Python scipy.stats.mstats.mode() Examples

def test_mode(self):
        a1 = [0,0,0,1,1,1,2,3,3,3,3,4,5,6,7]
        a2 = np.reshape(a1, (3,5))
        a3 = np.array([1,2,3,4,5,6])
        a4 = np.reshape(a3, (3,2))
        ma1 = ma.masked_where(ma.array(a1) > 2, a1)
        ma2 = ma.masked_where(a2 > 2, a2)
        ma3 = ma.masked_where(a3 < 2, a3)
        ma4 = ma.masked_where(ma.array(a4) < 2, a4)
        assert_equal(mstats.mode(a1, axis=None), (3,4))
        assert_equal(mstats.mode(a1, axis=0), (3,4))
        assert_equal(mstats.mode(ma1, axis=None), (0,3))
        assert_equal(mstats.mode(a2, axis=None), (3,4))
        assert_equal(mstats.mode(ma2, axis=None), (0,3))
        assert_equal(mstats.mode(a3, axis=None), (1,1))
        assert_equal(mstats.mode(ma3, axis=None), (2,1))
        assert_equal(mstats.mode(a2, axis=0), ([[0,0,0,1,1]], [[1,1,1,1,1]]))
        assert_equal(mstats.mode(ma2, axis=0), ([[0,0,0,1,1]], [[1,1,1,1,1]]))
        assert_equal(mstats.mode(a2, axis=-1), ([[0],[3],[3]], [[3],[3],[1]]))
        assert_equal(mstats.mode(ma2, axis=-1), ([[0],[1],[0]], [[3],[1],[0]]))
        assert_equal(mstats.mode(ma4, axis=0), ([[3,2]], [[1,1]]))
        assert_equal(mstats.mode(ma4, axis=-1), ([[2],[3],[5]], [[1],[1],[1]])) 

def majority_voting_rule(votes):
    """Applies the majority voting rule to the estimated votes.

    Parameters
    ----------
    votes : array of shape (n_samples, n_classifiers),
        The votes obtained by each classifier for each sample.

    Returns
    -------
    predicted_label : array of shape (n_samples)
        The label of each query sample predicted using the majority voting rule
    """
    # Omitting nan value in the predictions as they comes from removed
    # classifiers
    return mode(votes, axis=1)[0][:, 0] 

def test_mode(self):
        a1 = [0,0,0,1,1,1,2,3,3,3,3,4,5,6,7]
        a2 = np.reshape(a1, (3,5))
        a3 = np.array([1,2,3,4,5,6])
        a4 = np.reshape(a3, (3,2))
        ma1 = ma.masked_where(ma.array(a1) > 2, a1)
        ma2 = ma.masked_where(a2 > 2, a2)
        ma3 = ma.masked_where(a3 < 2, a3)
        ma4 = ma.masked_where(ma.array(a4) < 2, a4)
        assert_equal(mstats.mode(a1, axis=None), (3,4))
        assert_equal(mstats.mode(a1, axis=0), (3,4))
        assert_equal(mstats.mode(ma1, axis=None), (0,3))
        assert_equal(mstats.mode(a2, axis=None), (3,4))
        assert_equal(mstats.mode(ma2, axis=None), (0,3))
        assert_equal(mstats.mode(a3, axis=None), (1,1))
        assert_equal(mstats.mode(ma3, axis=None), (2,1))
        assert_equal(mstats.mode(a2, axis=0), ([[0,0,0,1,1]], [[1,1,1,1,1]]))
        assert_equal(mstats.mode(ma2, axis=0), ([[0,0,0,1,1]], [[1,1,1,1,1]]))
        assert_equal(mstats.mode(a2, axis=-1), ([[0],[3],[3]], [[3],[3],[1]]))
        assert_equal(mstats.mode(ma2, axis=-1), ([[0],[1],[0]], [[3],[1],[0]]))
        assert_equal(mstats.mode(ma4, axis=0), ([[3,2]], [[1,1]]))
        assert_equal(mstats.mode(ma4, axis=-1), ([[2],[3],[5]], [[1],[1],[1]]))

        a1_res = mstats.mode(a1, axis=None)

        # test for namedtuple attributes
        attributes = ('mode', 'count')
        check_named_results(a1_res, attributes, ma=True) 

def test_mode_modifies_input(self):
        # regression test for gh-6428: mode(..., axis=None) may not modify
        # the input array
        im = np.zeros((100, 100))
        im[:50, :] += 1
        im[:, :50] += 1
        cp = im.copy()
        a = mstats.mode(im, None)
        assert_equal(im, cp) 

def median(x: pd.Series, w: Union[Window, int] = Window(None, 0)) -> pd.Series:
    """
    Median value of series over given window

    :param x: series: timeseries
    :param w: Window or int: size of window and ramp up to use. e.g. Window(22, 10) where 22 is the window size
              and 10 the ramp up value. Window size defaults to length of series.
    :return: timeseries of median value

    **Usage**

    Computes the `median <https://en.wikipedia.org/wiki/Median>`_ value over a given window. For each window, this
    function will return the middle value when all elements in the window are sorted. If the number of observations in
    the window is even, will return the average of the middle two values. If the window size is greater than the
    available data, will return median of available values:

    :math:`d = \\frac{w-1}{2}`

    :math:`R_t = \\frac{X_{\lfloor t-d \\rfloor} + X_{\lceil t-d \\rceil}}{2}`

    where :math:`w` is the size of the rolling window. If window is not provided, computes median over the full series

    **Examples**

    Generate price series and compute median over :math:`22` observations

    >>> prices = generate_series(100)
    >>> median(prices, 22)

    **See also**

    :func:`mean` :func:`mode`
    """
    w = normalize_window(x, w)
    assert x.index.is_monotonic_increasing, "series index is monotonic increasing"
    if isinstance(w.w, pd.DateOffset):
        values = [x.loc[(x.index > idx - w.w) & (x.index <= idx)].median() for idx in x.index]
        return apply_ramp(pd.Series(values, index=x.index, dtype=np.dtype(float)), w)
    else:
        return apply_ramp(x.rolling(w.w, 0).median(), w) 

def mode(x: pd.Series, w: Union[Window, int] = Window(None, 0)) -> pd.Series:
    """
    Most common value in series over given window

    :param x: series: timeseries
    :param w: Window or int: size of window and ramp up to use. e.g. Window(22, 10) where 22 is the window size
              and 10 the ramp up value. Window size defaults to length of series.
    :return: timeseries of mode value

    **Usage**

    Computes the `mode <https://en.wikipedia.org/wiki/Mode_(statistics)>`_ over a given window. For each window, this
    function will return the most common value of all elements in the window. If there are multiple values with the same
    frequency of occurrence, will return the smallest value.

    If window is not provided, computes mode over the full series.

    **Examples**

    Generate price series and compute mode over :math:`22` observations

    >>> prices = generate_series(100)
    >>> mode(prices, 22)

    **See also**

    :func:`mean` :func:`median`
    """
    w = normalize_window(x, w)
    assert x.index.is_monotonic_increasing, "series index is monotonic increasing"
    if isinstance(w.w, pd.DateOffset):
        values = [stats.mode(x.loc[(x.index > idx - w.w) & (x.index <= idx)]).mode[0] for idx in x.index]
        return apply_ramp(pd.Series(values, index=x.index, dtype=np.dtype(float)), w)
    else:
        return apply_ramp(x.rolling(w.w, 0).apply(lambda y: stats.mode(y).mode, raw=True), w) 

def mean(x: Union[pd.Series, List[pd.Series]], w: Union[Window, int] = Window(None, 0)) -> pd.Series:
    """
    Arithmetic mean of series over given window

    :param x: series: a timeseries or an array of timeseries
    :param w: Window or int: size of window and ramp up to use. e.g. Window(22, 10) where 22 is the window size
              and 10 the ramp up value. Window size defaults to length of series.
    :return: timeseries of mean value

    **Usage**

    Calculates `arithmetic mean <https://en.wikipedia.org/wiki/Arithmetic_mean>`_ of the series over a rolling window

    If a timeseries is provided:

    :math:`R_t = \\frac{\sum_{i=t-w+1}^{t} X_i}{N}`

    where :math:`N` is the number of observations in each rolling window, :math:`w`.

    If an array of timeseries is provided:

    :math:`R_t = \\frac{\sum_{i=t-w+1}^{t} {\sum_{j=1}^{n}} X_{ij}}{N}`

    where :math:`n` is the number of series, and :math:`N` is the number of observations in each rolling window,
    :math:`w`.

    If window is not provided, computes rolling mean over the full series. If the window size is greater than the
    available data, will return mean of available values.

    **Examples**

    Generate price series and compute mean over :math:`22` observations

    >>> prices = generate_series(100)
    >>> mean(prices, 22)

    **See also**

    :func:`median` :func:`mode`

    """
    if isinstance(x, list):
        x = pd.concat(x, axis=1)
    w = normalize_window(x, w)
    assert x.index.is_monotonic_increasing, "series index is monotonic increasing"
    if isinstance(w.w, pd.DateOffset):
        values = [np.nanmean(x.loc[(x.index > idx - w.w) & (x.index <= idx)]) for idx in x.index]
    else:
        values = [np.nanmean(x.iloc[max(idx - w.w + 1, 0): idx + 1]) for idx in range(0, len(x))]
    return apply_ramp(pd.Series(values, index=x.index, dtype=np.dtype(float)), w)