"""
Class for training and predicting with built-in or SKLL models.
:author: Jeremy Biggs (jbiggs@ets.org)
:author: Anastassia Loukina (aloukina@ets.org)
:author: Nitin Madnani (nmadnani@ets.org)
:organization: ETS
"""
import logging
import pickle
from math import log10, sqrt
from os.path import join
from typing import Any, Dict, List, Optional, Tuple
import joblib
import numpy as np
import pandas as pd
import statsmodels.api as sm
from numpy.random import RandomState
from scipy.optimize import nnls
from sklearn.linear_model import LassoCV
from skll.data import FeatureSet
from skll.learner import Learner
from statsmodels.regression.linear_model import RegressionResults
from .analyzer import Analyzer
from .configuration_parser import Configuration
from .container import DataContainer, DatasetDict
from .preprocessor import FeaturePreprocessor
from .utils.metrics import compute_expected_scores_from_model
from .utils.models import is_skll_model
from .writer import DataWriter
[docs]
class Modeler:
"""
Class to train model and generate predictions with built-in or SKLL models.
Note
----
The learner and scaling-/trimming-related attributes are set to ``None``.
"""
def __init__(self, logger: Optional[logging.Logger] = None) -> None:
"""
Initialize empty Modeler object.
Parameters
----------
logger : Optional[logging.Logger]
Logger object to use for logging messages. If ``None``, a new logger
instance will be created.
Defaults to ``None``.
"""
self.feature_info: Optional[pd.DataFrame] = None
self.learner: Optional[Learner] = None
self.trim_min: Optional[float] = None
self.trim_max: Optional[float] = None
self.trim_tolerance: Optional[float] = None
self.train_predictions_mean: Optional[float] = None
self.train_predictions_sd: Optional[float] = None
self.h1_mean: Optional[float] = None
self.h1_sd: Optional[float] = None
if logger is None:
self.logger = logging.getLogger(__name__)
else:
self.logger = logger
[docs]
def save(self, model_path: str) -> None:
"""
Save an instance of this class to disk.
Parameters
----------
model_path : str
Destination path for model file
"""
joblib.dump(self, model_path)
[docs]
@classmethod
def load_from_file(cls, path: str) -> "Modeler":
"""
Load a ``Modeler`` object from a file on disk.
The file must contain either a ``Modeler`` or a SKLL ``Learner``,
in which case a ``Modeler`` object will be created from the
``Learner``.
Parameters
----------
path : str
File path from which to load the modeler object.
Returns
-------
model : Modeler
A ``Modeler`` instance.
Raises
------
ValueError
If ``path`` does not end with ".model".
"""
if not path.lower().endswith(".model"):
raise ValueError(
f"The file `{path}` does not end with the proper extension. Please "
"make sure that it is a `.model` file."
)
with open(path, "rb") as model_file:
modeler_or_learner = joblib.load(model_file)
if isinstance(modeler_or_learner, Modeler):
return modeler_or_learner
# If not a Modeler object, try to load as if it were a SKLL
# Learner object (for backward compatibility)
learner = Learner.from_file(path)
return cls.load_from_learner(learner)
[docs]
@classmethod
def load_from_learner(cls, learner: Learner) -> "Modeler":
"""
Create a new ``Modeler`` object with a pre-populated learner.
Parameters
----------
learner : skll.learner.Learner
A SKLL Learner object.
Returns
-------
modeler : Modeler
The newly created ``Modeler`` object.
Raises
------
TypeError
If ``learner`` is not a SKLL Learner instance.
"""
if not isinstance(learner, Learner):
raise TypeError(
"The `learner` argument must be a `SKLL Learner` instance, "
f"not `{type(learner)}`."
)
# Create Modeler instance
modeler = Modeler()
modeler.learner = learner
return modeler
[docs]
@staticmethod
def model_fit_to_dataframe(fit: RegressionResults) -> pd.DataFrame:
"""
Extract fit metrics from a ``statsmodels`` fit object into a data frame.
Parameters
----------
fit : statsmodels.regression.linear_model.RegressionResults
Model fit object obtained from a linear model trained using
``statsmodels.OLS``.
Returns
-------
df_fit : pandas.DataFrame
The output data frame with the main model fit metrics.
"""
df_fit = pd.DataFrame({"N responses": [int(fit.nobs)]})
df_fit["N features"] = int(fit.df_model)
df_fit["R2"] = fit.rsquared
df_fit["R2_adjusted"] = fit.rsquared_adj
return df_fit
[docs]
@staticmethod
def ols_coefficients_to_dataframe(coefs: pd.Series) -> pd.DataFrame:
"""
Convert series containing OLS coefficients to a data frame.
Parameters
----------
coefs : pandas.Series
Series with feature names in the index and the coefficient
values as the data, obtained from a linear model trained
using ``statsmodels.OLS``.
Returns
-------
df_coef : pandas.DataFrame
Data frame with two columns: the feature name and
the coefficient value.
Note
----
The first row in the output data frame is always for the intercept
and the rest are sorted by feature name.
"""
# first create a sorted data frame for all the non-intercept features
non_intercept_columns = [c for c in coefs.index if c != "const"]
df_non_intercept = pd.DataFrame(
coefs.filter(non_intercept_columns), columns=["coefficient"]
)
df_non_intercept.index.name = "feature"
df_non_intercept = df_non_intercept.sort_index()
df_non_intercept.reset_index(inplace=True)
# now create a data frame that just has the intercept
df_intercept = pd.DataFrame([{"feature": "Intercept", "coefficient": coefs["const"]}])
# append the non-intercept frame to the intercept one
df_coef = pd.concat([df_intercept, df_non_intercept], sort=True, ignore_index=True)
# we always want to have the feature column first
df_coef = df_coef[["feature", "coefficient"]]
return df_coef
[docs]
@staticmethod
def skll_learner_params_to_dataframe(learner: Learner) -> pd.DataFrame:
"""
Extract parameters from the given SKLL learner into a data frame.
Parameters
----------
learner : skll.learner.Learner
A SKLL learner object.
Returns
-------
df_coef : pandas.DataFrame
The data frame containing the model parameters from the given SKLL
Learner object.
Note
----
1. We use the ``coef_`` attribute of the scikit-learn model underlying
the SKLL learner instead of the latter's ``model_params`` attribute.
This is because ``model_params`` ignores zero coefficients, which we
do not want.
2. The first row in the output data frame is always for the intercept
and the rest are sorted by feature name.
"""
# get the intercept, coefficients, and feature names
intercept = learner.model.intercept_
coefficients = learner.model.coef_
feature_names = learner.feat_vectorizer.get_feature_names_out()
# first create a sorted data frame for all the non-intercept features
df_non_intercept = pd.DataFrame({"feature": feature_names, "coefficient": coefficients})
df_non_intercept = df_non_intercept.sort_values(by=["feature"])
# now create a data frame that just has the intercept
df_intercept = pd.DataFrame([{"feature": "Intercept", "coefficient": intercept}])
# append the non-intercept frame to the intercept one
df_coef = pd.concat([df_intercept, df_non_intercept], sort=True, ignore_index=True)
# we always want to have the feature column first
df_coef = df_coef[["feature", "coefficient"]]
return df_coef
[docs]
def create_fake_skll_learner(self, df_coefficients: pd.DataFrame) -> Learner:
"""
Create a fake SKLL linear regression learner from given coefficients.
Parameters
----------
df_coefficients : pandas.DataFrame
The data frame containing the linear coefficients we want to create
the fake SKLL model with.
Returns
-------
learner: skll.learner.Learner
SKLL Learner object representing a ``LinearRegression`` model
with the specified coefficients.
"""
# initialize a random number generator
randgen = RandomState(1234567890)
# iterate over the coefficients
coefdict = {}
for feature, coefficient in df_coefficients.itertuples(index=False):
if feature == "Intercept":
intercept = coefficient
else:
# exclude NA coefficients
if coefficient == np.nan:
self.logger.warning(
f"No coefficient was estimated for {feature}. "
f"This is likely due to exact collinearity in "
f"the model. This feature will not be used for "
f"model building"
)
else:
coefdict[feature] = coefficient
learner = Learner("LinearRegression")
num_features = len(coefdict) # excluding the intercept
fake_feature_values = randgen.rand(num_features)
fake_features = [dict(zip(coefdict, fake_feature_values))]
fake_fs = FeatureSet("fake", ids=["1"], labels=[1.0], features=fake_features)
learner.train(fake_fs, grid_search=False)
# now create its parameters from the coefficients from the built-in model
learner.model.coef_ = learner.feat_vectorizer.transform(coefdict).toarray()[0]
learner.model.intercept_ = intercept
return learner
[docs]
def train_linear_regression(
self, df_train: pd.DataFrame, feature_columns: List[str]
) -> Tuple[Learner, RegressionResults, pd.DataFrame, List[str]]:
"""
Train a "LinearRegression" model.
This model is a simple linear regression model.
Parameters
----------
df_train : pandas.DataFrame
Data frame containing the features on which to train the model.
feature_columns : List[str]
A list of feature columns to use in training the model.
Returns
-------
learner : skll.learner.Learner
The SKLL learner object.
fit : statmodels.regression.linear_model.RegressionResults
A ``statsmodels`` regression results object.
df_coef : pandas.DataFrame
Data frame containing the model coefficients.
used_features : List[str]
A list of features used in the final model.
"""
# get the feature columns
X = df_train[feature_columns]
# add the intercept
X = sm.add_constant(X)
# fit the model
fit = sm.OLS(df_train["sc1"], X).fit()
df_coef = self.ols_coefficients_to_dataframe(fit.params)
learner = self.create_fake_skll_learner(df_coef)
# we used all the features
used_features = feature_columns
return learner, fit, df_coef, used_features
[docs]
def train_equal_weights_lr(
self, df_train: pd.DataFrame, feature_columns: List[str]
) -> Tuple[Learner, RegressionResults, pd.DataFrame, List[str]]:
"""
Train an "EqualWeightsLR" model.
This model assigns the same weight to all features.
Parameters
----------
df_train : pandas.DataFrame
Data frame containing the features on which to train the model.
feature_columns : List[str]
A list of feature columns to use in training the model.
Returns
-------
learner : skll.learner.Learner
The SKLL learner object.
fit : statsmodels.regression.linear_model.RegressionResults
A ``statsmodels`` regression results object.
df_coef : pandas.DataFrame
Data frame containing the model coefficients.
used_features : List[str]
A list of features used in the final model.
"""
# we first compute a single feature that is simply the sum of all features
df_train_eqwt = df_train.copy()
df_train_eqwt["sumfeature"] = df_train_eqwt[feature_columns].apply(np.sum, axis=1)
# train a plain Linear Regression model
X = df_train_eqwt["sumfeature"]
X = sm.add_constant(X)
fit = sm.OLS(df_train_eqwt["sc1"], X).fit()
# get the coefficient for the summed feature and the intercept
coef = fit.params["sumfeature"]
const = fit.params["const"]
# now we need to assign this coefficient to all of the original
# features and create a fake SKLL learner with these weights
original_features = [
c for c in df_train_eqwt.columns if c not in ["sc1", "sumfeature", "spkitemid"]
]
coefs = pd.Series(dict([(origf, coef) for origf in original_features] + [("const", const)]))
df_coef = self.ols_coefficients_to_dataframe(coefs)
# create fake SKLL learner with these coefficients
learner = self.create_fake_skll_learner(df_coef)
# we used all the features
used_features = feature_columns
return learner, fit, df_coef, used_features
[docs]
def train_rebalanced_lr(
self, df_train: pd.DataFrame, feature_columns: List[str]
) -> Tuple[Learner, RegressionResults, pd.DataFrame, List[str]]:
"""
Train a "RebalancedLR" model.
This model balances empirical weights by changing betas (adapted
from `here <https://stats.stackexchange.com/questions/30876/
how-to-convert-standardized-coefficients-to-unstandardized-coefficients>`_).
Parameters
----------
df_train : pandas.DataFrame
Data frame containing the features on which to train the model.
feature_columns : List[str]
A list of feature columns to use in training the model.
Returns
-------
learner : skll.learner.Learner
The SKLL learner object.
fit : statsmodels.regression.linear_model.RegressionResults
A ``statsmodels`` regression results object.
df_coef : pandas.DataFrame
Data frame containing the model coefficients.
used_features : List[str]
A list of features used in the final model.
"""
# train a plain Linear Regression model
X = df_train[feature_columns]
X = sm.add_constant(X)
fit = sm.OLS(df_train["sc1"], X).fit()
# convert the model parameters into a data frame
df_params = self.ols_coefficients_to_dataframe(fit.params)
df_params = df_params.set_index("feature")
# compute the betas for the non-intercept coefficients
df_weights = df_params.loc[feature_columns]
df_betas = df_weights.copy()
df_train_std = df_train[feature_columns].std()
df_betas["coefficient"] = (
df_weights["coefficient"].multiply(df_train_std, axis="index") / df_train["sc1"].std()
)
# replace each negative beta with delta and adjust
# all the positive betas to account for this
RT = 0.05
df_positive_betas = df_betas[df_betas["coefficient"] > 0]
df_negative_betas = df_betas[df_betas["coefficient"] < 0]
delta = np.sum(df_positive_betas["coefficient"]) * RT / len(df_negative_betas)
df_betas["coefficient"] = df_betas.apply(
lambda row: row["coefficient"] * (1 - RT) if row["coefficient"] > 0 else delta,
axis=1,
)
# rescale the adjusted betas to get the new coefficients
df_coef = df_betas["coefficient"] * df_train["sc1"].std()
df_coef = df_coef.divide(df_train[feature_columns].std(), axis="index")
# add the intercept back to the new coefficients
df_coef["Intercept"] = df_params.loc["Intercept"].coefficient
df_coef = df_coef.sort_index().reset_index()
df_coef.columns = ["feature", "coefficient"]
# create fake SKLL learner with these coefficients
learner = self.create_fake_skll_learner(df_coef)
# we used all the features
used_features = feature_columns
return learner, fit, df_coef, used_features
[docs]
def train_lasso_fixed_lambda_then_lr(
self, df_train: pd.DataFrame, feature_columns: List[str]
) -> Tuple[Learner, RegressionResults, pd.DataFrame, List[str]]:
"""
Train a "LassoFixedLambdaThenLR" model.
First do feature selection using lasso regression with
a fixed lambda and then use only those features to train
a second linear regression
Parameters
----------
df_train : pandas.DataFrame
Data frame containing the features on which to train the model.
feature_columns : List[str]
A list of feature columns to use in training the model.
Returns
-------
learner : skll.learner.Learner
The SKLL learner object
fit : statsmodels.regression.linear_model.RegressionResults
A ``statsmodels`` regression results object.
df_coef : pandas.DataFrame
The model coefficients in a data_frame
used_features : List[str]
A list of features used in the final model.
"""
# train a Lasso Regression model with this featureset with a preset lambda
p_lambda = sqrt(len(df_train) * log10(len(feature_columns)))
# create a SKLL FeatureSet instance from the given data frame
fs_train = FeatureSet.from_data_frame(
df_train[feature_columns + ["sc1"]], "train", labels_column="sc1"
)
# note that 'alpha' in sklearn is different from this lambda
# so we need to normalize looking at the sklearn objective equation
p_alpha = p_lambda / len(df_train)
l_lasso = Learner("Lasso", model_kwargs={"alpha": p_alpha, "positive": True})
l_lasso.train(fs_train, grid_search=False)
# get the feature names that have the non-zero coefficients
non_zero_features = list(l_lasso.model_params[0].keys())
# now train a new vanilla linear regression with just the non-zero features
X = df_train[non_zero_features]
X = sm.add_constant(X)
fit = sm.OLS(df_train["sc1"], X).fit()
# get the coefficients data frame
df_coef = self.ols_coefficients_to_dataframe(fit.params)
# create fake SKLL learner with these coefficients
learner = self.create_fake_skll_learner(df_coef)
# we used only the non-zero features
used_features = non_zero_features
return learner, fit, df_coef, used_features
[docs]
def train_positive_lasso_cv_then_lr(
self, df_train: pd.DataFrame, feature_columns: List[str]
) -> Tuple[Learner, RegressionResults, pd.DataFrame, List[str]]:
"""
Train a "PositiveLassoCVThenLR" model.
First do feature selection using lasso regression optimized
for log likelihood using cross validation and then use only
those features to train a second linear regression.
Parameters
----------
df_train : pandas.DataFrame
Data frame containing the features on which to train the model.
feature_columns : List[str]
A list of feature columns to use in training the model.
Returns
-------
learner : skll.learner.Learner
The SKLL learner object.
fit : statsmodels.regression.linear_model.RegressionResults
A statsmodels regression results object.
df_coef : pandas.DataFrame
Data frame containing the model coefficients.
used_features : List[str]
A list of features used in the final model.
"""
# train a LassoCV outside of SKLL since it's not exposed there
X = df_train[feature_columns].values
y = df_train["sc1"].values
clf = LassoCV(cv=10, positive=True, random_state=1234567890)
model = clf.fit(X, y)
# get the non-zero features from this model
non_zero_features = []
for feature, coefficient in zip(feature_columns, model.coef_):
if coefficient != 0:
non_zero_features.append(feature)
# now train a new linear regression with just these non-zero features
X = df_train[non_zero_features]
X = sm.add_constant(X)
fit = sm.OLS(df_train["sc1"], X).fit()
# convert the model parameters into a data frame
df_coef = self.ols_coefficients_to_dataframe(fit.params)
# create fake SKLL learner with these coefficients
learner = self.create_fake_skll_learner(df_coef)
# we used only the non-zero features
used_features = non_zero_features
return learner, fit, df_coef, used_features
[docs]
def train_non_negative_lr(
self, df_train: pd.DataFrame, feature_columns: List[str]
) -> Tuple[Learner, RegressionResults, pd.DataFrame, List[str]]:
"""
Train an "NNLR" model.
To do this, we first do feature selection using non-negative
least squares (NNLS) and then use only its non-zero features to
train another linear regression (LR) model. We do the regular LR
at the end since we want an LR object so that we have access to R^2
and other useful statistics. There should be no difference between
the non-zero coefficients from NNLS and the coefficients that end
up coming out of the subsequent LR.
Parameters
----------
df_train : pandas.DataFrame
Data frame containing the features on which to train the model.
feature_columns : List[str]
A list of feature columns to use in training the model.
Returns
-------
learner : skll.learner.Learner
The SKLL learner object.
fit : statsmodels.regression.linear_model.RegressionResults
A statsmodels regression results object.
df_coef : pandas.DataFrame
Data frame containing the model coefficients.
used_features : List[str]
A list of features used in the final model.
"""
# add an intercept to the features manually
X = df_train[feature_columns].values
intercepts = np.ones((len(df_train), 1))
X_plus_intercept = np.concatenate([intercepts, X], axis=1)
y = df_train["sc1"].values
# fit an NNLS model on this data
coefs, _ = nnls(X_plus_intercept, y)
# check whether the intercept is set to 0 and if so then we need
# to flip the sign and refit the model to ensure that it is always
# kept in the model
if coefs[0] == 0:
intercepts = -1 * np.ones((len(df_train), 1))
X_plus_intercept = np.concatenate([intercepts, X], axis=1)
coefs, _ = nnls(X_plus_intercept, y)
# separate the intercept and feature coefficients
# intercept = coefs[0]
coefficients = coefs[1:].tolist()
# get the non-zero features from this model
non_zero_features = []
for feature, coefficient in zip(feature_columns, coefficients):
if coefficient != 0:
non_zero_features.append(feature)
# now train a new linear regression with just these non-zero features
X = df_train[non_zero_features]
X = sm.add_constant(X)
fit = sm.OLS(df_train["sc1"], X).fit()
# convert this model's parameters to a data frame
df_coef = self.ols_coefficients_to_dataframe(fit.params)
# create fake SKLL learner with these coefficients
learner = self.create_fake_skll_learner(df_coef)
# we used only the non-zero features
used_features = non_zero_features
return learner, fit, df_coef, used_features
[docs]
def train_non_negative_lr_iterative(
self, df_train: pd.DataFrame, feature_columns: List[str]
) -> Tuple[Learner, RegressionResults, pd.DataFrame, List[str]]:
"""
Train an "NNLR_iterative" model.
For applications where there is a concern that standard NNLS may not
converge, an alternate method of training NNLR by iteratively fitting
OLS models, checking the coefficients, and dropping negative
coefficients. First, fit an OLS model. Then, identify any variables
whose coefficients are negative. Drop these variables from the model.
Finally, refit the model. If any coefficients are still negative, set
these to zero.
Parameters
----------
df_train : pandas.DataFrame
Data frame containing the features on which to train the model.
feature_columns : List[str]
A list of feature columns to use in training the model.
Returns
-------
learner : skll.learner.Learner
The SKLL learner object.
fit : statsmodels.regression.linear_model.RegressionResults
A statsmodels regression results object.
df_coef : pandas.DataFrame
Data frame containing the model coefficients.
used_features : List[str]
A list of features used in the final model.
"""
X = df_train[feature_columns]
X = sm.add_constant(X)
y = df_train["sc1"]
fit = sm.OLS(y, X).fit()
positive_features = []
for name, value in fit.params.items():
if value >= 0 and name != "const":
positive_features.append(name)
X = df_train[positive_features]
X = sm.add_constant(X)
fit = sm.OLS(y, X).fit()
# if any parameters are still negative, set them to zero
params = fit.params.copy()
params = params.drop("const")
if not (params >= 0).all():
fit.params[(fit.params < 0) & (fit.params.index != "const")] = 0
# convert this model's parameters to a data frame
df_coef = self.ols_coefficients_to_dataframe(fit.params)
# create fake SKLL learner with these coefficients
learner = self.create_fake_skll_learner(df_coef)
# we used only the non-zero features
used_features = positive_features
return learner, fit, df_coef, used_features
[docs]
def train_lasso_fixed_lambda_then_non_negative_lr(
self, df_train: pd.DataFrame, feature_columns: List[str]
) -> Tuple[Learner, RegressionResults, pd.DataFrame, List[str]]:
"""
Train an "LassoFixedLambdaThenNNLR" model.
First do feature selection using lasso regression and positive
only weights. Then fit an NNLR (see above) on those features.
Parameters
----------
df_train : pandas.DataFrame
Data frame containing the features on which to train the model.
feature_columns : List[str]
A list of feature columns to use in training the model.
Returns
-------
learner : skll.learner.Learner
The SKLL learner object.
fit : statsmodels.regression.linear_model.RegressionResults
A statsmodels regression results object.
df_coef : pandas.DataFrame
Data frame containing the model coefficients.
used_features : List[str]
A list of features used in the final model.
"""
# train a Lasso Regression model with a preset lambda
p_lambda = sqrt(len(df_train) * log10(len(feature_columns)))
# create a SKLL FeatureSet instance from the given data frame
fs_train = FeatureSet.from_data_frame(
df_train[feature_columns + ["sc1"]], "train", labels_column="sc1"
)
# note that 'alpha' in sklearn is different from this lambda
# so we need to normalize looking at the sklearn objective equation
p_alpha = p_lambda / len(df_train)
l_lasso = Learner("Lasso", model_kwargs={"alpha": p_alpha, "positive": True})
l_lasso.train(fs_train, grid_search=False)
# get the feature names that have the non-zero coefficients
non_zero_features = list(l_lasso.model_params[0].keys())
# now train an NNLS regression using these non-zero features
# first add an intercept to the features manually
X = df_train[feature_columns].values
intercepts = np.ones((len(df_train), 1))
X_plus_intercept = np.concatenate([intercepts, X], axis=1)
y = df_train["sc1"].values
# fit an NNLS model on this data
coefs, _ = nnls(X_plus_intercept, y)
# check whether the intercept is set to 0 and if so then we need
# to flip the sign and refit the model to ensure that it is always
# kept in the model
if coefs[0] == 0:
intercepts = -1 * np.ones((len(df_train), 1))
X_plus_intercept = np.concatenate([intercepts, X], axis=1)
coefs, _ = nnls(X_plus_intercept, y)
# separate the intercept and feature coefficients
# even though we do not use intercept in the code
# we define it here for readability
# intercept = coefs[0]
coefficients = coefs[1:].tolist()
# get the non-zero features from this model
non_zero_features = []
for feature, coefficient in zip(feature_columns, coefficients):
if coefficient != 0:
non_zero_features.append(feature)
# now train a new linear regression with just these non-zero features
X = df_train[non_zero_features]
X = sm.add_constant(X)
fit = sm.OLS(df_train["sc1"], X).fit()
# convert this model's parameters into a data frame
df_coef = self.ols_coefficients_to_dataframe(fit.params)
# create fake SKLL learner with these coefficients
learner = self.create_fake_skll_learner(df_coef)
# we used only the positive features
used_features = non_zero_features
return learner, fit, df_coef, used_features
[docs]
def train_lasso_fixed_lambda(
self, df_train: pd.DataFrame, feature_columns: List[str]
) -> Tuple[Learner, None, pd.DataFrame, List[str]]:
"""
Train a "LassoFixedLambda" model.
This is a Lasso model with a fixed lambda.
Parameters
----------
df_train : pandas.DataFrame
Data frame containing the features on which to train the model.
feature_columns : List[str]
A list of feature columns to use in training the model.
Returns
-------
learner : skll.learner.Learner
The SKLL learner object
fit : ``None``
This is always ``None`` since there is no OLS model fitted in this case.
df_coef : pandas.DataFrame
Data frame containing the model coefficients.
used_features : List[str]
A list of features used in the final model.
"""
# train a Lasso Regression model with a preset lambda
p_lambda = sqrt(len(df_train) * log10(len(feature_columns)))
# create a SKLL FeatureSet instance from the given data frame
fs_train = FeatureSet.from_data_frame(
df_train[feature_columns + ["sc1"]], "train", labels_column="sc1"
)
# note that 'alpha' in sklearn is different from this lambda
# so we need to normalize looking at the sklearn objective equation
alpha = p_lambda / len(df_train)
learner = Learner("Lasso", model_kwargs={"alpha": alpha, "positive": True})
learner.train(fs_train, grid_search=False)
# convert this model's parameters to a data frame
df_coef = self.skll_learner_params_to_dataframe(learner)
# there's no OLS fit object in this case
fit = None
# we used all the features
used_features = feature_columns
return learner, fit, df_coef, used_features
[docs]
def train_positive_lasso_cv(
self, df_train: pd.DataFrame, feature_columns: List[str]
) -> Tuple[Learner, None, pd.DataFrame, List[str]]:
"""
Train a "PositiveLassoCV" model.
Do feature selection using lasso regression optimized for
log likelihood using cross validation. All coefficients
are constrained to have positive values.
Parameters
----------
df_train : pandas.DataFrame
Data frame containing the features on which to train the model.
feature_columns : List[str]
A list of feature columns to use in training the model.
Returns
-------
learner : skll.learner.Learner
The SKLL learner object
fit : ``None``
This is always ``None`` since there is no OLS model fitted in this case.
df_coef : pandas.DataFrame
Data frame containing the model coefficients.
used_features : List[str]
A list of features used in the final model.
"""
# train a LassoCV outside of SKLL since it's not exposed there
X = df_train[feature_columns].values
y = df_train["sc1"].values
clf = LassoCV(cv=10, positive=True, random_state=1234567890)
model = clf.fit(X, y)
# save the non-zero model coefficients and intercept to a data frame
non_zero_features, non_zero_feature_values = [], []
for feature, coefficient in zip(feature_columns, model.coef_):
if coefficient != 0:
non_zero_features.append(feature)
non_zero_feature_values.append(coefficient)
# initialize the coefficient data frame with just the intercept
df_coef = pd.DataFrame([("Intercept", model.intercept_)])
df_coef = pd.concat(
[df_coef, pd.DataFrame(zip(non_zero_features, non_zero_feature_values))],
ignore_index=True,
)
df_coef.columns = ["feature", "coefficient"]
# create a fake SKLL learner with these non-zero weights
learner = self.create_fake_skll_learner(df_coef)
# there's no OLS fit object in this case
fit = None
# we used only the non-zero features
used_features = non_zero_features
return learner, fit, df_coef, used_features
[docs]
def train_score_weighted_lr(
self, df_train: pd.DataFrame, feature_columns: List[str]
) -> Tuple[Learner, RegressionResults, pd.DataFrame, List[str]]:
"""
Train a "ScoreWeightedLR" model.
This is a linear regression model weighted by score.
Parameters
----------
df_train : pandas.DataFrame
Data frame containing the features on which to train the model.
feature_columns : List[str]
A list of feature columns to use in training the model.
Returns
-------
learner : skll.learner.Learner
The SKLL learner object
fit : statsmodels.regression.linear_model.RegressionResults
A statsmodels regression results object.
df_coef : pandas.DataFrame
Data frame containing the model coefficients.
used_features : List[str]
A list of features used in the final model.
"""
# train weighted least squares regression
# get the feature columns
X = df_train[feature_columns]
# add the intercept
X = sm.add_constant(X)
# define the weights as inverse proportion of total
# number of data points for each score
score_level_dict = df_train["sc1"].value_counts()
expected_proportion = 1 / len(score_level_dict)
score_weights_dict = {
sc1: expected_proportion / count for sc1, count in score_level_dict.items()
}
weights = [score_weights_dict[sc1] for sc1 in df_train["sc1"]]
# fit the model
fit = sm.WLS(df_train["sc1"], X, weights=weights).fit()
df_coef = self.ols_coefficients_to_dataframe(fit.params)
learner = self.create_fake_skll_learner(df_coef)
# we used all the features
used_features = feature_columns
return learner, fit, df_coef, used_features
[docs]
def train_builtin_model(
self,
model_name: str,
df_train: pd.DataFrame,
experiment_id: str,
filedir: str,
file_format: str = "csv",
) -> Learner:
"""
Train one of the :ref:`built-in linear regression models <builtin_models>`.
Parameters
----------
model_name : str
Name of the built-in model to train.
df_train : pandas.DataFrame
Data frame containing the features on which to train
the model. The data frame must contain the ID column
named "spkitemid" and the numeric label column named
"sc1".
experiment_id : str
The experiment ID.
filedir : str
Path to the `output` experiment output directory.
file_format : str
The format in which to save files. One of {``"csv"``, ``"tsv"``,
``"xlsx"``}.
Defaults to ``"csv"``.
Returns
-------
learner : skll.learner.Learner
SKLL ``LinearRegression`` Learner object containing the coefficients
learned by training the built-in model.
"""
# get the columns that actually contain the feature values
feature_columns = [c for c in df_train.columns if c not in ["spkitemid", "sc1"]]
# LinearRegression
if model_name == "LinearRegression":
result = self.train_linear_regression(df_train, feature_columns)
# EqualWeightsLR
elif model_name == "EqualWeightsLR":
result = self.train_equal_weights_lr(df_train, feature_columns)
# RebalancedLR
elif model_name == "RebalancedLR":
result = self.train_rebalanced_lr(df_train, feature_columns)
# LassoFixedLambdaThenLR
elif model_name == "LassoFixedLambdaThenLR":
result = self.train_lasso_fixed_lambda_then_lr(df_train, feature_columns)
# PositiveLassoCVThenLR
elif model_name == "PositiveLassoCVThenLR":
result = self.train_positive_lasso_cv_then_lr(df_train, feature_columns)
# NNLR
elif model_name == "NNLR":
result = self.train_non_negative_lr(df_train, feature_columns)
# NNLRIterative
elif model_name == "NNLRIterative":
result = self.train_non_negative_lr_iterative(df_train, feature_columns)
# LassoFixedLambdaThenNNLR
elif model_name == "LassoFixedLambdaThenNNLR":
result = self.train_lasso_fixed_lambda_then_non_negative_lr(df_train, feature_columns)
# LassoFixedLambda
elif model_name == "LassoFixedLambda":
result = self.train_lasso_fixed_lambda(df_train, feature_columns)
# PositiveLassoCV
elif model_name == "PositiveLassoCV":
result = self.train_positive_lasso_cv(df_train, feature_columns)
# ScoreWeightedLR
elif model_name == "ScoreWeightedLR":
result = self.train_score_weighted_lr(df_train, feature_columns)
writer = DataWriter(experiment_id)
datasets = []
# unpack all results
learner, fit, df_coef, used_features = result
# add raw coefficients to frame list
datasets.append(DatasetDict({"name": "coefficients", "frame": df_coef}))
# compute the standardized and relative coefficients (betas) for the
# non-intercept features and save to a file
df_betas = df_coef.set_index("feature").loc[used_features]
df_betas = (
df_betas.multiply(df_train[used_features].std(), axis="index") / df_train["sc1"].std()
)
df_betas.columns = ["standardized"]
df_betas["relative"] = df_betas / sum(abs(df_betas["standardized"]))
df_betas.reset_index(inplace=True)
# add betas to frame list
datasets.append(DatasetDict({"name": "betas", "frame": df_betas}))
# save the OLS fit object and its summary to files
if fit:
ols_file = join(filedir, f"{experiment_id}.ols")
summary_file = join(filedir, f"{experiment_id}_ols_summary.txt")
with open(ols_file, "wb") as olsf, open(summary_file, "w") as summf:
pickle.dump(fit, olsf)
summf.write(str(fit.summary()))
# create a data frame with main model fit metrics and save to the file
df_model_fit = self.model_fit_to_dataframe(fit)
# add model_fit to frame list
datasets.append(DatasetDict({"name": "model_fit", "frame": df_model_fit}))
container = DataContainer(datasets=datasets)
writer.write_experiment_output(filedir, container, file_format=file_format)
self.learner = learner
return learner
[docs]
def train_skll_model(
self,
model_name: str,
df_train: pd.DataFrame,
custom_fixed_parameters: Optional[Dict[str, Any]] = None,
custom_objective: Optional[str] = None,
predict_expected_scores: bool = False,
skll_grid_search_jobs: int = 1,
) -> Tuple[Learner, str]:
"""
Train a SKLL classification or regression model.
Parameters
----------
model_name : str
Name of the SKLL model to train.
df_train : pandas.DataFrame
Data frame containing the features on which to train the model.
custom_fixed_parameters : Optional[Dict[str, Any]]
A dictionary containing any fixed parameters for the SKLL model.
Defaults to ``None``.
custom_objective : Optional[str]
Name of custom user-specified objective. If not specified
or ``None``, "neg_mean_squared_error" is used as the objective.
Defaults to ``None``.
predict_expected_scores : bool
Whether we want the trained classifiers to predict expected scores.
Defaults to ``False``.
skll_grid_search_jobs : int
Number of folds to run in parallel when using SKLL grid search.
Defaults to 1.
Returns
-------
learner: skll.learner.Learner
A SKLL Learner object of the appropriate type.
objective: str
The chosen tuning objective.
"""
# Instantiate the given SKLL learner and set its probability value
# and fixed parameters appropriately
model_kwargs = custom_fixed_parameters if custom_fixed_parameters is not None else {}
learner = Learner(
model_name, model_kwargs=model_kwargs, probability=predict_expected_scores
)
# get the features, IDs, and labels from the given data frame
feature_columns = [c for c in df_train.columns if c not in ["spkitemid", "sc1"]]
features = df_train[feature_columns].to_dict(orient="records")
ids = df_train["spkitemid"].tolist()
labels = df_train["sc1"].tolist()
# create a FeatureSet and train the model
fs = FeatureSet("train", ids=ids, labels=labels, features=features)
# If we are training a SKLL regressor, then we want to use either the
# user-specified objective or `neg_mean_squared_error`. If it's SKLL
# classifier, then the choice is between the user-specified objective
# and `f1_score_micro`.
if learner.model_type._estimator_type == "regressor":
objective = "neg_mean_squared_error" if not custom_objective else custom_objective
else:
objective = "f1_score_micro" if not custom_objective else custom_objective
learner.train(
fs, grid_search=True, grid_objective=objective, grid_jobs=skll_grid_search_jobs
)
# TODO: compute betas for linear SKLL models?
self.learner = learner
# return the SKLL learner object and the chosen objective
return learner, objective
[docs]
def train(
self,
configuration: Configuration,
data_container: DataContainer,
filedir: str,
file_format: str = "csv",
):
"""
Train the given model on the given data and save the results.
The main driver function to train the given model on the given
data and save the results in the given directories using the
given experiment ID as the prefix.
Parameters
----------
configuration : Configuration
A configuration object containing "experiment_id" and "model_name"
parameters.
data_container : DataContainer
A data container object containing "train_preprocessed_features"
data set.
filedir : str
Path to the "output" experiment output directory.
file_format : str
The format in which to save files. One of {``"csv"``, ``"tsv"``,
``"xlsx"``}.
Defaults to ``"csv"``.
Returns
-------
model : skll.learner.Learner
The trained SKLL Learner object.
"""
Analyzer.check_param_names(configuration, ["model_name", "experiment_id"])
Analyzer.check_frame_names(data_container, ["train_preprocessed_features"])
model_name = configuration["model_name"]
experiment_id = configuration["experiment_id"]
df_train = data_container["train_preprocessed_features"]
args = [model_name, df_train]
# add user-specified SKLL objective to the arguments if we are
# training a SKLL model
if is_skll_model(model_name):
kwargs = {
"custom_fixed_parameters": configuration["skll_fixed_parameters"],
"custom_objective": configuration["skll_objective"],
"predict_expected_scores": configuration["predict_expected_scores"],
"skll_grid_search_jobs": configuration["skll_grid_search_jobs"],
}
model, chosen_objective = self.train_skll_model(*args, **kwargs)
configuration["skll_objective"] = chosen_objective
else:
kwargs = {"file_format": file_format}
args.extend([experiment_id, filedir])
model = self.train_builtin_model(*args, **kwargs)
return model
[docs]
def predict(
self,
df: pd.DataFrame,
min_score: Optional[float] = None,
max_score: Optional[float] = None,
predict_expected: bool = False,
) -> pd.DataFrame:
"""
Get raw predictions from given SKLL model on data in given data frame.
Parameters
----------
df : pandas.DataFrame
Data frame containing features on which to make the predictions.
The data must contain pre-processed feature values, an ID column
named "spkitemid", and a label column named "sc1".
min_score : Optional[float]
Minimum score level to be used if computing expected scores.
If ``None``, trying to compute expected scores will raise an
exception.
Defaults to ``None``.
max_score : Optional[float]
Maximum score level to be used if computing expected scores.
If ``None``, trying to compute expected scores will raise an
exception.
Defaults to ``None``.
predict_expected : bool
Predict expected scores for classifiers that return probability
distributions over score. This will be ignored with a warning
if the specified model does not support probability distributions.
Note also that this assumes that the score range consists of
contiguous integers - starting at ``min_score`` and ending at
``max_score``.
Defaults to ``False``.
Returns
-------
df_predictions : pandas.DataFrame
Data frame containing the raw predictions, the IDs, and the human
scores.
Raises
------
ValueError
If no model has been trained yet.
ValueError
If the model cannot predict probability distributions and
``predict_expected`` is set to ``True``.
ValueError
If the score range specified by ``min_score`` and ``max_score``
does not match what the model predicts in its probability
distribution.
ValueError
If ``predict_expected`` is ``True`` but ``min_score`` and
``max_score`` are ``None``.
"""
model = self.learner
if model is None:
raise ValueError("No model has been trained yet.")
feature_columns = [c for c in df.columns if c not in ["spkitemid", "sc1"]]
features = df[feature_columns].to_dict(orient="records")
ids = df["spkitemid"].tolist()
# if we have the labels, save them in the featureset
labels = None
if "sc1" in df:
labels = df["sc1"].tolist()
fs = FeatureSet("data", ids=ids, labels=labels, features=features)
# if we are predicting expected scores, then call a different function
if predict_expected:
if min_score is None or max_score is None:
raise ValueError("Must specify 'min_score' and 'max_score' for expected scores.")
else:
predictions = compute_expected_scores_from_model(
model, fs, int(min_score), int(max_score)
)
else:
predictions = model.predict(fs)
df_predictions = pd.DataFrame()
df_predictions["spkitemid"] = ids
df_predictions["raw"] = predictions
# save the labels in the dataframe if they existed in the first place
if labels:
df_predictions["sc1"] = labels
return df_predictions
[docs]
def predict_train_and_test(
self, df_train: pd.DataFrame, df_test: pd.DataFrame, configuration: Configuration
) -> Tuple[Configuration, DataContainer]:
"""
Generate raw, scaled, and trimmed predictions on given data.
Parameters
----------
df_train : pandas.DataFrame
Data frame containing the pre-processed training set features.
df_test : pandas.DataFrame
Data frame containing the pre-processed test set features.
configuration : Configuration
A configuration object containing "trim_max" and "trim_min" keys.
Returns
-------
configuration : Configuration
A copy of the given configuration object also containing the
"train_predictions_mean", "train_predictions_sd", "human_labels_mean",
"human_labels_sd", "trim_min", and "trim_max" parameters.
data_container : DataContainer
A data container object containing the "pred_train", "pred_test",
and "postprocessing_params" data sets.
"""
Analyzer.check_param_names(configuration, ["trim_max", "trim_min", "trim_tolerance"])
(
trim_min,
trim_max,
trim_tolerance,
) = configuration.get_trim_min_max_tolerance()
# At this point, `trim_min` and `trim_max` are guaranteed to be
# valid numeric values but let's confirm to satisfy mypy
assert trim_min is not None and trim_max is not None and trim_tolerance is not None
predict_expected_scores = configuration["predict_expected_scores"]
df_train_predictions = self.predict(
df_train,
min_score=trim_min,
max_score=trim_max,
predict_expected=predict_expected_scores,
)
df_test_predictions = self.predict(
df_test,
min_score=trim_min,
max_score=trim_max,
predict_expected=predict_expected_scores,
)
# get the mean and SD of the training set predictions
train_predictions_mean = df_train_predictions["raw"].mean()
train_predictions_sd = df_train_predictions["raw"].std()
# get the mean and SD of the human labels
human_labels_mean = df_train["sc1"].mean()
human_labels_sd = df_train["sc1"].std()
self.logger.info("Processing train set predictions.")
feature_preprocessor = FeaturePreprocessor()
df_train_predictions = feature_preprocessor.process_predictions(
df_train_predictions,
train_predictions_mean,
train_predictions_sd,
human_labels_mean,
human_labels_sd,
trim_min,
trim_max,
trim_tolerance,
)
self.logger.info("Processing test set predictions.")
df_test_predictions = feature_preprocessor.process_predictions(
df_test_predictions,
train_predictions_mean,
train_predictions_sd,
human_labels_mean,
human_labels_sd,
trim_min,
trim_max,
trim_tolerance,
)
df_postproc_params = pd.DataFrame(
[
{
"trim_min": trim_min,
"trim_max": trim_max,
"trim_tolerance": trim_tolerance,
"h1_mean": human_labels_mean,
"h1_sd": human_labels_sd,
"train_predictions_mean": train_predictions_mean,
"train_predictions_sd": train_predictions_sd,
}
]
)
datasets = [
DatasetDict({"name": "pred_train", "frame": df_train_predictions}),
DatasetDict({"name": "pred_test", "frame": df_test_predictions}),
DatasetDict({"name": "postprocessing_params", "frame": df_postproc_params}),
]
# configuration options that are entirely for internal use
internal_options_dict = {
"train_predictions_mean": train_predictions_mean,
"train_predictions_sd": train_predictions_sd,
"human_labels_mean": human_labels_mean,
"human_labels_sd": human_labels_sd,
}
new_configuration = configuration.copy()
for key, value in internal_options_dict.items():
new_configuration[key] = value
return new_configuration, DataContainer(datasets=datasets)
[docs]
def get_feature_names(self) -> Optional[List[str]]:
"""
Get the feature names, if available.
Returns
-------
feature_names : Optional[List[str]]
A list of feature names, or None if no learner was trained.
"""
if self.learner is not None:
return self.learner.feat_vectorizer.get_feature_names_out().tolist()
return None
[docs]
def get_intercept(self) -> Optional[float]:
"""
Get the intercept of the model, if available.
Returns
-------
intercept : Optional[float]
The intercept of the model.
"""
if self.learner is not None:
return self.learner.model.intercept_
return None
[docs]
def get_coefficients(self) -> Optional[np.ndarray]:
"""
Get the coefficients of the model, if available.
Returns
-------
coefficients : Optional[np.ndarray]
The coefficients of the model, if available.
"""
if self.learner is not None:
return self.learner.model.coef_
return None
[docs]
def scale_coefficients(self, configuration: Configuration) -> DataContainer:
"""
Scale coefficients using human scores & training set predictions.
This procedure approximates what is done in operational
setting but does not apply trimming to predictions.
Parameters
----------
configuration : Configuration
A configuration object containing the "train_predictions_mean",
"train_predictions_sd", and "human_labels_sd" parameters.
Returns
-------
data_container : DataContainer
A container object containing the "coefficients_scaled"
dataset. The frame for this dataset contains the scaled
coefficients and the feature names, along with the
intercept.
Raises
------
RuntimeError
If the model is non-linear and no coefficients
are available.
"""
Analyzer.check_param_names(
configuration,
["train_predictions_mean", "train_predictions_sd", "human_labels_sd"],
)
train_predictions_mean = configuration["train_predictions_mean"]
train_predictions_sd = configuration["train_predictions_sd"]
h1_sd = configuration["human_labels_sd"]
feature_names = self.get_feature_names()
# try to get the model coefficients, if available
try:
coefficients = self.get_coefficients()
except AttributeError:
raise RuntimeError("no coefficients available for this model.")
intercept = self.get_intercept()
# scale the coefficients and the intercept
scaled_coefficients = coefficients * h1_sd / train_predictions_sd
# adjust the intercept to set the mean predicted score
# to the mean of the training variable
new_intercept = intercept * (h1_sd / train_predictions_sd)
new_intercept += train_predictions_mean * (1 - h1_sd / train_predictions_sd)
intercept_and_feature_names = (
["Intercept"] + feature_names if feature_names else ["Intercept"]
)
intercept_and_feature_values = [new_intercept] + list(scaled_coefficients)
# create a data frame with new values
df_scaled_coefficients = pd.DataFrame(
{
"feature": intercept_and_feature_names,
"coefficient": intercept_and_feature_values,
},
columns=["feature", "coefficient"],
)
scaled_dataset = [
DatasetDict({"name": "coefficients_scaled", "frame": df_scaled_coefficients})
]
return DataContainer(datasets=scaled_dataset)