alpha |
Number
|
<optional>
|
Multiplier of the regularization
term in the objective of SGD. |
l1_ratio |
Number
|
<optional>
|
Fraction of L1 regularization
in sum of L1 and L2 regularizations. |
max_iter |
Number
|
<optional>
|
Maximum number of iterations
for which to train the model. |
eta0 |
Number
|
<optional>
|
Learning rate for the algorithm. |
penalty |
String
|
<optional>
|
Type of penalty to use. Supported
- 'elasticnet': both L1 and L2 penalty are used; Ratio between the
two losses is specified using `l1_ratio` parameter. |
loss |
String
|
<optional>
|
Loss to use during training. Supported
losses are:
- 'squared_loss': squared error, l(y, f) = 1/n \sum_{i = 1..n}(y_i - f_i)^2
This loss is often used for regression problems.
- 'hinge': Hinge loss, l(y, f) = 1/n \sum_{i = 1..n} max(0, 1 - y_i * f_i)
This loss assumes that y_i \in {-1, 1}. Most of the time this loss is used
for classification problems. |