Ctrl+K

watex.methods.MXS#

class watex.methods.MXS(kname=None, aqname=None, threshold=None, method='naive', trailer='*', keep_label_0=False, random_state=42, n_groups=3, sep=None, prefix=None, **kws)[source]#

Mixture Learning Strategy (MXS)

The use of machine learning for k-parameter prediction seems an alternative way to reduce the cost of data collection thereby saving money. However, the borehole data comes with a lot of missing k since the parameter is strongly tied to the aquifer after the pumping test. In other words, the k-parameter collection is feasible if the layer in the well is an aquifer. Unfortunately, predicting some samples of k in a large set of missing data remains an issue using the classical supervised learning methods. We, therefore propose an alternative approach called a mixture learning strategy (MXS) to solve these double issues. It entails predicting upstream a naïve group of aquifers (NGA) combined with the real values k to counterbalance the missing values and yield an optimal prediction score. The method, first, implies the K-Means and Hierarchical Agglomerative Clustering (HAC) algorithms. K-Means and HAC are used for NGA label predicting necessary the MXS label merging.

Parameters:

  • kname (str, int) –

    Name of permeability coefficient columns. kname allows to retrieve the

    permeability coefficient ‘k’ in a specific dataframe. If integer is passed, it assumes the index of the dataframe fits the ‘k’ columns. Note that integer value must not be out the dataframe size along axis 1. Commonly

    kname needs to be supplied when a dataframe is passed as a positional

    or keyword argument.

  • aqname (str, optional,) –

    Name of aquifer group column. aqname allows to retrieve the

    aquifer group arr_aq value in a specific dataframe. Commonly

    aqname needs to be supplied when a dataframe is passed as a positional

    or keyword argument. Note that it is not mandatory to have a group of aquifer in the log data. It is needed only if the label similarity needs to be calculated.

  • threshold (float, default=None) – The threshold from which, label in ‘k’ array can be considered similar than the one in NGA labels ‘y_pred’. The default is ‘None’ which means none rule is considered and the high preponderence or occurence in the data compared to other labels is considered as the most representative and similar. Setting the rule instead by fixing the threshold is recommended especially in a huge dataset.

  • n_groups (int, default=3) – The number of aquifer n_groups to form as well as the number of centroids to generate. If a idea about the number of aquifer group in the areas, it should be used instead. Hiwever, it is recommended to validate this number using the ‘elbow plot’ or the ‘silhouette plot’ or the Hierachical Agglomerative Clustering dendrogram. Refer to plot_elbow() or plotSilhouette() or :func:~.watex.view.plotDendrogram` for plotting purpose.

  • keep_label_0 (bool, default=False) –

    The prediction already include the label 0. However, including 0 in

    the predicted label refers to ‘k=0’ i.e. no permeability coefficient equals to 0, which is not True in principle, because all rocks have a permeability coefficient ‘k’. Here we considered ‘k=0’ as an undefined permeability coefficient. Therefore, ‘0’ , can be exclude since, it can also considered as a missing ‘k’-value. If predicted ‘0’ is in the target it should mean a missing ‘k’-value rather than being a concrete label. Therefore, to avoid any confusion, ‘0’ is altered to ‘1’ so the value +1 is used to move forward all class labels thereby excluding the ‘0’ label. To force include 0 in the label, set keep_label_0 to True.

    sep: str, default’’

    Separator between the true labels ‘y_true’ and predicted NGA labels. Sep is used to rewrite the MXS labels. Mostly the MXS labels is a combinaison with the true label of permeability coefficient ‘k’ and the label of NGA to compose new similarity labels. For instance

    >>> true_labels=['k1', 'k2', 'k3'] ; NGA_labels =['II', 'I', 'UV']
>>> # gives
>>> MXS_labels= ['k1_II', 'k2_I', 'k3_UV']

    where the seperator sep is set to _. This happens especially when one of the label (NGA or true_labels) is not a numeric datatype and a similariy is found between ‘k1’ and ‘II’, ‘k2’ and ‘I’ and so on.

    prefix: str, default=’’

    prefix is used to rename the true_labels i.e the true valid-k. For instance:

    >>> k_valid =[1, 2, ..] -> k_new = [k1, k2, ...]

    where ‘k’ is the prefix.

    method: str [‘naive’, ‘strict’], default=’naive’

    The kind of strategy to compute the representativity of a label in the predicted array ‘y_pred’. It can also be ‘strict’. Indeed:

    • naive computes the importance of the label by the number of its

      occurence for this specific label in the array ‘y_true’. It does not take into account of the occurence of other existing labels. This is usefull for unbalanced class labels in y_true.

    • strict computes the importance of the label by the number of

      occurence in the whole valid y_true i.e. under the total of occurence of all the labels that exist in the whole ‘arra_aq’. This can give a suitable anaylse results if the data is not unbalanced for each labels in y_pred.

    trailer: str, default=’*’

    The Mixture strategy marker to differentiate the existing class label in ‘y_true’ with the predicted labels ‘y_pred’ especially when the the same class labels are also present the true label with the same label-identifier name. This usefull to avoid any confusion for both labels in y_true and y_pred for better demarcation and distinction. Note that if the trailer`is set to ``None` and both y_true and y_pred are numeric data, the labels in y_pred are systematically renamed to be distinct with the ones in the ‘y_true’. For instance

    >>> true_labels=[1, 2, 3] ; NGA_labels =[0, 1, 2]
>>> # with trailer , MXS labels should be
>>>  MXS_labels= ['0', '1*', '2*', '3'] # 1 and 2 are in true_labels
>>> # with no trailer
>>> MXS_labels= [0, 4, 5, 3] # 1 and 2 have been changed to [4, 5]

  • verbose (int, default is 0) – Control the level of verbosity. Higher value lead to more messages.

Examples

>>> from watex.datasets import load_hlogs
>>> from watex.methods.hydro import MXS
>>> hdata= load_hlogs (as_frame =True)
>>> # drop the 'remark' columns since there is no valid data
>>> hdata.drop (columns ='remark', inplace =True)
>>> mxs = MXS (kname ='k').fit(hdata)
>>> # predict the default NGA
>>> mxs.predictNGA() # default prediction with n_groups =3
>>> # make MXS labels using the default 'k' categorization
>>> ymxs=mxs.makeyMXS(categorize_k=True, default_func=True)
>>> mxs.yNGA_ [62:74]
Out[43]: array([1, 2, 2, 2, 3, 1, 2, 1, 2, 2, 1, 2])
>>> ymxs[62:74]
Out[44]: array([ 1, 22, 22, 22,  3,  1, 22,  1, 22, 22,  1, 22])
>>> # to get the label similariry , need to provide the
>>> # the column name of aquifer group and fit again like
>>> mxs = MXS (kname ='k', aqname ='aquifer_group').fit(hdata)
>>> sim = mxs.labelSimilarity()
>>> sim
Out[47]: [(0, 'II')] # group II and label 0 are very similar
On this page
Show Source