updated gridsearch

eval_regularisation.py

@@ -5,29 +5,30 @@
 
 
 def plot_heatmap(df, col, out_dir, title=""):
-    scores = df[col].values
-    scores = np.array(scores).reshape(len(df["C"].unique()), len(df["gamma"].unique()))
-    #plt.figure(figsize=(8, 6))
-    #plt.subplots_adjust(left=.2, right=0.95, bottom=0.15, top=0.95)
-    fig, ax = plt.subplots()
-    im = ax.imshow(scores[::-1, :], interpolation='nearest')
-    # im = ax.imshow(scores, interpolation='nearest',
-    #            norm=MidpointNormalize(vmin=-.2, midpoint=0.5))
-    ax.set_xlabel('gamma')
-    ax.set_ylabel('C')
-    fig.colorbar(im)
-    ax.set_xticks(np.arange(len(df["gamma"].unique())),
-               [np.format_float_scientific(i, 1) for i in df["gamma"].unique()], rotation=45)
-    ax.set_yticks(np.arange(len(df["C"].unique()))[::-1],
-               [np.format_float_scientific(i, ) for i in df["C"].unique()])
-    ax.set_title(f'Regularisation AUC\n{title}')
-    fig.tight_layout()
-    fig.show()
-    out_dir.mkdir(parents=True, exist_ok=True)
-    filename = f"heatmap_{col}_{title}.png".replace(":", "_").replace(" ", "_")
-    filepath = out_dir / filename
-    print(filepath)
-    fig.savefig(filepath)
+    df = df.fillna("linear")
+    for g in df["gamma"].unique():
+        df_ = df[df["gamma"] == g]
+        scores = df_[col].values
+        scores = np.array(scores).reshape(len(df_["C"].unique()), len(df_["gamma"].unique()))
+        #plt.figure(figsize=(8, 6))
+        #plt.subplots_adjust(left=.2, right=0.95, bottom=0.15, top=0.95)
+        fig, ax = plt.subplots()
+        im = ax.imshow(scores[::-1, :], interpolation='nearest')
+        # im = ax.imshow(scores, interpolation='nearest',
+        #            norm=MidpointNormalize(vmin=-.2, midpoint=0.5))
+        # ax.set_xlabel('gamma')
+        ax.set_ylabel('C')
+        fig.colorbar(im)
+        ax.set_xticks(np.arange(len(df_["gamma"].unique())), [i for i in df_["gamma"].unique()], rotation=45)
+        ax.set_yticks(np.arange(len(df_["C"].unique()))[::-1],
+                   [np.format_float_scientific(i, ) for i in df_["C"].unique()])
+        ax.set_title(f'Regularisation AUC\n{title}')
+        fig.tight_layout()
+        fig.show()
+        out_dir.mkdir(parents=True, exist_ok=True)
+        filename = f"heatmap_{g}_{col}_{title}.png".replace(":", "_").replace(" ", "_")
+        filepath = out_dir / filename
+        print(filepath)
 
 
 def plot_fig(df, col, out_dir, title=""):

main.py

@@ -10,13 +10,13 @@ def main(
     exp_temporal: bool = False,
     exp_cross_farm: bool = False,
     weather_exp: bool = False,
-    regularisation_exp: bool = False,
+    regularisation_exp: bool = True,
     output_dir: Path = Path("output"),
     delmas_dir_mrnn: Path = Path("datasets/delmas_dataset4_mrnn_7day"),
     cedara_dir_mrnn: Path = Path("datasets/cedara_datasetmrnn7_23"),
     n_job: int = 5,
-    enable_regularisation: bool = False,
-    export_hpc_string: bool = False,
+    enable_regularisation: bool = True,
+    export_hpc_string: bool = True,
     plot_2d_space: bool = False
 ):
     """Thesis script runs all key experiments for data exploration chapter
@@ -33,8 +33,8 @@ def main(
             for steps in steps_list:
                 slug = "_".join(steps)
                 for clf in ["rbf"]:
-                    for i_day in [6]:
-                        for a_day in [7]:
+                    for i_day in [1,2,3,4,5,6]:
+                        for a_day in [1,2,3,4,5,6,7]:
                             if i_day >= a_day:
                                 continue
                             for cv in ["RepeatedKFold"]:
@@ -63,7 +63,7 @@ def main(
                                         ],
                                         study_id=farm_id,
                                         export_fig_as_pdf=False,
-                                        plot_2d_space=True,
+                                        plot_2d_space=plot_2d_space,
                                         pre_visu=False,
                                         export_hpc_string=export_hpc_string,
                                         skip=False,

model/svm.py

@@ -916,9 +916,9 @@ def cross_validate_svm_fast(
         if kernel in ["linear", "rbf"]:
             if enable_regularisation:
                 print("Grid search...")
-                svc = SVC(kernel=kernel, probability=True)
-                parameters["kernel"] = [kernel]
-                clf = GridSearchCV(svc, parameters, return_train_score=True, cv=10, scoring='roc_auc', n_jobs=-1)
+                svc = SVC(probability=True)
+                # parameters["kernel"] = [kernel]
+                clf = GridSearchCV(svc, parameters, return_train_score=True, cv=5, scoring='roc_auc', n_jobs=-1)
             else:
                 if C is not None and gamma is not None:
                     clf = SVC(kernel=kernel, probability=True, C=C, gamma=gamma)

var.py

@@ -42,11 +42,15 @@
 #     ],
 # }
 
-parameters = {
-    "C": [1e-100, 1.7782794100389228e-75, 3.1622776601683795e-50, 5.623413251903491e-25, 10.0],
-    "gamma": [1e-100, 1e-25, 1e+50, 1e+125, 1e+200]
-}
+parameters = [
+  {'C': [0.000001, 0.00001, 0.0001, 0.001, 0.01, 0.1, 1, 10, 100, 1000], 'kernel': ['linear']},
+  {'C': [0.000001, 0.00001, 0.0001, 0.001, 0.01, 0.1, 1, 10, 100, 1000], 'gamma': ['scale'], 'kernel': ['rbf']},
+ ]
 
+# parameters = [
+#   {'C': [0.0001, 100], 'kernel': ['linear']},
+#   {'C': [0.0001, 100], 'gamma': ['scale'], 'kernel': ['rbf']},
+#  ]
 
 transponders_delmas = [
     "40101310316",