Merge pull request #39 from martinvonk/dev

Update main to v0.4.2

doc/examples/example01_indices.ipynb

@@ -27,7 +27,7 @@
      "text": [
       "Versions\n",
       "python: 3.10.12\n",
-      "spei: 0.4.0\n",
+      "spei: 0.4.1\n",
       "numpy: 1.26.4\n",
       "scipy: 1.12.0\n",
       "matplotlib: 3.8.3\n",
@@ -131,7 +131,7 @@
    ],
    "source": [
     "f = 90  # days\n",
-    "series = prec.rolling(f\"{f}D\", min_periods=f).sum().dropna()\n",
+    "series = prec.rolling(f, min_periods=f).sum().dropna()\n",
     "series"
    ]
   },

src/spei/_version.py

@@ -1,7 +1,7 @@
 from importlib import metadata
 from platform import python_version
 
-__version__ = "0.4.1"
+__version__ = "0.4.2"
 
 
 def show_versions() -> str:

src/spei/dist.py

@@ -102,7 +102,7 @@ def cdf(self) -> Series:
         """Compute cumulative density function of a Scipy Continuous Distribution"""
         if self.pars is not None:
             cdf = self.dist.cdf(
-                self.data.values, self.pars, loc=self.loc, scale=self.scale
+                self.data.values, *self.pars, loc=self.loc, scale=self.scale
             )
         else:
             cdf = self.dist.cdf(self.data.values, loc=self.loc, scale=self.scale)
@@ -117,7 +117,7 @@ def pdf(self) -> Series:
         data_pdf = self.data.sort_values()
         if self.pars is not None:
             pdf = self.dist.pdf(
-                data_pdf.values, self.pars, loc=self.loc, scale=self.scale
+                data_pdf.values, *self.pars, loc=self.loc, scale=self.scale
             )
         else:
             pdf = self.dist.pdf(data_pdf.values, loc=self.loc, scale=self.scale)

src/spei/si.py

@@ -30,6 +30,10 @@ def sgi(
     series: pandas.Series
         Pandas time series of the groundwater levels. Time series index
         should be a pandas DatetimeIndex.
+    fit_freq : str, optional, default=None
+        Frequency for fitting the distribution. Default is None in which case
+        the frequency of the series is inferred. If this fails a monthly
+        frequency is used.
 
     Returns
     -------
@@ -75,9 +79,26 @@ def spi(
         However, for the SPI generally the Gamma probability density
         function is recommended. Other appropriate choices could be the
         lognormal, log-logistic (fisk) or PearsonIII distribution.
-    prob_zero: bool
-        Option to correct the distribution if x=0 is not in probability
-        density function. E.g. the case with the Gamma distriubtion.
+    timescale : int, optional, default=0
+        Size of the moving window over which the series is summed. If zero, no
+        summation is performed over the time series. If the time series
+        frequency is daily, then one would provide timescale=30 for SI1,
+        timescale=90 for SI3, timescale=180 for SI6 etc.
+    fit_freq : str, optional, default=None
+        Frequency for fitting the distribution. Default is None in which case
+        the frequency of the series is inferred. If this fails a monthly
+        frequency is used.
+    fit_window : int, optional, default=0
+        Window size for fitting data in fit_freq frequency's unit. Default is
+        zero in which case only data within the fit_freq is considered. If
+        larger than zero data data within the window is used to fit the
+        distribution for the series. fit_window must be a odd number larger
+        than 3 when used.
+    prob_zero : bool, default=True
+        Option to correct the distribution if x=0 is not in probability density
+        function. E.g. the case with the Gamma distriubtion. If True, the
+        probability of zero values in the series is calculated by the
+        occurence.
 
     Returns
     -------
@@ -124,6 +145,24 @@ def spei(
         However, for the SPEI generally the log-logistic (fisk) probability
         density function is recommended. Other appropriate choices could be
         the lognormal or PearsonIII distribution.
+    timescale : int, optional, default=0
+        Size of the moving window over which the series is summed. If zero, no
+        summation is performed over the time series. If the time series
+        frequency is daily, then one would provide timescale=30 for SI1,
+        timescale=90 for SI3, timescale=180 for SI6 etc.
+    fit_freq : str, optional, default=None
+        Frequency for fitting the distribution. Default is None in which case
+        the frequency of the series is inferred. If this fails a monthly
+        frequency is used.
+    fit_window : int, optional, default=0
+        Window size for fitting data in fit_freq frequency's unit. Default is
+        zero in which case only data within the fit_freq is considered. If
+        larger than zero data data within the window is used to fit the
+        distribution for the series. fit_window must be a odd number larger
+        than 3 when used.
+    prob_zero : bool, default=False
+        Flag indicating whether the probability of zero values in the series is
+        calculated by the occurence.
 
     Returns
     -------
@@ -170,6 +209,24 @@ def ssfi(
         However, for the SSFI generally the gamma probability density
         function is recommended. Other appropriate choices could be the
         normal, lognormal, pearsonIII, GEV or  Gen-Logistic distribution.
+    timescale : int, optional, default=0
+        Size of the moving window over which the series is summed. If zero, no
+        summation is performed over the time series. If the time series
+        frequency is daily, then one would provide timescale=30 for SI1,
+        timescale=90 for SI3, timescale=180 for SI6 etc.
+    fit_freq : str, optional, default=None
+        Frequency for fitting the distribution. Default is None in which case
+        the frequency of the series is inferred. If this fails a monthly
+        frequency is used.
+    fit_window : int, optional, default=0
+        Window size for fitting data in fit_freq frequency's unit. Default is
+        zero in which case only data within the fit_freq is considered. If
+        larger than zero data data within the window is used to fit the
+        distribution for the series. fit_window must be a odd number larger
+        than 3 when used.
+    prob_zero : bool, default=False
+        Flag indicating whether the probability of zero values in the series is
+        calculated by the occurence.
 
     Returns
     -------
@@ -207,9 +264,10 @@ class SI:
     dist : ContinuousDist
         The SciPy continuous distribution associated with the data.
     timescale : int, optional, default=0
-        Rolling window timescale in days over which the series is summed. For
-        SI1 the user would provide timescale=30, for SI3: timescale=90, SI6:
-        timescale=180 etc.
+        Size of the moving window over which the series is summed. If zero, no
+        summation is performed over the time series. If the time series
+        frequency is daily, then one would provide timescale=30 for SI1,
+        timescale=90 for SI3, timescale=180 for SI6 etc.
     fit_freq : str, optional, default=None
         Frequency for fitting the distribution. Default is None in which case
         the frequency of the series is inferred. If this fails a monthly
@@ -258,7 +316,7 @@ def __post_init__(self) -> None:
 
         if self.timescale > 0:
             self.series = (
-                self.series.rolling(f"{self.timescale}D", min_periods=self.timescale)
+                self.series.rolling(self.timescale, min_periods=self.timescale)
                 .sum()
                 .dropna()
                 .copy()

src/spei/utils.py

@@ -64,7 +64,7 @@ def infer_frequency(index: Union[Index, DatetimeIndex]) -> str:
         logging.info(
             "Could not infer frequency from index, using monthly frequency instead"
         )
-        inf_freq = "ME" if pd_version >= "2.1.0" else "M"
+        inf_freq = "ME" if pd_version >= "2.2.0" else "M"
     else:
         logging.info(f"Inferred frequency '{inf_freq}' from index")
 
@@ -81,7 +81,7 @@ def group_yearly_df(series: Series) -> DataFrame:
     """
     strfstr: str = "%m-%d %H:%M:%S"
     grs = {}
-    freq = "YE" if pd_version >= "2.1.0" else "Y"
+    freq = "YE" if pd_version >= "2.2.0" else "Y"
     for year_timestamp, gry in series.groupby(Grouper(freq=freq)):
         index = validate_index(gry.index)
         gry.index = to_datetime(