py2ls 0.2.4.25__py3-none-any.whl → 0.2.4.27__py3-none-any.whl

py2ls/corr.py

@@ -0,0 +1,475 @@
+# correlations
+import numpy as np
+import pandas as pd
+
+from scipy.interpolate import interp1d
+import statsmodels.api as sm
+from scipy.stats import pearsonr, spearmanr
+# (1) pd.Series(data1).rolling(window=winsize).corr(pd.Series(data2))
+def corr_pd_roll(x,y,window=3):
+    """
+    winsize = 10
+    series_1 = pd.Series(res_spin.spin["freq"])
+    series_2 = pd.Series(res_spin.spin["pk2pk"])
+    # Compute the rolling correlation coefficient
+    rolling_correlation = corr_pd_roll(series_1, series_2,window=winsize)
+    Returns:
+        rolling_correlation
+    """
+    # pd.Series() data1 and data2 
+    series_1 = pd.Series(x)
+    series_2 = pd.Series(y)
+    # Compute the rolling correlation coefficient
+    rolling_correlation = series_1.rolling(window=window).corr(series_2)
+    print(f" corr_pd_roll correlation to check similarity, \nwindow is {window}, cannot be 1")
+    return rolling_correlation
+# Sliding window: a sliding window with incremental updates. This method is computationally 
+# efficient compared to recalculating the correlation coefficient for each window.
+def corr_sliding_window(x, y, window=1):
+    """
+    corr_sliding_window a sliding window with incremental updates. 
+    This method is computationally efficient compared to recalculating 
+    the correlation coefficient for each window.
+
+    series_1 = pd.Series(res_spin.spin["freq"])
+    series_2 = pd.Series(res_spin.spin["pk2pk"])
+    sliding_corr=corr_sliding_window(series_1, series_2, window=3)
+    Returns:
+        sliding_corr: r values
+    """
+    # Initialize variables
+    n = len(x)
+    sum_x = np.sum(x[:window])
+    sum_y = np.sum(y[:window])
+    sum_xy = np.sum(x[:window] * y[:window])
+    sum_x_sq = np.sum(x[:window] ** 2)
+    sum_y_sq = np.sum(y[:window] ** 2)
+
+    # Compute the initial correlation coefficient
+    corr = [
+        (n * sum_xy - sum_x * sum_y)
+        / np.sqrt((n * sum_x_sq - sum_x**2) * (n * sum_y_sq - sum_y**2))
+    ]
+
+    # Update correlation coefficient for each subsequent window
+    for i in range(1, n - window + 1):
+        sum_x += x[i + window - 1] - x[i - 1]
+        sum_y += y[i + window - 1] - y[i - 1]
+        sum_xy += np.dot(x[i : i + window], y[i : i + window]) - np.dot(
+            x[i - 1 : i + window - 1], y[i - 1 : i + window - 1]
+        )
+        sum_x_sq += np.sum(x[i : i + window] ** 2) - np.sum(
+            x[i - 1 : i + window - 1] ** 2
+        )
+        sum_y_sq += np.sum(y[i : i + window] ** 2) - np.sum(
+            y[i - 1 : i + window - 1] ** 2
+        )
+
+        # Compute the correlation coefficient for the current window
+        corr.append(
+            (window * sum_xy - sum_x * sum_y)
+            / np.sqrt(
+                (window * sum_x_sq - sum_x**2)
+                * (window * sum_y_sq - sum_y**2)
+            )
+        )
+
+    return np.array(corr)
+
+
+# Fourier Transform for correlation analysis
+# Compute the cross-power spectral density (CPSD) between the two time series.
+# Compute the power spectral density (PSD) of each time series separately.
+# Divide the CPSD by the square root of the product of the individual PSDs to obtain the cross-correlation function.
+# Apply the inverse Fourier Transform to obtain the correlation coefficient as a function of time.
+def corr_fft(x, y):
+    """
+    corr_fft _summary_
+
+    Args:
+        x (_type_): _description_
+        y (_type_): _description_
+    series_1 = pd.Series(res_spin.spin["freq"])
+    series_2 = pd.Series(res_spin.spin["pk2pk"])
+    r=corr_fft(series_1, series_2)
+    Returns:
+        r: r values
+    """
+    # Compute FFT of each time series
+    fft_x = np.fft.fft(x)
+    fft_y = np.fft.fft(y)
+
+    # Compute cross-power spectral density
+    cpsd = fft_x * np.conj(fft_y)
+
+    # Compute power spectral density of each time series
+    psd_x = np.abs(fft_x) ** 2
+    psd_y = np.abs(fft_y) ** 2
+
+    # Compute cross-correlation function
+    cross_corr = np.fft.ifft(cpsd / np.sqrt(psd_x * psd_y))
+    return cross_corr.real
+
+# Exponentially Weighted Moving Average (EWMA)
+# You can use exponentially weighted moving average to compute the correlation coefficient continuously over time. This method assigns exponentially decreasing weights
+# to the past observations, giving more weight to recent observations. Here's an example of how you can implement it:
+def corr_ewma(x, y, smth=0.1):  # alpha is the smth factor
+    """
+    smth = 0.1  # default
+    # Compute the EWMA correlation coefficient
+    series_1 = pd.Series(res_spin.spin["freq"])
+    series_2 = pd.Series(res_spin.spin["pk2pk"])
+    ewma_correlation = corr_ewma(series_1, series_2, smth=smth)
+
+    Args:
+        x (_type_): data1
+        y (_type_): data2
+        smth (float, optional): alpha is the smth factor. Defaults to 0.1.
+
+    Returns:
+        ewma_correlation: r values
+    """
+    corr = []
+    corr.append(np.corrcoef(x, y)[0, 1])
+    for i in range(1, len(x)):
+        corr.append(
+            smth * np.corrcoef(x[: i + 1], y[: i + 1])[0, 1] + (1 - smth) * corr[i - 1]
+        )
+    return np.array(corr)
+
+# Recursive Formulas
+# where each new value is computed based on the previous one. This method is similar to 
+# rolling window functions but calculates each new value efficiently without re-computing 
+# the entire window.
+def corr_recursive(x, y):
+    corr = []
+    corr.append(np.corrcoef(x[:2], y[:2])[0, 1])
+    for i in range(2, len(x)):
+        corr.append(
+            (i - 1) / i * corr[-1] + 1 / i * np.corrcoef(x[: i + 1], y[: i + 1])[0, 1]
+        )
+    return np.array(corr)
+
+# adaptive or online algorithm
+# One such algorithm is the Online Pearson Correlation Coefficient algorithm, which updates the correlation coefficient as new data points become available without the need for storing or reprocessing past data.
+class ContinuousOnlinePearsonCorrelation:
+    """ 
+    x = pd.Series(res_spin.spin["freq"])
+    y = pd.Series(res_spin.spin["pk2pk"])
+
+    # Initialize ContinuousOnlinePearsonCorrelation
+    continuous_online_corr = ContinuousOnlinePearsonCorrelation()
+    for i, j in zip(x, y):
+        continuous_online_corr.update(i, j)
+
+    print("Continuous correlation coefficients:")
+    print(continuous_online_corr.correlation_values[:10])
+    """
+    def __init__(self):
+        self.n = 0
+        self.mean_x = 0
+        self.mean_y = 0
+        self.m2_x = 0
+        self.m2_y = 0
+        self.cov_xy = 0
+        self.correlation_values = []
+
+    def update(self, x, y):
+        self.n += 1
+        delta_x = x - self.mean_x
+        delta_y = y - self.mean_y
+        self.mean_x += delta_x / self.n
+        self.mean_y += delta_y / self.n
+        delta2_x = x - self.mean_x
+        delta2_y = y - self.mean_y
+        self.m2_x += delta_x * delta2_x
+        self.m2_y += delta_y * delta2_y
+        self.cov_xy += delta_x * delta_y * (self.n - 1) / self.n
+        if self.m2_x > 0 and self.m2_y > 0:
+            correlation = self.cov_xy / (self.m2_x**0.5 * self.m2_y**0.5)
+            self.correlation_values.append(correlation)
+
+
+"""
+# what if the two data series with different sample rate. how to do the correlation?
+
+If the two data series have different sample rates, you can still compute the correlation between them. However, you need to ensure that they are synchronized or resampled to a common time grid before calculating the correlation.
+
+general approach to handle data series with different sample rates:
+
+(1) Resample both data series to a common time grid using interpolation or other resampling techniques.
+
+(2) Compute the correlation between the resampled data series.
+
+example:
+    series_1 = pd.Series(res_spin.spin["freq"])
+    series_2 = pd.Series(res_spin.spin["pk2pk"])
+    series_3 = resample_data(res_spin.spin["freq"], 1000, 12)
+    series_3 = pd.Series(series_3)
+    series_4 = resample_data(res_spin.spin["pk2pk"], 1000, 12)
+    series_4 = pd.Series(series_4)
+    window_size = 10
+    resample_sliding_corr = sliding_window_corr(series_3, series_4, window_size)
+"""
+
+def corr_interp_sliding(
+    x, y, x_timestamps, y_timestamps, window_size
+):
+    """
+    Using interpolation to align timestamps followed by sliding window computation of the correlation coefficient
+
+    Args:
+        x (np.array): _description_
+        y (_type_): _description_
+        x_timestamps (int): _description_
+        y_timestamps (int): _description_
+        window_size (int): sliding window
+    # Example data
+        x = np.random.randn(10000)  # sampled at 1000 Hz
+        y = np.random.randn(120)  # sampled at 12 Hz
+
+        x_timestamps = np.linspace(0, 10, 10000)  # EEG timestamps
+        y_timestamps = np.linspace(0, 10, 120)  # Glucose timestamps
+
+        # Set the window size for sliding window correlation computation
+        window_size = 100  # Adjust according to your needs
+
+        # Compute continuous correlation coefficients using interpolation and sliding window
+        continuous_correlation = corr_interp_sliding(
+            x, y, x_timestamps, y_timestamps, window_size
+        )
+
+        print("Continuous correlation coefficients:")
+        print(continuous_correlation)
+    Returns:
+        continuous_correlation: r value
+    """
+    # Interpolate y data onto x timestamps
+    interp_func = interp1d(y_timestamps, y, kind="linear", fill_value="extrapolate")
+    y_interp = interp_func(x_timestamps)
+
+    # Compute correlation coefficient using sliding window
+    n = len(x)
+    corr_values = []
+
+    for i in range(n - window_size + 1):
+        x_window = x[i : i + window_size]
+        y_window = y_interp[i : i + window_size]
+
+        # Calculate correlation coefficient for the current window
+        correlation = np.corrcoef(x_window, y_window)[0, 1]
+        corr_values.append(correlation)
+
+    return np.array(corr_values)
+
+
+"""
+Autocorrelation is used in various fields and applications, including:
+
+Time Series Analysis: Autocorrelation is fundamental in time series analysis 
+for understanding the structure and patterns in sequential data. It helps identify 
+seasonality, trends, and other repeating patterns within the data.
+
+Modeling and Forecasting: Autocorrelation informs the selection of appropriate models 
+for forecasting future values of a time series. Models such as autoregressive 
+integrated moving average (ARIMA) and seasonal autoregressive integrated moving 
+average (SARIMA) rely on autocorrelation patterns to capture dependencies between 
+observations.
+
+Quality Control: In manufacturing and process control, autocorrelation analysis 
+is used to detect correlations between successive measurements. Deviations from 
+expected autocorrelation patterns can indicate process instability or abnormalities.
+
+Signal Processing: Autocorrelation is used in signal processing for tasks such as 
+speech recognition, audio processing, and seismic analysis to analyze time-domain 
+signals and extract useful information about signal characteristics.
+
+Overall, autocorrelation provides valuable insights into the temporal dependencies
+and behavior of time series data, enabling better understanding, modeling, and prediction 
+of sequential phenomena.
+"""
+def autocorr_np(x, lag=1):
+    """
+    autocorr_np : use np.correlate(x)
+
+    Args:
+        x (_type_): _description_
+        lag (_type_): _description_
+    # Example data
+        data = np.random.randn(100)
+
+        # Compute autocorrelation at lag 1
+        lag_1_autocorr = autocorr_np(data, 1)
+        print("Autocorrelation at lag 1:", lag_1_autocorr)
+    Returns:
+        lag_corr: r value
+    """
+    n = len(x)
+    mean = np.mean(x)
+    var = np.var(x)
+    x = x - mean
+    lag_corr = np.correlate(x, x, mode="full") / (var * n)
+    return lag_corr[n - 1 : n + lag]
+
+def autocorr_pd(data,max_lag=10):
+    """
+    Compute autocorrelation of a 1D numpy array.
+
+    Parameters:
+    data (numpy.ndarray): 1D array containing the data.
+
+    # Example data
+        data_series = np.random.randn(100)
+        autocorr_series = autocorr_pd(data_series)
+        print("Autocorrelation:", autocorr_series)
+    Returns:
+    float: Autocorrelation value.
+    """
+    # Compute mean and centered data
+    mean = np.mean(data)
+    centered_data = data - mean
+    
+    # Compute autocovariance at lag 0
+    auto_covariance_0 = np.mean(centered_data ** 2)
+    
+    # Compute autocorrelation values for a range of lags
+    autocorr_values = np.zeros(max_lag + 1)
+    for lag in range(max_lag + 1):
+        if lag == 0:
+            autocorr_values[lag] = 1.0
+        else:
+            auto_covariance_lag = np.mean(centered_data[:-lag] * centered_data[lag:])
+            autocorr_values[lag] = auto_covariance_lag / auto_covariance_0
+    
+    return autocorr_values
+
+
+def autocorr_statsmodels(data, nlags=1):
+    """
+    Compute autocorrelation of a 1D numpy array using StatsModels.
+
+    Parameters:
+        data (numpy.ndarray): 1D array containing the data.
+        nlags (int): Number of lags for which to compute autocorrelation (default: 1).
+    # Example data
+        data_array = np.random.randn(100)
+        autocorr_array = compute_autocorrelation(data_array, nlags=1)
+        print("Autocorrelation at lag 1:", autocorr_array)
+    Returns:
+        autocorr_array(float): Autocorrelation value at the specified lag.
+    """
+    # Compute autocorrelation using StatsModels
+    autocorr_result = sm.tsa.acf(data, nlags=nlags)
+
+    return autocorr_result
+
+
+"""
+cross-correlation
+
+Cross-correlation is a statistical method used to measure the similarity between two 
+time series by comparing them at different time lags. Unlike autocorrelation, which 
+measures the similarity of a time series with itself at different lags, cross-correlation 
+measures the similarity between two different time series.
+
+Cross-correlation has several applications, including:
+Signal Processing: In signal processing, cross-correlation is used to detect similarities 
+between different signals or to find the time delay between them. It is widely used in 
+fields such as audio processing, radar signal processing, and image processing.
+
+Time Series Analysis: Cross-correlation helps identify relationships and dependencies between
+different time series data. It is used in fields such as economics, finance, and environmental 
+science to analyze the interactions between various variables over time.
+
+Pattern Recognition: Cross-correlation is used in pattern recognition tasks to match and 
+compare patterns in different datasets. It is employed in fields such as speech recognition, 
+pattern matching, and machine vision.
+"""
+
+def cross_corr_np(x, y,mode='same'):
+    """
+    cross_corr_np _summary_
+
+    Args:
+        x (_type_): _description_
+        y (_type_): _description_
+        mode: default 'same', returns the same lengh  "full",  in NumPy, setting the mode parameter to "full" returns the 
+        cross-correlation of x and y at each position of their overlap, with the result 
+        being twice the length of the original sequences minus 1.
+    # Example data
+    x = np.random.randn(100)
+    y = np.random.randn(100) 
+    cross_corr_values = cross_corr_np(x, y)
+    print("Cross-correlation values:", cross_corr_values[:4])
+    Returns:
+        _type_: _description_
+    """
+    n = len(x)
+    mean_x = np.mean(x)
+    mean_y = np.mean(y)
+    x = x - mean_x
+    y = y - mean_y
+    cross_corr_values = np.correlate(x, y, mode=mode) / (np.std(x) * np.std(y) * n)
+    return cross_corr_values
+
+def cross_corr_pd(x, y):
+    """
+    Compute cross-correlation coefficient between two pandas Series.
+    Example:
+    x=np.random.randn(100)
+    y=np.random.randn(100)
+    cross_corr_values = cross_corr_pd(x,y)
+    print("Cross-correlation:", cross_corr_values)
+    Returns:
+    cross_corr_values(float): Cross-correlation coefficient between the two Series.
+    """
+    Series1=pd.Series(x)
+    Series2=pd.Series(y)
+    # Compute cross-correlation using pandas
+    cross_corr_value = Series1.corr(Series2)
+    
+    return cross_corr_value
+
+def cross_corr_scipy(x, y):
+    from scipy.signal import correlate
+    cross_corr_values = correlate(x, y)
+    print("Cross-correlation values:", cross_corr_values[:4])
+    return cross_corr_values
+
+"""Autocorrelation is used in various fields and applications, including:
+
+Time Series Analysis: Autocorrelation is fundamental in time series analysis for understanding the structure and patterns in sequential data. It helps identify seasonality, trends, and other repeating patterns within the data.
+
+Modeling and Forecasting: Autocorrelation informs the selection of appropriate models for forecasting future values of a time series. Models such as autoregressive integrated moving average (ARIMA) and seasonal autoregressive integrated moving average (SARIMA) rely on autocorrelation patterns to capture dependencies between observations.
+
+Quality Control: In manufacturing and process control, autocorrelation analysis is used to detect correlations between successive measurements. Deviations from expected autocorrelation patterns can indicate process instability or abnormalities.
+
+Signal Processing: Autocorrelation is used in signal processing for tasks such as speech recognition, audio processing, and seismic analysis to analyze time-domain signals and extract useful information about signal characteristics.
+
+Overall, autocorrelation provides valuable insights into the temporal dependencies and behavior of time series data, enabling better understanding, modeling, and prediction of sequential phenomena."""
+def autocorr(x, lag):
+    n = len(x)
+    mean = np.mean(x)
+    var = np.var(x)
+    x = x - mean
+    corr = np.correlate(x, x, mode="full") / (var * n)
+    return corr[n - 1 : n + lag]
+
+"""
+General correlation
+e.g., Pearson correlation or Spearman correlation
+"""
+def corr(x, y, method='pearson'):
+    if method.lower() in ['pe','pear','pearson','peson','pearon']:
+        r, p = pearsonr(x, y)
+        print("Pearson correlation coefficient:", r)
+        print("Pearson p-value:", p)
+        return r,p
+    elif method.lower() in ['spear','sp','spea','spearman','speaman']:
+        r, p = spearmanr(x, y)
+        print("Spearman correlation coefficient:", r)
+        print("Spearman p-value:", p)
+        return r,p
+    else:
+        print(f"{method} is not supported, do you mean 'pearson' or 'spearman'")
+        return None, None