Please enable JavaScript.
Coggle requires JavaScript to display documents.
Regression - Coggle Diagram
Regression
Model Evaluation
Metrics
score( R-squared )
co-efficient of determination
Value = 1 - u/v ;
u: residual sum of squares
= (Xw-y).(Xw-y)
v: total sum of squares
= (y - y_pred_mean).(y-y_pred_mean)
Best possible score = 1 ; when u = 0
constant model: = 0 ; u = v
< 0 => A bad model
0 and < 1: A good model
mean_absolute_errror
mean_squared_error
r2_score
mean_squared_log_error
mean_absolute_percentage_error
median_absolute_error
max_error
Models
LinearRegression
(sklearn.linear_model)
SGDRegressor
(sklearn.linear_model)
Constructor
Parameters
loss
penalty
L1 ( == Lasso)
L2 ( == Ridge)
elasticnet
( a convex combination of L1 and L2)
learning_rate
After every iteration, lr recalculates as:
eta = eta_0 / pow(t, power_t)
early_stopping
shuffle
eta0
Initial Learning Rate
power_t
Scaling learning rate in invscaling
tol
n_iter_no_change
validation_fraction
average
warm_start
Attributes
.coef_
intercept_
DummyRegressor
(sklearn.dummy)
Ridge
( / RidgeCV )
Constructor
Parameters
alpha
Lasso
( / LassoCV )
Constructor
Parameters
alpha
Logistic Regression
Constructor
Parameters
penalty