Merge branch 'asymptote_transform' of https://github.com/cvxgrp/spcqe …

…into asymptote_transform merging new changes
cvxgrp · Jan 10, 2025 · 1c5d549 · 1c5d549
2 parents 8130ed4 + 7de7f95
commit 1c5d549
Show file tree

Hide file tree

Showing 3 changed files with 131 additions and 105 deletions.
diff --git a/notebooks/Aramis/extrapolation_demo.ipynb b/notebooks/Aramis/extrapolation_demo.ipynb
diff --git a/src/spcqe/extrapolate_asymptotic.py b/src/spcqe/extrapolate_asymptotic.py
@@ -16,7 +16,7 @@
 from scipy.interpolate import interp1d
 
 dist = sps.norm()
-XSOLAR = 0
+XSOLAR = -0.1
 YSOLAR = dist.ppf(0.99999)
 
 def get_asymptote_parameters_out(x0, _y0, x1, _y1, yasympt):
@@ -182,14 +182,13 @@ def plot_tails(ax, sig, quantiles, fit_quantiles, transf, method, key, index, ex
         elif method == 'linear':
             extrap_ys = linear_interp(extrap_values)
     sig_values = np.linspace(sig_m, sig_M, 1000)
-    ax.scatter(fit_quantiles[index], dist.ppf(quantiles), color='C3', marker='+', s=40, label='quantiles setpoints')
     if h_per_day is not None:
         idxs_days = np.arange(-n_days, n_days+1, 1) * h_per_day + index
         idxs_days = np.array([i for i in idxs_days if i != index])
         idxs_hours = np.arange(-n_hours, n_hours+1, 1) + index
         idxs_hours = np.array([i for i in idxs_hours if i != index])
-        ax.scatter(sig[idxs_days], transf[idxs_days], color='green', s=20, label='nearby days')
-        ax.scatter(sig[idxs_hours], transf[idxs_hours], color='orange', s=20, label='nearby hours')
+        ax.scatter(sig[idxs_days], transf[idxs_days], color='green', s=30, label='nearby days')
+        ax.scatter(sig[idxs_hours], transf[idxs_hours], color='orange', s=30, label='nearby hours')
     ax.axvline(sig[index], color='C3', linestyle='--', label='value to transform')
     ax.plot(sig_values, linear_interp(sig_values), label='linear transform')
     ax.plot(
@@ -199,6 +198,7 @@ def plot_tails(ax, sig, quantiles, fit_quantiles, transf, method, key, index, ex
         color='black',
         label='extrapolation transform'
         )
+    ax.scatter(fit_quantiles[index], dist.ppf(quantiles), color='C3', marker='+', s=150, label='quantiles setpoints')
     ax.scatter(sig[index], transf[index], color='black', marker='x', label='transformed value')
     ax.set_title(f'Transfer function from dilated signal to normal distribution - {key} tail')
     ax.legend()

diff --git a/src/spcqe/quantiles.py b/src/spcqe/quantiles.py
@@ -352,40 +352,6 @@ def inverse_fit_extrapolate(self, X, Z, new_quantiles):
 
         return Z
 
-#     TODO: not used anymore. Should be removed ?
-#     def x_expand(self, xi, tix):
-#        """
-#        
-#        """
-#        xin = np.atleast_1d(xi)
-#        h1 = np.ones_like(xin)
-#        h2 = xin
-#        h3up = [np.clip(xin - kn, 0, np.inf) for kn in self.fit_quantiles[tix][1:-1]]
-#        basis = np.r_[[h1, h2] + h3up].T
-#     return basis
-
-#    TODO: not used anymore. Should be removed ?
-#    def score(self, X, y=None):
-#        data = np.asarray(X)
-#        if len(data) != self.length and y is None:
-#            raise ValueError(
-#                "If not transforming the original fit data set, a time index must be passed as y"
-#            )
-#        # get correct basis matrix and quantile estimates for time period of prediction
-#        if y is not None:
-#            new_quantiles = self.predict(y)
-#        else:
-#            new_quantiles = self.fit_quantiles
-#        data = data[:, np.newaxis]
-#        q = self.quantiles[np.newaxis, :]
-#        score = np.sum(
-#            np.trapz(
-#                0.5 * np.abs(data - new_quantiles) + (q - 0.5) * (data - new_quantiles),
-#                x=self.quantiles,
-#            )
-#        )
-#        return score
-
     def extend_basis(self, t):
         """
         Extend the basis matrix to a new time index.
@@ -398,11 +364,12 @@ def extend_basis(self, t):
                 self.num_harmonics,
                 t,
                 self.periods,
+                standing_wave=self.standing_wave,
+                trend=self.trend,
                 max_cross_k=self.max_cross_k,
                 custom_basis=self.custom_basis,
             )
             return new_basis
-            # TODO: check if this is the correct way to extend the basis
         else:
             raise NotImplementedError(
                 "Extending the basis to time before the training data not currently supported"