finds.recipes.filters
Filtering functions
Copyright 2022, Terence Lim
MIT License
- finds.recipes.filters.fft_align(X: ndarray) Tuple[source]
Find best alignment and cross-correlation of all pairs of columns
- Parameters:
X – array with time series in columns
- Returns:
Max cross-correlations, best lags, indices of all pairs of columns
Notes:
Apply convolution theorem to compute cross-correlations at all lags
For each pair of series, the lag with largest correlation is assumed to be the displacement which aligns the presentation of the two series
Examples:
>>> fft_align(np.hstack((X[:-1], X[1:])))
- finds.recipes.filters.fft_correlation(X: ndarray, Y: ndarray | None) Series[source]
Compute cross-correlations of two series, using Convolution Theorem
- Parameters:
X – series of observations
Y – series of observations
- Returns:
Series of cross-correlation values at every displacement lag
Notes: - Cross-correlation[n] = sum_m^N f[m] g[n + m] - equals Convolution (f * g)[n] = sum_m^N f[m] g[n - m]
Examples
>>> statsmodels.tsa.stattools.acf(X) >>> fft_correlate(X, X)
- finds.recipes.filters.fft_neweywest(X: ndarray) List[source]
Compute Newey-West weighted cross-correlation of all pairs of columns
- Parameters:
X – array with series in columns
- Returns:
List of Newey-west weighted cross-correlations
Notes:
First apply convolution theorem to compute all cross-autocorrelations,
Then for each pair of series, compute Newey-west weighted correlation
- finds.recipes.filters.fractile_split(values: Iterable, pct: Iterable, keys: Iterable | None = None, ascending: bool = False) List[int][source]
Sort and assign values into fractiles
- Parameters:
values – input array to assign to fractiles
pct – list of percentiles between 0 and 100
keys – key values to determine breakpoints, use values if None
ascending – if True, assign to fractiles in ascending order
- Returns:
list of fractile assignments (starting at 1 with smallest values)
- finds.recipes.filters.is_outlier(x: Any, method: str = 'iq10', fences: bool = False) array[source]
Test if elements of x are column-wise outliers
- Parameters:
x – Input array to test element-wise
method – method to filter, in {‘iq{D}’, ‘tukey’, ‘farout’}
fences – If True, return (low, high) values of fence
- Returns:
boolean indicator array if element is column-wise outlier or compute fences
Notes: - ‘iq{D}’ - screen by iq range median +/- [D times (Q3-Q1)] - ‘tukey’ - [Q1 - 1.5(Q3-Q1), Q3 + 1.5(Q3-Q1)] - ‘farout’ - tukey with 3IQ
- finds.recipes.filters.remove_outliers(X: DataFrame, method: str = 'iq10', verbose: bool = False) DataFrame[source]
Set column-wise outliers to np.nan
- Parameters:
X – Input array to test element-wise
method – method to filter outliers, in {‘iq{D}’, ‘tukey’, ‘farout’}
- Returns:
DataFrame with outliers set to NaN
Notes: - ‘iq{D}’ - within [median +/- D times (Q3-Q1)] - ‘tukey’ - within [Q1 - 1.5(Q3-Q1), Q3 + 1.5(Q3-Q1)] - ‘farout’ - within [Q1 - 3(Q3-Q1), Q3 + 3(Q3-Q1)]
- finds.recipes.filters.weighted_average(df: DataFrame, weights: str = '') Series[source]
Weighted means of data frame
- Parameters:
df – DataFrame containing values, and optional weights, in columns
weights – Column name to use as weights
- Returns:
Series of weighted means
Notes: - ignores NaN’s using numpy.ma