return RegressionModelsMethods(self._df)
@classmethod
def _predict(cls, df, estimator, *args, **kwargs):
data = df.data.values
eval_MSE = kwargs.get('eval_MSE', False)
if eval_MSE:
y, MSE = estimator.predict(data, *args, **kwargs)
if y.ndim == 1:
y = df._constructor_sliced(y, index=df.index)
MSE = df._constructor_sliced(MSE, index=df.index)
else:
y = df._constructor(y, index=df.index)
MSE = df._constructor(MSE, index=df.index)
return y, MSE
else:
y = estimator.predict(data, *args, **kwargs)
if y.ndim == 1:
y = df._constructor_sliced(y, index=df.index)
else:
y = df._constructor(y, index=df.index)
return y
class RegressionModelsMethods(_AccessorMethods):
_module_name = 'sklearn.gaussian_process.regression_models'
_regression_methods = ['constant', 'linear', 'quadratic']
_attach_methods(RegressionModelsMethods, _wrap_data_func, _regression_methods)
if is_integer_dtype(values):
# integer raises an error in normalize
values = values.astype(np.float)
result = func(values, *args, **kwargs)
result = self._constructor(result,
index=self._data.index,
columns=self._data.columns)
else:
# ModelSeries
values = np.atleast_2d(self._df.values)
if is_integer_dtype(values):
values = values.astype(np.float)
result = func(values, *args, **kwargs)
result = self._constructor(result[0],
index=self._df.index,
name=self._df.name)
return result
f.__doc__ = ("""
Call ``%s`` using automatic mapping.
- ``X``: ``ModelFrame.data``
""" % func_name)
return f
_attach_methods(PreprocessingMethods, _wrap_func, _preprocessing_methods)
# Pairwise metrics
@property
def pairwise(self):
"""Not implemented"""
raise NotImplementedError
# y_true and y_pred
_classification_methods = ['accuracy_score', 'classification_report',
'hamming_loss', 'jaccard_similarity_score',
'matthews_corrcoef', 'zero_one_loss']
_regression_methods = ['explained_variance_score', 'mean_absolute_error',
'mean_squared_error', 'r2_score']
_cluster_methods = ['mutual_info_score']
_true_pred_methods = (_classification_methods + _regression_methods +
_cluster_methods)
_attach_methods(MetricsMethods, _wrap_target_pred_func, _true_pred_methods)
# methods which doesn't take additional arguments
_cluster_methods_noargs = ['adjusted_mutual_info_score',
'adjusted_rand_score',
'completeness_score',
'homogeneity_completeness_v_measure',
'homogeneity_score',
'normalized_mutual_info_score',
'v_measure_score']
_attach_methods(MetricsMethods, _wrap_target_pred_noargs, _cluster_methods_noargs)
- ``y``: ``ModelFrame.target``
"""
func = self._module.lasso_stability_path
data = self._data
target = self._target
alpha_grid, scores_path = func(data.values,
y=target.values,
*args,
**kwargs)
scores_path = self._constructor(scores_path, index=data.columns)
return alpha_grid, scores_path
def orthogonal_mp_gram(self, *args, **kwargs):
"""
Call ``sklearn.linear_model.orthogonal_mp_gram`` using automatic mapping.
- ``Gram``: ``ModelFrame.data.T.dot(ModelFrame.data)``
- ``Xy``: ``ModelFrame.data.T.dot(ModelFrame.target)``
"""
func = self._module.orthogonal_mp_gram
data = self._data.values
target = self._target.values
gram = data.T.dot(data)
Xy = data.T.dot(target)
coef = func(gram, Xy, *args, **kwargs)
return coef
_lm_methods = ['orthogonal_mp']
_attach_methods(LinearModelMethods, _wrap_data_target_func, _lm_methods)
Call ``sklearn.linear_model.lasso_stability_path`` using automatic mapping.
- ``X``: ``ModelFrame.data``
- ``y``: ``ModelFrame.target``
"""
func = self._module.lasso_stability_path
data = self._data
target = self._target
alpha_grid, scores_path = func(data.values, y=target.values, *args, **kwargs)
scores_path = self._constructor(scores_path, index=data.columns)
return alpha_grid, scores_path
def orthogonal_mp_gram(self, *args, **kwargs):
"""
Call ``sklearn.linear_model.orthogonal_mp_gram`` using automatic mapping.
- ``Gram``: ``ModelFrame.data.T.dot(ModelFrame.data)``
- ``Xy``: ``ModelFrame.data.T.dot(ModelFrame.target)``
"""
func = self._module.orthogonal_mp_gram
data = self._data.values
target = self._target.values
gram = data.T.dot(data)
Xy = data.T.dot(target)
coef = func(gram, Xy, *args, **kwargs)
return coef
_lm_methods = ['orthogonal_mp']
_attach_methods(LinearModelMethods, _wrap_data_target_func, _lm_methods)
labels = self._constructor_sliced(labels, index=data.index)
return cluster_centers, labels
def spectral_clustering(self, *args, **kwargs):
"""
Call ``sklearn.cluster.spectral_clustering`` using automatic mapping.
- ``affinity``: ``ModelFrame.data``
"""
func = self._module.spectral_clustering
data = self._data
labels = func(data.values, *args, **kwargs)
labels = self._constructor_sliced(labels, index=data.index)
return labels
# Biclustering
@property
def bicluster(self):
"""Property to access ``sklearn.cluster.bicluster``"""
return self._bicluster
@cache_readonly
def _bicluster(self):
return _AccessorMethods(self._df,
module_name='sklearn.cluster.bicluster')
_cluster_methods = ['estimate_bandwidth', 'ward_tree']
_attach_methods(ClusterMethods, _wrap_data_func, _cluster_methods)
- ``x``: ``ModelFrame.target_name``
""" % func_name)
return f
def _wrap_data_plot(func, func_name):
"""
Wrapper for plotting with data
"""
def f(self, *args, **kwargs):
return func(data=self._df, *args, **kwargs)
f.__doc__ = (
"""
Call ``%s`` using automatic mapping.
- ``data``: ``ModelFrame``
""" % func_name)
return f
_xy_plots = ['jointplot', 'lmplot', 'regplot', 'residplot']
_attach_methods(SeabornMethods, _wrap_xy_plot, _xy_plots)
_categorical_plots = ['factorplot', 'boxplot', 'violinplot', 'stripplot',
'pointplot', 'barplot']
_attach_methods(SeabornMethods, _wrap_categorical_plot, _categorical_plots)
_data_plots = ['pairplot']
_attach_methods(SeabornMethods, _wrap_data_plot, _data_plots)
#!/usr/bin/env python
from pandas_ml.core.accessor import _AccessorMethods, _attach_methods, _wrap_data_func
class NeighborsMethods(_AccessorMethods):
"""
Accessor to ``sklearn.neighbors``.
"""
_module_name = 'sklearn.neighbors'
_neighbor_methods = ['kneighbors_graph', 'radius_neighbors_graph']
_attach_methods(NeighborsMethods, _wrap_data_func, _neighbor_methods)
return f
def _wrap_data_plot(func, func_name):
"""
Wrapper for plotting with data
"""
def f(self, *args, **kwargs):
return func(data=self._df, *args, **kwargs)
f.__doc__ = (
"""
Call ``%s`` using automatic mapping.
- ``data``: ``ModelFrame``
"""
% func_name
)
return f
_xy_plots = ["jointplot", "lmplot", "regplot", "residplot"]
_attach_methods(SeabornMethods, _wrap_xy_plot, _xy_plots)
_categorical_plots = ["factorplot", "boxplot", "violinplot", "stripplot", "pointplot", "barplot"]
_attach_methods(SeabornMethods, _wrap_categorical_plot, _categorical_plots)
_data_plots = ["pairplot"]
_attach_methods(SeabornMethods, _wrap_data_plot, _data_plots)
# Pairwise metrics
@property
def pairwise(self):
"""Not implemented"""
raise NotImplementedError
# y_true and y_pred
_classification_methods = [
'accuracy_score', 'classification_report', 'hamming_loss', 'jaccard_score',
'matthews_corrcoef', 'zero_one_loss'
]
_regression_methods = [
'explained_variance_score', 'mean_absolute_error', 'mean_squared_error',
'r2_score'
]
_cluster_methods = ['mutual_info_score']
_true_pred_methods = (_classification_methods + _regression_methods +
_cluster_methods)
_attach_methods(MetricsMethods, _wrap_target_pred_func, _true_pred_methods)
# methods which doesn't take additional arguments
_cluster_methods_noargs = [
'adjusted_mutual_info_score', 'adjusted_rand_score', 'completeness_score',
'homogeneity_completeness_v_measure', 'homogeneity_score',
'normalized_mutual_info_score', 'v_measure_score'
]
_attach_methods(MetricsMethods, _wrap_target_pred_noargs,
_cluster_methods_noargs)
#!/usr/bin/env python
from pandas_ml.core.accessor import _AccessorMethods, _attach_methods, _wrap_data_func
class NeighborsMethods(_AccessorMethods):
"""
Accessor to ``sklearn.neighbors``.
"""
_module_name = 'sklearn.neighbors'
_neighbor_methods = ['kneighbors_graph', 'radius_neighbors_graph']
_attach_methods(NeighborsMethods, _wrap_data_func, _neighbor_methods)
if isinstance(self._df, ModelFrame):
values = self._data.values
if is_integer_dtype(values):
# integer raises an error in normalize
values = values.astype(np.float)
result = func(values, *args, **kwargs)
result = self._constructor(result, index=self._data.index,
columns=self._data.columns)
else:
# ModelSeries
values = np.atleast_2d(self._df.values)
if is_integer_dtype(values):
values = values.astype(np.float)
result = func(values, *args, **kwargs)
result = self._constructor(result[0], index=self._df.index,
name=self._df.name)
return result
f.__doc__ = (
"""
Call ``%s`` using automatic mapping.
- ``X``: ``ModelFrame.data``
""" % func_name)
return f
_attach_methods(PreprocessingMethods, _wrap_func, _preprocessing_methods)
#!/usr/bin/env python
from pandas_ml.core.accessor import _AccessorMethods, _attach_methods, _wrap_data_target_func
class FeatureSelectionMethods(_AccessorMethods):
"""
Accessor to ``sklearn.feature_selection``.
"""
_module_name = 'sklearn.feature_selection'
_fs_methods = ['chi2', 'f_classif', 'f_regression']
_attach_methods(FeatureSelectionMethods, _wrap_data_target_func, _fs_methods)
return RegressionModelsMethods(self._df)
@classmethod
def _predict(cls, df, estimator, *args, **kwargs):
data = df.data.values
eval_MSE = kwargs.get('eval_MSE', False)
if eval_MSE:
y, MSE = estimator.predict(data, *args, **kwargs)
if y.ndim == 1:
y = df._constructor_sliced(y, index=df.index)
MSE = df._constructor_sliced(MSE, index=df.index)
else:
y = df._constructor(y, index=df.index)
MSE = df._constructor(MSE, index=df.index)
return y, MSE
else:
y = estimator.predict(data, *args, **kwargs)
if y.ndim == 1:
y = df._constructor_sliced(y, index=df.index)
else:
y = df._constructor(y, index=df.index)
return y
class RegressionModelsMethods(_AccessorMethods):
_module_name = 'sklearn.gaussian_process.regression_models'
_regression_methods = ['constant', 'linear', 'quadratic']
_attach_methods(RegressionModelsMethods, _wrap_data_func, _regression_methods)
cluster_centers, labels = func(data.values, *args, **kwargs)
labels = self._constructor_sliced(labels, index=data.index)
return cluster_centers, labels
def spectral_clustering(self, *args, **kwargs):
"""
Call ``sklearn.cluster.spectral_clustering`` using automatic mapping.
- ``affinity``: ``ModelFrame.data``
"""
func = self._module.spectral_clustering
data = self._data
labels = func(data.values, *args, **kwargs)
labels = self._constructor_sliced(labels, index=data.index)
return labels
# Biclustering
@property
def bicluster(self):
"""Property to access ``sklearn.cluster.bicluster``"""
return self._bicluster
@cache_readonly
def _bicluster(self):
return _AccessorMethods(self._df, module_name='sklearn.cluster.bicluster')
_cluster_methods = ['estimate_bandwidth', 'ward_tree']
_attach_methods(ClusterMethods, _wrap_data_func, _cluster_methods)
