Tune Weibull priors

camirmas · Apr 14, 2023 · aa95ee8 · aa95ee8
1 parent 5114602
commit aa95ee8
Show file tree

Hide file tree

Showing 6 changed files with 594 additions and 550 deletions.
diff --git a/REStats/models/iec_power_curve.py b/REStats/models/iec_power_curve.py
@@ -20,7 +20,7 @@ def calc_iec_power_curve(data: pd.DataFrame, bin_size: float = 0.5) -> pd.DataFr
         raise ValueError("Input DataFrame must contain 'wind_speed' and 'power' columns.")
 
     min_ws = np.floor(np.min(data.wind_speed))
-    max_ws = np.ceil(np.max(data.wind_speed) + bin_size)
+    max_ws = np.ceil(np.max(data.wind_speed)) + bin_size
     wt_bins = np.arange(min_ws, max_ws, bin_size)
     wt_groups = data.groupby(pd.cut(data.wind_speed, wt_bins))
     wt_iec = wt_groups.mean()

diff --git a/REStats/models/weibull.py b/REStats/models/weibull.py
@@ -19,8 +19,10 @@ def weibull_model(data):
     Returns:
         None
     """
-    shape = pyro.sample("shape", dist.Gamma(2, 0.5))
-    scale = pyro.sample("scale", dist.Gamma(1, 0.5))
+    mu = np.log(data.mean())
+
+    shape = pyro.sample("shape", dist.Gamma(2, 1))
+    scale = pyro.sample("scale", dist.LogNormal(mu, 0.5))
 
     with pyro.plate("data", len(data)):
         pyro.sample("obs", dist.Weibull(scale, shape), obs=data)
@@ -94,7 +96,7 @@ def plot_prior_samples(idata_wb):
     shapes = idata_wb.prior.shape[0, :10]
     scales = idata_wb.prior.scale[0, :10]
 
-    x = np.linspace(0, 5, 500)
+    x = np.linspace(0, 12, 500)
 
     def weib(x, scale, shape):
         return (shape / scale) * (x / scale)**(shape - 1) * np.exp(-(x / scale)**shape)

diff --git a/REStats/notebooks/forecast.ipynb b/REStats/notebooks/forecast.ipynb
diff --git a/REStats/notebooks/power_curve.ipynb b/REStats/notebooks/power_curve.ipynb
diff --git a/REStats/notebooks/sim.ipynb b/REStats/notebooks/sim.ipynb
diff --git a/tests/models/test_iec_power_curve.py b/tests/models/test_iec_power_curve.py
@@ -18,7 +18,7 @@ def test_calc_iec_power_curve():
     output.reset_index(drop=True, inplace=True)
 
     # Prepare the expected output
-    wt_bins = np.arange(np.min(wind_speed), np.max(wind_speed), bin_size)
+    wt_bins = np.arange(np.min(wind_speed), np.ceil(np.max(wind_speed)) + bin_size, bin_size)
     wt_groups = data.groupby(pd.cut(data.wind_speed, wt_bins))
     expected_output = wt_groups.mean().reset_index(drop=True)