| | |
- GLMPerSegment(formula=None, data=None, input_columns=None, response_column=None, attribute_data=None, parameter_data=None, family='GAUSSIAN', iter_max=300, batch_size=10, lambda1=0.02, alpha=0.15, iter_num_no_change=50, tolerance=0.001, intercept=True, class_weights='0:1.0, 1:1.0', learning_rate=None, initial_eta=0.05, decay_rate=0.25, decay_steps=5, momentum=0.0, nesterov=True, iteration_mode='BATCH', partition_column=None, **generic_arguments)
- DESCRIPTION:
The GLM() function is used to train the whole data set as one model. The
GLMPerSegment() function is a partition-by-key function to create a single
model for each partition.
The following operations are supported for GLMPerSegment():
* Gaussian linear regression.
* Binomial logistic regression for binary classification.
* Iteration modes batch and epoch.
* Regularization using L1, L2 and Elasticnet.
* Mini-batch gradient descent for numeric optimization algorithm.
* Training support with and without intercept.
* Class-weighted modeling.
* Learning rate support for mini-batch gradient descent using constant,
dynamic and hybrid gradients.
* Learning rate optimization algorithms for mini-batch gradient
descent using plain, momentum, and nesterov gradients.
Notes:
* The order column can be optionally applied to guarantee the result in each
run is deterministic. The situation of indeterministic result in a partition
can occur if the "batch_size" argument is less than the number of rows in
the partition. However, adding order columns can influence the performance.
* A model is generated from the GLMPerSegment(), and a model from GLM()
should be the same when the "batch_size" argument is not less than the
number of rows in the corresponding partition of the model.
* GLMPerSegment() takes all features as numeric input. Categorical columns
must be converted to numeric columns as preprocessing step, such as using
OneHotEncodingFit()/OneHotEncodingTransform(), OrdinalEncodingFit()/
OrdinalEncodingTransform(), and TargetEncodingFit()/TargetEncodingTransform().
* Any observation with a missing value in an input column is ignored and
not used fortraining. You can use some imputation function, such as
SimpleImputeFit()/SimpleImputeTransform() to do imputation of missing values.
* Best practice is to standardize the dataset before using GLMPerSegment().
Standardization, also known as feature scaling, makes a better model and
converges quicker.
* Model evaluation metrics of MSE, Loglikelihood, AIC, and BIC are generated by
GLMPerSegment(). For additional regression and classification metrics, you
should use RegressionEvaluator(), ClassificationEvaluator() and ROC()
functions as post-processing step.
* GLMPerSegment() supports binary classification only.
* "response_column" for classification accepts values of 0 and 1 for two
classes in the response column.
* A maximum of 2046 features are supported due to the limitation imposed
by the maximum number of columns (2048) in a input data.
* "batch_size" and "learning_rate" are directly related. With a larger "batch_size",
"learning_rate" can be increased for faster training with fewer iterations.
* "iter_num" and "iter_num_no_change" are criteria used to stop learning. To force
the function to run through all iterations, disable the "iter_num_no_change"
by specifying "iter_num_no_change" to 0.
* User need to try different combinations to find the best values for a particular
use case.
* When an unsupported data type is passed in "input_columns" or "response_column",
the error message is of the following format:
Unsupported data type for column index n in argument InputColumns.
In the message, n refers to the index of the column based on an input to the
function comprising "input_columns" and "response_column only. This is due to
the rest of the columns in input are not needed by the function and internal
optimizer does not expose them to the function. Due to this, n might be different
from the actual index in the input data.
PARAMETERS:
formula:
Required Argument when "input_columns" and "response_column" are not
provided, optional otherwise.
Specifies a string consisting of "formula" which is the model to be fitted.
Only basic formula of the "col1 ~ col2 + col3 +..." form are
supported and all variables must be from the same teradataml
DataFrame object.
Notes:
* The function only accepts numeric features. User must convert the
categorical features to numeric values, before passing to the formula.
* In case categorical features are passed to formula, those are ignored,
and only numeric features are considered.
* Provide either "formula" argument or "input_columns" and
"response_column" arguments.
Types: str
data:
Required Argument.
Specifies the teradataml DataFrame containing the input data.
Types: teradataml DataFrame
attribute_data:
Optional Argument.
Specifies the teradataml DataFrame containing a subset of features
to be used with respect to each partition.
Types: teradataml DataFrame
parameter_data:
Optional Argument.
Specifies the teradataml DataFrame containing a subset of parameters
to be used with respect to each partition.
Types: teradataml DataFrame
input_columns:
Required argument when "response_column" is provided and "formula" is not
provided, optional otherwise.
Specifies the name(s) of the teradataml DataFrame column(s) that need
to be used for training the model (predictors, features, or
independent variables).
Types: str OR list of Strings (str)
response_column:
Required argument when "response_column" is provided and "formula" is not
provided, optional otherwise.
Specifies the name of the column that contains the class label for binary
classification when "family" is 'Binomial', or target value (dependent
variable) for 'Regression' when "family" is 'Gaussian'.
Types: str
family:
Optional Argument.
Specifies the distribution exponential family.
Permitted Values:
* BINOMIAL
* GAUSSIAN
Default Value: GAUSSIAN
Types: str
iter_max:
Optional Argument.
Specifies the maximum number of iterations over the training data
batches. If batch size is 0, then "iter_max" equals the number of epochs.
Note:
* The "iter_max" must be in the range [1, 10000000].
Default Value: 300
Types: int
batch_size:
Optional Argument.
Specifies the number of observations (training samples) to be parsed in
one mini-batch. Must be a non-negative integer value. A value of 0
indicates no mini-batches, so the entire input is processed in each
iteration, and the algorithm becomes (Batch) Gradient Descent. A
value higher than the number of rows on any partition also default
to Batch Gradient Descent.
Note:
* The "iter_max" must be in the range [0, 2147483647].
Default Value: 10
Types: int
lambda1:
Optional Argument.
Specifies the amount of regularization to be added. The higher the
value, the stronger the regularization. It is also used to compute
learning rate when "learning_rate" is set to 'Optimal'.
A value of '0' means no regularization.
Notes:
* The "lambda1" must be in the range [0, 1e7].
* The "lambda1" must be a non-negative float value.
Default Value: 0.02
Types: float OR int
alpha:
Optional Argument.
Specifies the Elasticnet parameter for penalty computation. It is
only effective if "lambda1" is greater than 0. The value represents
the contribution ratio of L1 in the penalty. A value of 1.0 indicates
L1 (LASSO) only, a value of 0 indicates L2 (Ridge) only, and a value
in between is a combination of L1 and L2.
Note:
* The "alpha" must be in the range [0, 1].
Default Value: 0.15
Types: float OR int
iter_num_no_change:
Optional Argument.
Specifies the number of iterations with no improvement in loss, including
the "tolerance", to stop training. A value of 0 indicates no early
stopping and the algorithm continues until "iter_max" iterations are reached.
Note:
* The "iter_num_no_change" must be in the range [0, 2147483647].
Default Value: 50
Types: int
tolerance:
Optional Argument.
Specifies the stopping criteria in terms of loss function improvement.
Training stops when loss is greater than best_loss – tolerance for
"iter_num_no_change" times.
Notes:
* The "tolerance" must be in the range [1e-7, 1e7].
* The "tolerance" works only when "iter_num_no_change"
is greater than 0.
* The "tolerance" must be a non-negative value.
Default Value: 0.001
Types: float OR int
intercept:
Optional Argument.
Specifies intercept should be estimated based on whether data is
already centered or not.
Default Value: True
Types: bool
class_weights:
Optional Argument.
Specifies the weights associated with classes. The format is
'0:weight,1:weight'. For example, '0:1.0,1:0.5' gives twice
as much weight to each observation in class 0. If the weight of
a class is omitted, then it is assumed to be 1.0.
Note:
* The "class_weights" argument works only when "family" is
'Binomial'.
Default Value: 0:1.0, 1:1.0
Types: str
learning_rate:
Optional Argument.
Specifies the Learning rate algorithm.
Permitted Values:
* CONSTANT
* OPTIMAL
* INVTIME
* ADAPTIVE
Default Value:
* 'INVTIME' when "family" is set to 'Gaussian'
* 'OPTIMAL' when "family" is set to 'Binomial'
Types: str
initial_eta:
Optional Argument.
Specifies the initial value of eta for learning rate. For constant,
this value is the learning rate for all iterations.
Note:
* The "initial_eta" must be in the range [1e-7, 1e7].
Default Value: 0.05.
Types: float OR int
decay_rate:
Optional Argument.
Specifies the decay rate for learning rate (invtime and adaptive).
Note:
* The "decay_rate" must be in the range [1e-7, 1e7].
Default Value: 0.25.
Types: float OR int
decay_steps:
Optional Argument.
Specifies the decay steps (number of iterations) for adaptive learning rate.
The learning rate changes by decay rate after this many number of iterations.
Note:
* The "decay_steps" must be in the range [1, 2147483647].
Default Value: 5
Types: int
momentum:
Optional Argument.
Specifies the value to use for momentum learning rate optimizer.
A larger value indicates higher momentum contribution. A value of 0
means momentum optimizer is disabled. For a good momentum contribution,
a value between 0.6-0.95 is recommended.
Note:
* The "momentum" must be in the range [0, 1].
Default Value: 0.0
Types: float OR int
nesterov:
Optional Argument.
Specifies the indicator that nesterov optimization should be
applied to Momentum Optimizer or not. Default is True when momentum
is greater than 0, otherwise False.
Default Value: True
Types: bool
iteration_mode:
Optional Argument.
Specifies the iteration mode.
Permitted Values:
* Batch: One iteration per batch. After processing rows in a
batch, update the weight of the parameters and proceed to the
next iteration.
* Epoch: One iteration per epoch. After processing all rows
in a partition (with one or more batches), update the weight
of the parameters and proceed to the next epoch.
Default Value: Batch
Types: str OR list of Strings (str)
partition_column:
Optional Argument.
Specifies the name of the "input_columns" on which to partition the input.
The name should be consistent with the "data_partition_column". If the
"data_partition_column" is unicode with foreign language characters, then
it is necessary to specify "partition_column" argument.
Types: str
**generic_arguments:
Specifies the generic keyword arguments SQLE functions accept. Below
are the generic keyword arguments:
persist:
Optional Argument.
Specifies whether to persist the results of the
function in a table or not. When set to True,
results are persisted in a table; otherwise,
results are garbage collected at the end of the
session.
Default Value: False
Types: bool
volatile:
Optional Argument.
Specifies whether to put the results of the
function in a volatile table or not. When set to
True, results are stored in a volatile table,
otherwise not.
Default Value: False
Types: bool
Function allows the user to partition, hash, order or local
order the input data. These generic arguments are available
for each argument that accepts teradataml DataFrame as
input and can be accessed as:
* "<input_data_arg_name>_partition_column" accepts str or
list of str (Strings)
* "<input_data_arg_name>_hash_column" accepts str or list
of str (Strings)
* "<input_data_arg_name>_order_column" accepts str or list
of str (Strings)
* "local_order_<input_data_arg_name>" accepts boolean
Note:
These generic arguments are supported by teradataml if
the underlying SQLE Engine function supports, else an
exception is raised.
RETURNS:
Instance of GLMPerSegment.
Output teradataml DataFrames can be accessed using attribute
references, such as GLMPerSegmentObj.<attribute_name>.
Output teradataml DataFrame attribute name is:
result
RAISES:
TeradataMlException, TypeError, ValueError
EXAMPLES:
# Notes:
# 1. Get the connection to Vantage to execute the function.
# 2. One must import the required functions mentioned in
# the example from teradataml.
# 3. Function will raise error if not supported on the Vantage
# user is connected to.
# Load the example data.
load_example_data("decisionforestpredict", ["housing_train"])
load_example_data("teradataml", ["housing_train_attribute", "housing_train_parameter"])
# Create teradataml DataFrame objects.
housing_train = DataFrame.from_table("housing_train")
housing_train_attribute = DataFrame.from_table("housing_train_attribute")
housing_train_parameter = DataFrame.from_table("housing_train_parameter")
# Check the list of available analytic functions.
display_analytic_functions()
# Filter the rows from train dataset with homestyle as Classic and Eclectic.
binomial_housing_train = DataFrame('housing_train', index_label="homestyle")
binomial_housing_train = binomial_housing_train.filter(like = 'ic', axis = 'rows')
# GLMPerSegment() function requires features in numeric format for processing,
# so dropping the non-numeric columns.
binomial_housing_train = binomial_housing_train.drop(columns=["driveway", "recroom",
"gashw", "airco", "prefarea",
"fullbase"])
gaussian_housing_train = binomial_housing_train.drop(columns="homestyle")
# Transform the train dataset categorical values to encoded values.
housing_train_ordinal_encodingfit = OrdinalEncodingFit(
target_column='homestyle',
data=binomial_housing_train)
housing_train_ordinal_encodingtransform = OrdinalEncodingTransform(
data=binomial_housing_train,
object=housing_train_ordinal_encodingfit.result,
accumulate=["sn", "price", "lotsize",
"bedrooms", "bathrms", "stories"])
# Example 1: Train the model using the 'Gaussian' family.
GLMPerSegment_out_1 = GLMPerSegment(data=gaussian_housing_train,
data_partition_column="stories",
input_columns=['garagepl', 'lotsize', 'bedrooms', 'bathrms'],
response_column="price",
family="Gaussian",
iter_max=1000,
batch_size=9)
# Print the result DataFrame.
print(GLMPerSegment_out_1.result)
# Example 2: Train the model using the 'Binomial' family, formula argument and
# subset of features and parameters to be used with respect to
# "partition_id".
formula = "homestyle ~ price + lotsize + bedrooms + bathrms"
GLMPerSegment_out_2 = GLMPerSegment(data=housing_train_ordinal_encodingtransform.result,
data_partition_column="stories",
formula = formula,
attribute_data=housing_train_attribute,
attribute_data_partition_column="partition_id",
parameter_data=housing_train_parameter,
parameter_data_partition_column="partition_id",
family="Binomial",
iter_max=100)
# Print the result DataFrame.
print(GLMPerSegment_out_2.result)
|