{ "id": 21957, "title": "gridsearch for feature weighting", "url": "/competitions/facebook-v-predicting-check-ins/discussion/21957", "author_name": "simonm3", "post_date": "2016-06-29T13:27:18.137000", "votes": 0, "comment_count": 0, "views": 204, "content": "

Would be useful to run gridearch to help optimise feature weights prior to KNN. I have got this to work...sort of....but does anyone have a better way?

\n\n

The current solution requires pasting the feature list into the class every time the features change. Also it has to be first in the pipeline so I can pass a dataframe in to match up the column names with the weights. After the first pipeline step it becomes a np.array.

\n\n
class FeatureWeights(TransformerMixin, BaseEstimator):\n""" base class for applying weights to features """    \n\ndef __init__(self):\n    """ Override to explicitly list params \n        BaseEstimator does not allow *args, **kwargs """\n    pass\n\ndef transform(self, x):\n    """ NOTE assumes 1st in pipeline and passed pandas dataframe \n    otherwise cannot match numpy array columns to unordered\n    feature weights dict.\n    sklearn does not allow pass **params or set class variables after init\n    """\n    cols = x.columns\n    x = x.values.copy()\n    for i, col in enumerate(cols):\n        x[:,i] = x[:,i] * self.__dict__[col]\n    return x\n\ndef fit(self, x, y=None):\n    return self\n\nclass FBFeatureWeights(FeatureWeights):\n""" weights features for models such as KNN """\n# copy features into __init__ parameters \ndef __init__(self, x, y, time, accuracy, dayofweek, hour, month):\n    self.__dict__ = locals().copy()\n    del self.__dict__["self"]\n
", "messages": [ { "id": 125440, "postDate": "2016-06-29T13:27:18.137Z", "content": "

Would be useful to run gridearch to help optimise feature weights prior to KNN. I have got this to work...sort of....but does anyone have a better way?

\n\n

The current solution requires pasting the feature list into the class every time the features change. Also it has to be first in the pipeline so I can pass a dataframe in to match up the column names with the weights. After the first pipeline step it becomes a np.array.

\n\n
class FeatureWeights(TransformerMixin, BaseEstimator):\n""" base class for applying weights to features """    \n\ndef __init__(self):\n    """ Override to explicitly list params \n        BaseEstimator does not allow *args, **kwargs """\n    pass\n\ndef transform(self, x):\n    """ NOTE assumes 1st in pipeline and passed pandas dataframe \n    otherwise cannot match numpy array columns to unordered\n    feature weights dict.\n    sklearn does not allow pass **params or set class variables after init\n    """\n    cols = x.columns\n    x = x.values.copy()\n    for i, col in enumerate(cols):\n        x[:,i] = x[:,i] * self.__dict__[col]\n    return x\n\ndef fit(self, x, y=None):\n    return self\n\nclass FBFeatureWeights(FeatureWeights):\n""" weights features for models such as KNN """\n# copy features into __init__ parameters \ndef __init__(self, x, y, time, accuracy, dayofweek, hour, month):\n    self.__dict__ = locals().copy()\n    del self.__dict__["self"]\n
", "rawMarkdown": "Would be useful to run gridearch to help optimise feature weights prior to KNN. I have got this to work...sort of....but does anyone have a better way?\r\n\r\nThe current solution requires pasting the feature list into the class every time the features change. Also it has to be first in the pipeline so I can pass a dataframe in to match up the column names with the weights. After the first pipeline step it becomes a np.array.\r\n\r\n\r\n class FeatureWeights(TransformerMixin, BaseEstimator):\r\n \"\"\" base class for applying weights to features \"\"\" \r\n \r\n def __init__(self):\r\n \"\"\" Override to explicitly list params \r\n BaseEstimator does not allow *args, **kwargs \"\"\"\r\n pass\r\n \r\n def transform(self, x):\r\n \"\"\" NOTE assumes 1st in pipeline and passed pandas dataframe \r\n otherwise cannot match numpy array columns to unordered\r\n feature weights dict.\r\n sklearn does not allow pass **params or set class variables after init\r\n \"\"\"\r\n cols = x.columns\r\n x = x.values.copy()\r\n for i, col in enumerate(cols):\r\n x[:,i] = x[:,i] * self.__dict__[col]\r\n return x\r\n\r\n def fit(self, x, y=None):\r\n return self\r\n\r\n class FBFeatureWeights(FeatureWeights):\r\n \"\"\" weights features for models such as KNN \"\"\"\r\n # copy features into __init__ parameters \r\n def __init__(self, x, y, time, accuracy, dayofweek, hour, month):\r\n self.__dict__ = locals().copy()\r\n del self.__dict__[\"self\"]" } ], "comments": [], "raw_markdown_by_id": { "125440": "Would be useful to run gridearch to help optimise feature weights prior to KNN. I have got this to work...sort of....but does anyone have a better way?\r\n\r\nThe current solution requires pasting the feature list into the class every time the features change. Also it has to be first in the pipeline so I can pass a dataframe in to match up the column names with the weights. After the first pipeline step it becomes a np.array.\r\n\r\n\r\n class FeatureWeights(TransformerMixin, BaseEstimator):\r\n \"\"\" base class for applying weights to features \"\"\" \r\n \r\n def __init__(self):\r\n \"\"\" Override to explicitly list params \r\n BaseEstimator does not allow *args, **kwargs \"\"\"\r\n pass\r\n \r\n def transform(self, x):\r\n \"\"\" NOTE assumes 1st in pipeline and passed pandas dataframe \r\n otherwise cannot match numpy array columns to unordered\r\n feature weights dict.\r\n sklearn does not allow pass **params or set class variables after init\r\n \"\"\"\r\n cols = x.columns\r\n x = x.values.copy()\r\n for i, col in enumerate(cols):\r\n x[:,i] = x[:,i] * self.__dict__[col]\r\n return x\r\n\r\n def fit(self, x, y=None):\r\n return self\r\n\r\n class FBFeatureWeights(FeatureWeights):\r\n \"\"\" weights features for models such as KNN \"\"\"\r\n # copy features into __init__ parameters \r\n def __init__(self, x, y, time, accuracy, dayofweek, hour, month):\r\n self.__dict__ = locals().copy()\r\n del self.__dict__[\"self\"]" } }