packages
import pandas as pd
from atom import ATOMRegressor

# Import packages import pandas as pd from atom import ATOMRegressor

In [2]:

# Load the data
X = pd.read_csv("./datasets/abalone.csv")

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

# Load the data X = pd.read_csv("./datasets/abalone.csv") # Let's have a look X.head()

Out[2]:

	Sex	Length	Diameter	Height	Whole weight	Shucked weight	Viscera weight	Shell weight	Rings
0	M	0.455	0.365	0.095	0.5140	0.2245	0.1010	0.150	15
1	M	0.350	0.265	0.090	0.2255	0.0995	0.0485	0.070	7
2	F	0.530	0.420	0.135	0.6770	0.2565	0.1415	0.210	9
3	M	0.440	0.365	0.125	0.5160	0.2155	0.1140	0.155	10
4	I	0.330	0.255	0.080	0.2050	0.0895	0.0395	0.055	7

In [3]:

# Initialize atom for regression tasks
atom = ATOMRegressor(X, "Rings", verbose=2, warnings=False, random_state=42)

# Initialize atom for regression tasks atom = ATOMRegressor(X, "Rings", verbose=2, warnings=False, random_state=42)

<< ================== ATOM ================== >>
Algorithm task: regression.

Dataset stats ==================== >>
Shape: (4177, 9)
Memory: 509.72 kB
Scaled: False
Categorical features: 1 (12.5%)
Outlier values: 196 (0.7%)
-------------------------------------
Train set size: 3342
Test set size: 835

In [4]:

# Encode the categorical features
atom.encode()

# Encode the categorical features atom.encode()

Fitting Encoder...
Encoding categorical columns...
 --> OneHot-encoding feature Sex. Contains 3 classes.

In [5]:

# Plot the dataset's correlation matrix
atom.plot_correlation()

# Plot the dataset's correlation matrix atom.plot_correlation()

In [6]:

# Apply PCA for dimensionality reduction
atom.feature_selection(strategy="pca", n_features=6)

# Apply PCA for dimensionality reduction atom.feature_selection(strategy="pca", n_features=6)

Fitting FeatureSelector...
Performing feature selection ...
 --> Applying Principal Component Analysis...
   >>> Scaling features...
   >>> Explained variance ratio: 0.977

In [7]:

# Note that the fetaures are automatically renamed to Component 1, 2, etc...
atom.columns

# Note that the fetaures are automatically renamed to Component 1, 2, etc... atom.columns

Out[7]:

['component 1',
 'component 2',
 'component 3',
 'component 4',
 'component 5',
 'component 6',
 'Rings']

In [8]:

# Use the plotting methods to see the retained variance ratio
atom.plot_pca()

# Use the plotting methods to see the retained variance ratio atom.plot_pca()

In [9]:

atom.plot_components(figsize=(8, 6))

atom.plot_components(figsize=(8, 6))

Run the pipeline¶

In [10]:

atom.run(
    models=["Tree", "Bag", "ET"],
    metric="MSE",
    n_calls=5,
    n_initial_points=2,
    bo_params={"base_estimator": "GBRT"},
    n_bootstrap=5,
)

atom.run( models=["Tree", "Bag", "ET"], metric="MSE", n_calls=5, n_initial_points=2, bo_params={"base_estimator": "GBRT"}, n_bootstrap=5, )

Training ========================= >>
Models: Tree, Bag, ET
Metric: neg_mean_squared_error


Running BO for Decision Tree...
| call             |   criterion | splitter | max_depth | min_samples_split | min_samples_leaf | max_features | ccp_alpha | neg_mean_squared_error | best_neg_mean_squared_error |    time | total_time |
| ---------------- | ----------- | -------- | --------- | ----------------- | ---------------- | ------------ | --------- | ---------------------- | --------------------------- | ------- | ---------- |
| Initial point 1  | absolute_.. |   random |         7 |                 8 |               19 |         auto |    0.0161 |                -8.1076 |                     -8.1076 |  0.108s |     0.114s |
| Initial point 2  | absolute_.. |     best |         6 |                 3 |               12 |         None |       0.0 |                -6.4884 |                     -6.4884 |  0.244s |     0.628s |
| Iteration 3      | absolute_.. |     best |         6 |                 4 |                1 |         None |    0.0086 |                -6.8759 |                     -6.4884 |  0.261s |     1.089s |
| Iteration 4      |     poisson |     best |         7 |                 3 |               17 |         auto |    0.0039 |                -9.2175 |                     -6.4884 |  0.076s |     1.359s |
| Iteration 5      |     poisson |     best |         6 |                 7 |               14 |         None |    0.0004 |                -8.9588 |                     -6.4884 |  0.074s |     1.627s |
Bayesian Optimization ---------------------------
Best call --> Initial point 2
Best parameters --> {'criterion': 'absolute_error', 'splitter': 'best', 'max_depth': 6, 'min_samples_split': 3, 'min_samples_leaf': 12, 'max_features': None, 'ccp_alpha': 0.0}
Best evaluation --> neg_mean_squared_error: -6.4884
Time elapsed: 1.830s
Fit ---------------------------------------------
Train evaluation --> neg_mean_squared_error: -5.8483
Test evaluation --> neg_mean_squared_error: -7.021
Time elapsed: 0.277s
Bootstrap ---------------------------------------
Evaluation --> neg_mean_squared_error: -7.1589 ± 0.2396
Time elapsed: 1.181s
-------------------------------------------------
Total time: 3.289s


Running BO for Bagging...
| call             | n_estimators | max_samples | max_features | bootstrap | bootstrap_features | neg_mean_squared_error | best_neg_mean_squared_error |    time | total_time |
| ---------------- | ------------ | ----------- | ------------ | --------- | ------------------ | ---------------------- | --------------------------- | ------- | ---------- |
| Initial point 1  |          112 |         0.9 |          0.6 |     False |              False |                -6.5592 |                     -6.5592 |  0.978s |     0.984s |
| Initial point 2  |          131 |         0.5 |          0.5 |     False |              False |                -5.4837 |                     -5.4837 |  0.687s |     1.757s |
| Iteration 3      |          302 |         0.5 |          0.5 |      True |               True |                -6.0919 |                     -5.4837 |  1.236s |     3.176s |
| Iteration 4      |          191 |         0.5 |          0.5 |     False |              False |                -5.3972 |                     -5.3972 |  0.989s |     4.541s |
| Iteration 5      |          217 |         0.5 |          0.5 |     False |              False |                -4.9339 |                     -4.9339 |  1.136s |     5.882s |
Bayesian Optimization ---------------------------
Best call --> Iteration 5
Best parameters --> {'n_estimators': 217, 'max_samples': 0.5, 'max_features': 0.5, 'bootstrap': False, 'bootstrap_features': False}
Best evaluation --> neg_mean_squared_error: -4.9339
Time elapsed: 6.093s
Fit ---------------------------------------------
Train evaluation --> neg_mean_squared_error: -1.3974
Test evaluation --> neg_mean_squared_error: -5.7349
Time elapsed: 1.417s
Bootstrap ---------------------------------------
Evaluation --> neg_mean_squared_error: -5.9024 ± 0.058
Time elapsed: 6.126s
-------------------------------------------------
Total time: 13.638s


Running BO for Extra-Trees...
| call             | n_estimators |     criterion | max_depth | min_samples_split | min_samples_leaf | max_features | bootstrap | ccp_alpha | max_samples | neg_mean_squared_error | best_neg_mean_squared_error |    time | total_time |
| ---------------- | ------------ | ------------- | --------- | ----------------- | ---------------- | ------------ | --------- | --------- | ----------- | ---------------------- | --------------------------- | ------- | ---------- |
| Initial point 1  |          112 | absolute_er.. |         1 |                 9 |                7 |          0.6 |      True |    0.0161 |         0.6 |               -10.9838 |                    -10.9838 |  0.492s |     0.499s |
| Initial point 2  |          369 | absolute_er.. |      None |                 3 |               12 |         None |      True |    0.0347 |         0.7 |                -6.8941 |                     -6.8941 |  5.217s |     5.810s |
| Iteration 3      |          383 | absolute_er.. |      None |                 6 |               20 |         None |      True |    0.0271 |         0.8 |                -7.5893 |                     -6.8941 |  5.447s |    11.471s |
| Iteration 4      |          221 | absolute_er.. |         1 |                 4 |                1 |         log2 |      True |    0.0339 |         0.7 |                   -9.5 |                     -6.8941 |  0.929s |    12.611s |
| Iteration 5      |          470 | squared_error |      None |                 3 |               13 |          0.6 |      True |    0.0162 |         0.6 |                -5.8203 |                     -5.8203 |  0.525s |    13.345s |
Bayesian Optimization ---------------------------
Best call --> Iteration 5
Best parameters --> {'n_estimators': 470, 'criterion': 'squared_error', 'max_depth': None, 'min_samples_split': 3, 'min_samples_leaf': 13, 'max_features': 0.6, 'bootstrap': True, 'ccp_alpha': 0.0162, 'max_samples': 0.6}
Best evaluation --> neg_mean_squared_error: -5.8203
Time elapsed: 13.568s
Fit ---------------------------------------------
Train evaluation --> neg_mean_squared_error: -6.3561
Test evaluation --> neg_mean_squared_error: -6.613
Time elapsed: 0.565s
Bootstrap ---------------------------------------
Evaluation --> neg_mean_squared_error: -6.56 ± 0.0946
Time elapsed: 2.500s
-------------------------------------------------
Total time: 16.634s


Final results ==================== >>
Duration: 33.563s
-------------------------------------
Decision Tree --> neg_mean_squared_error: -7.1589 ± 0.2396 ~
Bagging       --> neg_mean_squared_error: -5.9024 ± 0.058 ~ !
Extra-Trees   --> neg_mean_squared_error: -6.56 ± 0.0946 ~

Analyze the results¶

In [11]:

# Use the errors or residuals plots to check the model performances
atom.plot_residuals()

# Use the errors or residuals plots to check the model performances atom.plot_residuals()

In [12]:

atom.plot_errors()

atom.plot_errors()

In [13]:

# Analyze the relation between the target response and the features
atom.n_jobs = 8  # The method can be slow...
atom.ET.plot_partial_dependence(columns=(0, (2, 3)), figsize=(12, 5))

# Analyze the relation between the target response and the features atom.n_jobs = 8 # The method can be slow... atom.ET.plot_partial_dependence(columns=(0, (2, 3)), figsize=(12, 5))