Change Log

v.1.4.10 Changes

• Added a function to construct features derived from PFS mutual information estimation that should be expected to be linearly related to the target.
• Fixed a global name conflict in kxy.learning.base_learners.

v.1.4.9 Changes

• Change the activation function used by PFS from ReLU to switch/SILU.
• Leaving it to the user to set the logging level.

v.1.4.8 Changes

• Froze the versions of all python packages in the docker file.

v.1.4.7 Changes

Changes related to optimizing Principal Feature Selection.

• Made it easy to change PFS' default learning parameters.
• Changed PFS' default learning parameters (learning rate is now 0.005 and epsilon 1e-04)
• Adding a seed parameter to PFS' fit for reproducibility.

To globally change the learning rate to 0.003, change Adam's epsilon to 1e-5, and the number of epochs to 25, do

from kxy.misc.tf import set_default_parameter
set_default_parameter('lr', 0.003)
set_default_parameter('epsilon', 1e-5)
set_default_parameter('epochs', 25)

To change the number epochs for a single iteration of PFS, use the epochs argument of the fit method of your PFS object. The fit method now also has a seed parameter you may use to make the PFS implementation deterministic.

Example:

from kxy.pfs import PFS
selector = PFS()
selector.fit(x, y, epochs=25, seed=123)

Alternatively, you may also use the kxy.misc.tf.set_seed method to make PFS deterministic.

v.1.4.6 Changes

Minor PFS improvements.

• Adding more (robust) mutual information loss functions.
• Exposing the learned total mutual information between principal features and target as an attribute of PFS.
• Exposing the number of epochs as a parameter of PFS' fit.

Change Log

v.1.4.9 Changes

• Change the activation function used by PFS from ReLU to switch/SILU.
• Leaving it to the user to set the logging level.

v.1.4.8 Changes

• Froze the versions of all python packages in the docker file.

v.1.4.7 Changes

Changes related to optimizing Principal Feature Selection.

• Made it easy to change PFS' default learning parameters.
• Changed PFS' default learning parameters (learning rate is now 0.005 and epsilon 1e-04)
• Adding a seed parameter to PFS' fit for reproducibility.

To globally change the learning rate to 0.003, change Adam's epsilon to 1e-5, and the number of epochs to 25, do

from kxy.misc.tf import set_default_parameter
set_default_parameter('lr', 0.003)
set_default_parameter('epsilon', 1e-5)
set_default_parameter('epochs', 25)

To change the number epochs for a single iteration of PFS, use the epochs argument of the fit method of your PFS object. The fit method now also has a seed parameter you may use to make the PFS implementation deterministic.

Example:

from kxy.pfs import PFS
selector = PFS()
selector.fit(x, y, epochs=25, seed=123)

Alternatively, you may also use the kxy.misc.tf.set_seed method to make PFS deterministic.

v.1.4.6 Changes

Minor PFS improvements.

• Adding more (robust) mutual information loss functions.
• Exposing the learned total mutual information between principal features and target as an attribute of PFS.
• Exposing the number of epochs as a parameter of PFS' fit.

Change Log

v.1.4.8 Changes

• Froze the versions of all python packages in the docker file.

v.1.4.7 Changes

Changes related to optimizing Principal Feature Selection.

• Made it easy to change PFS' default learning parameters.
• Changed PFS' default learning parameters (learning rate is now 0.005 and epsilon 1e-04)
• Adding a seed parameter to PFS' fit for reproducibility.

To globally change the learning rate to 0.003, change Adam's epsilon to 1e-5, and the number of epochs to 25, do

from kxy.misc.tf import set_default_parameter
set_default_parameter('lr', 0.003)
set_default_parameter('epsilon', 1e-5)
set_default_parameter('epochs', 25)

To change the number epochs for a single iteration of PFS, use the epochs argument of the fit method of your PFS object. The fit method now also has a seed parameter you may use to make the PFS implementation deterministic.

Example:

from kxy.pfs import PFS
selector = PFS()
selector.fit(x, y, epochs=25, seed=123)

Alternatively, you may also use the kxy.misc.tf.set_seed method to make PFS deterministic.

v.1.4.6 Changes

Minor PFS improvements.

• Adding more (robust) mutual information loss functions.
• Exposing the learned total mutual information between principal features and target as an attribute of PFS.
• Exposing the number of epochs as a parameter of PFS' fit.

Change Log

v.1.4.7 Changes

Changes related to optimizing Principal Feature Selection.

• Made it easy to change PFS' default learning parameters.
• Changed PFS' default learning parameters (learning rate is now 0.005 and epsilon 1e-04)
• Adding a seed parameter to PFS' fit for reproducibility.

To globally change the learning rate to 0.003, change Adam's epsilon to 1e-5, and the number of epochs to 25, do

from kxy.misc.tf import set_default_parameter
set_default_parameter('lr', 0.003)
set_default_parameter('epsilon', 1e-5)
set_default_parameter('epochs', 25)

To change the number epochs for a single iteration of PFS, use the epochs argument of the fit method of your PFS object. The fit method now also has a seed parameter you may use to make the PFS implementation deterministic.

Example:

from kxy.pfs import PFS
selector = PFS()
selector.fit(x, y, epochs=25, seed=123)

Alternatively, you may also use the kxy.misc.tf.set_seed method to make PFS deterministic.

v.1.4.6 Changes

Minor PFS improvements.

• Adding more (robust) mutual information loss functions.
• Exposing the learned total mutual information between principal features and target as an attribute of PFS.
• Exposing the number of epochs as a parameter of PFS' fit.

Changes

• Adding more (robust) mutual information loss functions.
• Exposing the learned total mutual information between principal features and target as an attribute of PFS.
• Exposing the number of epochs as a parameter of PFS' fit.

Fixing some package incompatibilities.

Adding Principal Feature Selection.

Adding regression root mean square error (RMSE) in the list of metrics whose achievable values we calculate.

Adding a maximum-entropy based classifier (kxy.MaxEntClassifier) and regressor (kxy.MaxEntRegressor) following the scikit-learn signature for fitting and predicting.

These models estimate the posterior mean E[u_y|x] and the posterior standard deviation sqrt(Var[u_y|x]) for any specific value of x, where the copula-uniform representations (u_y, u_x) follow the maximum-entropy distribution.

Predictions in the primal are derived from E[u_y|x].

• Regression analyses now fully support categorical variables.
• Foundations for multi-output regressions are laid.
• Categorical variables are now systematically encoded and treated as continuous, consistent with what's done at the learning stage.
• Regression and classification are further normalized, and most the compute for classification problems now takes place on the API side, and should be considerably faster.