stepwise

In [ ]:
stepwise(estimator,
         input_relation: (str, vDataFrame),
         X: list,
         y: str,
         criterion: str = "bic",
         direction: str = "backward",
         max_steps: int = 100,
         criterion_threshold: int = 3,
         drop_final_estimator: bool = True,
         x_order: str = "pearson",
         print_info: bool = True,
         show: bool = True,
         ax=None,
         **style_kwds,)

Uses the stepwise algorithm to find the most suitable number of features when fitting the estimator.

Parameters

Name Type Optional Description
estimator
object
❌
 Vertica estimator with a fit method. This can be either a regressor or a binary classifier. Multiclass classifiers are not supported.
input_relation
str / vDataFrame
❌
 Input Relation.
X
list
❌
 List of the predictor columns.
y
str
❌
 Response Column.
criterion
str
✓
 Criterion used to evaluate the model.
  • aic : Akaike’s Information Criterion.
  • bic : Bayesian Information Criterion.
direction
str
✓
 How to start the stepwise search. Can be done 'backward' or 'forward'.
max_steps
int
✓
 The maximum number of steps to be considered.
criterion_threshold
int
✓
 Threshold used when comparing the models criterions. If the difference is lesser than the threshold then the current 'best' model is changed.
drop_final_estimator
bool
✓
 If set to True, the final estimator will be dropped.
x_order
str
✓
 How to preprocess X before using the stepwise algorithm.
  • pearson : X is ordered based on the Pearson's correlation coefficient.
  • spearman : X is ordered based on the Spearman's correlation coefficient.
  • random : Shuffles the vector X before applying the stepwise algorithm.
  • none : Does not change X order.
print_info
bool
✓
 If set to True, prints the model information at each step.
show
bool
✓
 If set to True, the Stepwise graphic will be drawn.
ax
Matplotlib axes object
✓
 The axes to plot on.
**style_kwds
any
✓
 Any optional parameter to pass to the Matplotlib functions.

Returns

tablesample : An object containing the result. For more information, see utilities.tablesample.

Example

In [4]:
from verticapy.learn.linear_model import LogisticRegression
model = LogisticRegression(name = "public.LR_titanic",
                           tol = 1e-4,
                           max_iter = 100, 
                           solver = 'Newton')

from verticapy.learn.model_selection import stepwise

# backward
stepwise(model,
         input_relation = "public.titanic", 
         X = ["age", "fare", "parch", "pclass",], 
         y = "survived",)

Starting Stepwise

[Model 0] bic: -1574.6084573061446; Variables: ['"age"', '"parch"', '"fare"', '"pclass"']
[Model 1] bic: -1941.0852092129517; (-) Variable: "age"
[Model 2] bic: -1940.602698869082; (-) Variable: "parch"
[Model 3] bic: -1937.9168392321783; (-) Variable: "fare"

Selected Model

[Model 3] bic: -1937.9168392321783; Variables: ['"pclass"']
Out[4]:
bic
change
variable
importance
0-1574.6084573061446[null][null]0.0
1-1941.0852092129517-"age"0.0
2-1940.602698869082-"parch"0.3359363423491798
3-1937.9168392321783-"fare"1.8699658441482787
4-1797.4537281723472+"pclass"97.79409781350255
Rows: 1-5 | Columns: 6
In [5]:
# forward
stepwise(model,
         input_relation = "public.titanic", 
         X = ["age", "fare", "parch", "pclass",], 
         y = "survived",
         direction = "forward",)

Starting Stepwise

[Model 0] bic: -1797.4537281723472; Variables: []
[Model 1] bic: -1937.9168392321783; (+) Variable: "pclass"
[Model 2] bic: -1943.8577653611087; (+) Variable: "parch"

Selected Model

[Model 2] bic: -1943.8577653611087; Variables: ['"pclass"', '"parch"']
Out[5]:
features
bic
change
variable
importance
0[]-1797.4537281723472[null][null]0.0
1['pclass']-1937.9168392321783+"pclass"94.21370196806726
2['pclass', 'fare']-1940.602698869082-"fare"1.8015034513333006
3['pclass', 'parch']-1943.8577653611087+"parch"3.984794580599448
4['pclass', 'parch', 'age']-1580.6419110115492-"age"0.0
Rows: 1-5 | Columns: 6