Cross Sectional Imputation¶

Imputations for cross-sectional data.

impyute.imputation.cs.random(data, **kwargs)[source]

Fill missing values in with a randomly selected value from the same column.

Parameters: data: numpy.ndarray Data to impute. numpy.ndarray Imputed data.
impyute.imputation.cs.mean(data, **kwargs)[source]

Substitute missing values with the mean of that column.

Parameters: data: numpy.ndarray Data to impute. numpy.ndarray Imputed data.
impyute.imputation.cs.mode(data, **kwargs)[source]

Substitute missing values with the mode of that column(most frequent).

In the case that there is a tie (there are multiple, most frequent values) for a column randomly pick one of them.

Parameters: data: numpy.ndarray Data to impute. numpy.ndarray Imputed data.
impyute.imputation.cs.median(data, **kwargs)[source]

Substitute missing values with the median of that column(middle).

Parameters: data: numpy.ndarray Data to impute. numpy.ndarray Imputed data.
impyute.imputation.cs.mice(data, **kwargs)[source]

Multivariate Imputation by Chained Equations

Reference:
Buuren, S. V., & Groothuis-Oudshoorn, K. (2011). Mice: Multivariate Imputation by Chained Equations in R. Journal of Statistical Software, 45(3). doi:10.18637/jss.v045.i03

Implementation follows the main idea from the paper above. Differs in decision of which variable to regress on (here, I choose it at random). Also differs in stopping criterion (here the model stops after change in prediction from previous prediction is less than 10%).

Parameters: data: numpy.ndarray Data to impute. numpy.ndarray Imputed data.
impyute.imputation.cs.em(data, loops=50, **kwargs)[source]

Imputes given data using expectation maximization.

E-step: Calculates the expected complete data log likelihood ratio. M-step: Finds the parameters that maximize the log likelihood of the complete data.

Parameters: data: numpy.nd.array Data to impute. loops: int Number of em iterations to run before breaking. inplace: boolean If True, operate on the numpy array reference numpy.nd.array Imputed data.
impyute.imputation.cs.fast_knn(data, k=3, eps=0, p=2, distance_upper_bound=inf, leafsize=10, **kwargs)[source]

Impute using a variant of the nearest neighbours approach

Basic idea: Impute array with a basic mean impute and then use the resulting complete array to construct a KDTree. Use this KDTree to compute nearest neighbours. After finding k nearest neighbours, take the weighted average of them. Basically, find the nearest row in terms of distance

This approach is much, much faster than the other implementation (fit+transform for each subset) which is almost prohibitively expensive.

Parameters: data: numpy.ndarray 2D matrix to impute. k: int, optional Parameter used for method querying the KDTree class object. Number of neighbours used in the KNN query. Refer to the docs for [scipy.spatial.KDTree.query](https://docs.scipy.org/doc/scipy/reference/generated/scipy.spatial.KDTree.query.html). eps: nonnegative float, optional Parameter used for method querying the KDTree class object. From the SciPy docs: “Return approximate nearest neighbors; the kth returned value is guaranteed to be no further than (1+eps) times the distance to the real kth nearest neighbor”. Refer to the docs for [scipy.spatial.KDTree.query](https://docs.scipy.org/doc/scipy/reference/generated/scipy.spatial.KDTree.query.html). p : float, 1<=p<=infinity, optional Parameter used for method querying the KDTree class object. Straight from the SciPy docs: “Which Minkowski p-norm to use. 1 is the sum-of-absolute-values Manhattan distance 2 is the usual Euclidean distance infinity is the maximum-coordinate-difference distance”. Refer to the docs for [scipy.spatial.KDTree.query](https://docs.scipy.org/doc/scipy/reference/generated/scipy.spatial.KDTree.query.html). distance_upper_bound : nonnegative float, optional Parameter used for method querying the KDTree class object. Straight from the SciPy docs: “Return only neighbors within this distance. This is used to prune tree searches, so if you are doing a series of nearest-neighbor queries, it may help to supply the distance to the nearest neighbor of the most recent point.” Refer to the docs for [scipy.spatial.KDTree.query](https://docs.scipy.org/doc/scipy/reference/generated/scipy.spatial.KDTree.query.html). leafsize: int, optional Parameter used for construction of the KDTree class object. Straight from the SciPy docs: “The number of points at which the algorithm switches over to brute-force. Has to be positive”. Refer to the docs for [scipy.spatial.KDTree](https://docs.scipy.org/doc/scipy-0.14.0/reference/generated/scipy.spatial.KDTree.html) for more information. numpy.ndarray Imputed data.

Examples

>>> data = np.arange(25).reshape((5, 5)).astype(np.float)
>>> data[0][2] =  np.nan
>>> data
array([[ 0.,  1., nan,  3.,  4.],
[ 5.,  6.,  7.,  8.,  9.],
[10., 11., 12., 13., 14.],
[15., 16., 17., 18., 19.],
[20., 21., 22., 23., 24.]])
>> fast_knn(data, k=1) # Weighted average (by distance) of nearest 1 neighbour
array([[ 0.,  1.,  7.,  3.,  4.],
[ 5.,  6.,  7.,  8.,  9.],
[10., 11., 12., 13., 14.],
[15., 16., 17., 18., 19.],
[20., 21., 22., 23., 24.]])
>> fast_knn(data, k=2) # Weighted average of nearest 2 neighbours
array([[ 0.        ,  1.        , 10.08608891,  3.        ,  4.        ],
[ 5.        ,  6.        ,  7.        ,  8.        ,  9.        ],
[10.        , 11.        , 12.        , 13.        , 14.        ],
[15.        , 16.        , 17.        , 18.        , 19.        ],
[20.        , 21.        , 22.        , 23.        , 24.        ]])
>> fast_knn(data, k=3)
array([[ 0.        ,  1.        , 13.40249283,  3.        ,  4.        ],
[ 5.        ,  6.        ,  7.        ,  8.        ,  9.        ],
[10.        , 11.        , 12.        , 13.        , 14.        ],
[15.        , 16.        , 17.        , 18.        , 19.        ],
[20.        , 21.        , 22.        , 23.        , 24.        ]])
>> fast_knn(data, k=5) # There are at most only 4 neighbours. Raises error
...
IndexError: index 5 is out of bounds for axis 0 with size 5