Partager via

OnlineGradientDescentRegressor Class

  • Reference

Train a stochastic gradient descent model.

Inheritance
nimbusml.internal.core.linear_model._onlinegradientdescentregressor.OnlineGradientDescentRegressor
OnlineGradientDescentRegressor
nimbusml.base_predictor.BasePredictor
OnlineGradientDescentRegressor
sklearn.base.RegressorMixin
OnlineGradientDescentRegressor

Constructor

OnlineGradientDescentRegressor(normalize='Auto', caching='Auto', loss='squared', learning_rate=0.1, decrease_learning_rate=True, l2_regularization=0.0, number_of_iterations=1, initial_weights_diameter=0.0, reset_weights_after_x_examples=None, lazy_update=True, recency_gain=0.0, recency_gain_multiplicative=False, averaged=True, averaged_tolerance=0.01, initial_weights=None, shuffle=True, feature=None, label=None, **params)

Parameters

Name Description
feature

see Columns.
label

see Columns.
normalize

Specifies the type of automatic normalization used:

  • "Auto": if normalization is needed, it is performed automatically. This is the default choice.

  • "No": no normalization is performed.

  • "Yes": normalization is performed.

  • "Warn": if normalization is needed, a warning message is displayed, but normalization is not performed.

Normalization rescales disparate data ranges to a standard scale. Feature scaling insures the distances between data points are proportional and enables various optimization methods such as gradient descent to converge much faster. If normalization is performed, a MaxMin normalizer is used. It normalizes values in an interval [a, b] where -1 <= a <= 0 and 0 <= b <= 1 and b - a = 1. This normalizer preserves sparsity by mapping zero to zero.
caching

Whether trainer should cache input training data.
loss

The default is Hinge. Other choices are Exp, Log, SmoothedHinge. For more information, please see nimbusml.
learning_rate

Determines the size of the step taken in the direction of the gradient in each step of the learning process. This determines how fast or slow the learner converges on the optimal solution. If the step size is too big, you might overshoot the optimal solution. If the step size is too small, training takes longer to converge to the best solution.
decrease_learning_rate

Decrease learning rate.
l2_regularization

L2 Regularization Weight.
number_of_iterations

Number of iterations.
initial_weights_diameter

Sets the initial weights diameter that specifies the range from which values are drawn for the initial weights. These weights are initialized randomly from within this range. For example, if the diameter is specified to be d, then the weights are uniformly distributed between -d/2 and d/2. The default value is 0, which specifies that all the weights are set to zero.
reset_weights_after_x_examples

Number of examples after which weights will be reset to the current average.
lazy_update

Instead of updating averaged weights on every example, only update when loss is nonzero.
recency_gain

Extra weight given to more recent updates (do_lazy_updates` must be False).
recency_gain_multiplicative

Whether Recency Gain is multiplicative (vs. additive).
averaged

Do averaging?.
averaged_tolerance

The inexactness tolerance for averaging.
initial_weights

Initial Weights and bias, comma-separated.
shuffle

Whether to shuffle for each training iteration.
params

Additional arguments sent to compute engine.

Examples


   ###############################################################################
   # OnlineGradientDescentRegressor
   from nimbusml import Pipeline, FileDataStream
   from nimbusml.datasets import get_dataset
   from nimbusml.feature_extraction.categorical import OneHotVectorizer
   from nimbusml.linear_model import OnlineGradientDescentRegressor

   # data input (as a FileDataStream)
   path = get_dataset('infert').as_filepath()

   data = FileDataStream.read_csv(path)
   print(data.head())
   #    age  case education  induced  parity ... row_num  spontaneous  ...
   # 0   26     1    0-5yrs        1       6 ...       1            2  ...
   # 1   42     1    0-5yrs        1       1 ...       2            0  ...
   # 2   39     1    0-5yrs        2       6 ...       3            0  ...
   # 3   34     1    0-5yrs        2       4 ...       4            0  ...
   # 4   35     1   6-11yrs        1       3 ...       5            1  ...

   # define the training pipeline
   pipeline = Pipeline([
       OneHotVectorizer(columns={'edu': 'education'}),
       OnlineGradientDescentRegressor(feature=['parity', 'edu'], label='age')
   ])

   # train, predict, and evaluate
   metrics, predictions = pipeline.fit(data).test(data, output_scores=True)

   # print predictions
   print(predictions.head())
   #       Score
   # 0  28.103731
   # 1  21.805904
   # 2  28.103731
   # 3  25.584600
   # 4  33.743286
   # print evaluation metrics
   print(metrics)
   #    L1(avg)   L2(avg)  RMS(avg)  Loss-fn(avg)  R Squared
   # 0  4.452286  31.15933  5.582054      31.15933  -0.134398

Remarks

Stochastic gradient descent uses a simple yet efficient iterative technique to fit model coefficients using error gradients for convex loss functions (see Stochastic_gradient_descent).

The OnlineGradientDescentRegressor implements the standard (non- batch) SGD, with a choice of loss functions, and an option to update the weight vector using the average of the vectors seen over time (averaged argument is set to True by default).

Reference

Stochastic_gradient_descent

Methods

get_params

Get the parameters for this operator.

get_params

Get the parameters for this operator.

get_params(deep=False)

Parameters

Name Description
deep
Default value: False