1c PM - alpha correction

eXascaleInfolab · Jan 8, 2025 · dbe9131 · dbe9131
1 parent b8792a2
commit dbe9131
Show file tree

Hide file tree

Showing 32 changed files with 243 additions and 93 deletions.
diff --git a/.idea/workspace.xml b/.idea/workspace.xml
diff --git a/imputegap/recovery/__pycache__/benchmark.cpython-312.pyc b/imputegap/recovery/__pycache__/benchmark.cpython-312.pyc
diff --git a/imputegap/recovery/benchmark.py b/imputegap/recovery/benchmark.py
@@ -203,7 +203,7 @@ def generate_heatmap(self, scores_list, algos, sets, save_dir="./reports", displ
         plt.savefig(filepath, dpi=300, bbox_inches='tight')  # Save in HD with tight layout
 
         # Show the plot
-        if display :
+        if display:
             plt.tight_layout()
             plt.show()
             plt.close()
@@ -692,6 +692,6 @@ def eval(self, algorithms=["cdrec"], datasets=["eeg-alcohol"], patterns=["mcar"]
                 print("============================================================================\n\n\n\n\n\n")
 
         scores_list, algos, sets = self.avg_results(*run_storage)
-        _ = Benchmark().generate_heatmap(scores_list, algos, sets, save_dir=save_dir)
+        _ = Benchmark().generate_heatmap(scores_list, algos, sets, save_dir=save_dir, display=False)
 
         return run_storage, scores_list
diff --git a/imputegap/recovery/explainer.py b/imputegap/recovery/explainer.py
@@ -151,20 +151,17 @@ def print(shap_values, shap_details=None):
         None
         """
 
-        if shap_details is not None:
-            print("\n\nx_data (with", len(shap_details), "elements) : ")
-            for i, (input, _) in enumerate(shap_details):
-                print("\tFEATURES VALUES", i, "(", len(input), ") : ", *input)
-
-            print("\ny_data (with", len(shap_details), "elements) : ")
-
-            for i, (_, output) in enumerate(shap_details):
-                print(f"\tRMSE SERIES {i:<5} : {output:<15}")
+        #if shap_details is not None:
+            #print("\n\nx_data (with", len(shap_details), "elements) : ")
+            #for i, (input, _) in enumerate(shap_details):
+            #    print("\tFEATURES VALUES", i, "(", len(input), ") : ", *input)
+            #print("\ny_data (with", len(shap_details), "elements) : ")
+            #for i, (_, output) in enumerate(shap_details):
+            #    print(f"\tRMSE SERIES {i:<5} : {output:<15}")
 
         print("\n\nSHAP Results details : ")
         for (x, algo, rate, description, feature, category, mean_features) in shap_values:
-            print(
-                f"\tFeature : {x:<5} {algo:<10} with a score of {rate:<10} {category:<18} {description:<75} {feature}\n")
+            print(f"\tFeature : {x:<5} {algo:<10} with a score of {rate:<10} {category:<18} {description:<75} {feature}\n")
 
     def convert_results(tmp, file, algo, descriptions, features, categories, mean_features, to_save):
         """
@@ -243,14 +240,14 @@ def execute_shap_model(x_dataset, x_information, y_dataset, file, algorithm, spl
             Results of the SHAP explainer model.
         """
 
-        print("\n\nInitilization of the SHAP model with ", np.array(x_information).shape)
+        print("\n\nInitialization of the SHAP model with dimension", np.array(x_information).shape)
 
         path_file = "./assets/shap/"
         if not os.path.exists(path_file):
             path_file = "./imputegap" + path_file[1:]
 
         x_features, x_categories, x_descriptions = [], [], []
-        x_fs, x_cs, x_ds = [], [], []
+        x_fs, x_cs, x_ds, alphas = [], [], [], []
 
         for current_time_series in x_information:
             x_fs.clear()
@@ -310,28 +307,28 @@ def execute_shap_model(x_dataset, x_information, y_dataset, file, algorithm, spl
         os.makedirs(path_file, exist_ok=True)
         plt.savefig(alpha)
         plt.close()
-        print("\n\n\t\t\tGRAPH has benn computed : ", alpha)
+        alphas.append(alpha)
 
         shap.summary_plot(np.array(shval).T, np.array(x_test).T, feature_names=series_names, show=display)
         alpha = os.path.join(path_file + file + "_" + algorithm + "_shap_reverse_plot.png")
         plt.title("SHAP Features by Series")
         plt.savefig(alpha)
         plt.close()
-        print("\t\t\tGRAPH has benn computed : ", alpha)
+        alphas.append(alpha)
 
         shap.plots.waterfall(shval_x[0], show=display)
         alpha = os.path.join(path_file + file + "_" + algorithm + "_DTL_Waterfall.png")
         plt.title("SHAP Waterfall Results")
         plt.savefig(alpha)
         plt.close()
-        print("\t\t\tGRAPH has benn computed : ", alpha)
+        alphas.append(alpha)
 
         shap.plots.beeswarm(shval_x, show=display)
         alpha = os.path.join(path_file + file + "_" + algorithm + "_DTL_Beeswarm.png")
         plt.title("SHAP Beeswarm Results")
         plt.savefig(alpha)
         plt.close()
-        print("\t\t\tGRAPH has benn computed : ", alpha)
+        alphas.append(alpha)
 
         total_weights_for_all_algorithms = []
 
@@ -388,31 +385,31 @@ def execute_shap_model(x_dataset, x_information, y_dataset, file, algorithm, spl
         plt.title("SHAP details of geometry")
         plt.savefig(alpha)
         plt.close()
-        print("\t\t\tGRAPH has benn computed : ", alpha)
+        alphas.append(alpha)
 
         shap.summary_plot(np.array(transformation).T, np.array(transformationT).T, plot_size=(20, 10),
                           feature_names=transformationDesc, show=display)
         alpha = os.path.join(path_file + file + "_" + algorithm + "_shap_transformation_plot.png")
         plt.title("SHAP details of transformation")
         plt.savefig(alpha)
         plt.close()
-        print("\t\t\tGRAPH has benn computed : ", alpha)
+        alphas.append(alpha)
 
         shap.summary_plot(np.array(correlation).T, np.array(correlationT).T, plot_size=(20, 10),
                           feature_names=correlationDesc, show=display)
         alpha = os.path.join(path_file + file + "_" + algorithm + "_shap_correlation_plot.png")
         plt.title("SHAP details of correlation")
         plt.savefig(alpha)
         plt.close()
-        print("\t\t\tGRAPH has benn computed : ", alpha)
+        alphas.append(alpha)
 
         shap.summary_plot(np.array(trend).T, np.array(trendT).T, plot_size=(20, 8), feature_names=trendDesc,
                           show=display)
         alpha = os.path.join(path_file + file + "_" + algorithm + "_shap_trend_plot.png")
         plt.title("SHAP details of Trend")
         plt.savefig(alpha)
         plt.close()
-        print("\t\t\tGRAPH has benn computed : ", alpha)
+        alphas.append(alpha)
 
         aggregation_features.append(np.mean(geometry, axis=0))
         aggregation_features.append(np.mean(correlation, axis=0))
@@ -434,15 +431,15 @@ def execute_shap_model(x_dataset, x_information, y_dataset, file, algorithm, spl
         plt.gca().axes.get_xaxis().set_visible(False)
         plt.savefig(alpha)
         plt.close()
-        print("\t\t\tGRAPH has benn computed : ", alpha)
+        alphas.append(alpha)
 
         shap.summary_plot(np.array(aggregation_features).T, np.array(aggregation_test).T, feature_names=series_names,
                           show=display)
         alpha = os.path.join(path_file + file + "_" + algorithm + "_shap_aggregate_reverse_plot.png")
         plt.title("SHAP Aggregation Features by Series")
         plt.savefig(alpha)
         plt.close()
-        print("\t\t\tGRAPH has benn computed : ", alpha, "\n\n")
+        alphas.append(alpha)
 
         if verbose:
             print("\t\tSHAP Families details :")
@@ -461,6 +458,9 @@ def execute_shap_model(x_dataset, x_information, y_dataset, file, algorithm, spl
 
         total_weights_for_all_algorithms = np.append(total_weights_for_all_algorithms, total_weights_percent)
 
+        for alpha in alphas:
+            print("\n\n\t\t\tplot has been saved : ", alpha)
+
         results_shap = Explainer.convert_results(total_weights_for_all_algorithms, file, algorithm, x_descriptions,
                                                  x_features, x_categories, mean_features,
                                                  to_save=path_file + file + "_" + algorithm)

diff --git a/params/optimal_parameters_e_eeg-alcohol_cdrec.toml b/params/optimal_parameters_e_eeg-alcohol_cdrec.toml
@@ -1,4 +1,4 @@
 [cdrec]
-rank = 9
-epsilon = 0.0010055574182674504
-iteration = 489
+rank = 2
+epsilon = 2.9103006766169248e-5
+iteration = 976
diff --git a/params/optimal_parameters_e_eeg-alcohol_stmvl.toml b/params/optimal_parameters_e_eeg-alcohol_stmvl.toml
@@ -1,4 +1,4 @@
 [stmvl]
-window_size = 70
-gamma = 1.0621347452712664e-6
-alpha = 2
+window_size = 40
+gamma = 0.22504996184885026
+alpha = 6
diff --git a/requirements.txt b/requirements.txt
@@ -2,7 +2,7 @@ numpy==1.26.4
 matplotlib==3.7.5
 toml==0.10.2
 scikit-learn==1.3.2
-scipy==1.14.1
+#scipy==1.14.1
 setuptools==75.1.0
 tensorflow==2.17.0
 shap==0.44.1
@@ -11,6 +11,11 @@ scikit-optimize==0.10.2
 pyswarms==1.3.0
 pytest==8.3.3
 xlsxwriter==3.2.0
+
+# Specify scipy versions conditionally
+scipy==1.10.1; python_version < "3.10"
+scipy==1.14.1; python_version >= "3.10"
+
 types-toml
 types-setuptools
 wheel
diff --git a/tests/__pycache__/test_benchmarking.cpython-312-pytest-8.3.3.pyc b/tests/__pycache__/test_benchmarking.cpython-312-pytest-8.3.3.pyc
diff --git a/tests/__pycache__/test_explainer.cpython-312-pytest-8.3.3.pyc b/tests/__pycache__/test_explainer.cpython-312-pytest-8.3.3.pyc
diff --git a/tests/assets/shap/chlorine_cdrec_DTL_Beeswarm.png b/tests/assets/shap/chlorine_cdrec_DTL_Beeswarm.png
diff --git a/tests/assets/shap/chlorine_cdrec_DTL_Waterfall.png b/tests/assets/shap/chlorine_cdrec_DTL_Waterfall.png
diff --git a/tests/assets/shap/chlorine_cdrec_results.txt b/tests/assets/shap/chlorine_cdrec_results.txt
@@ -1,22 +1,22 @@
-Feature : 6     cdrec      with a score of 22.22      Geometry           Proportion of high incremental changes in the series              MD_hrv_classic_pnn40
-Feature : 15    cdrec      with a score of 10.24      Transformation     Power in the lowest 20% of frequencies                            SP_Summaries_welch_rect_area_5_1
-Feature : 10    cdrec      with a score of 9.27       Geometry           Goodness of exponential fit to embedding distance distribution    CO_Embed2_Dist_tau_d_expfit_meandiff
-Feature : 0     cdrec      with a score of 9.07       Geometry           5-bin histogram mode                                              DN_HistogramMode_5
-Feature : 5     cdrec      with a score of 8.31       Correlation        Time reversibility                                                CO_trev_1_num
-Feature : 21    cdrec      with a score of 7.84       Trend              Error of 3-point rolling mean forecast                            FC_LocalSimple_mean3_stderr
-Feature : 2     cdrec      with a score of 7.65       Correlation        First 1/e crossing of the ACF                                     CO_f1ecac
-Feature : 14    cdrec      with a score of 6.67       Geometry           Negative outlier timing                                           DN_OutlierInclude_n_001_mdrmd
-Feature : 1     cdrec      with a score of 5.0        Geometry           10-bin histogram mode                                             DN_HistogramMode_10
-Feature : 8     cdrec      with a score of 3.75       Geometry           Transition matrix column variance                                 SB_TransitionMatrix_3ac_sumdiagcov
-Feature : 4     cdrec      with a score of 3.11       Correlation        Histogram-based automutual information (lag 2, 5 bins)            CO_HistogramAMI_even_2_5
-Feature : 17    cdrec      with a score of 2.06       Trend              Entropy of successive pairs in symbolized series                  SB_MotifThree_quantile_hh
-Feature : 3     cdrec      with a score of 1.63       Correlation        First minimum of the ACF                                          CO_FirstMin_ac
-Feature : 12    cdrec      with a score of 1.58       Correlation        Change in autocorrelation timescale after incremental differencing FC_LocalSimple_mean1_tauresrat
-Feature : 13    cdrec      with a score of 1.55       Geometry           Positive outlier timing                                           DN_OutlierInclude_p_001_mdrmd
-Feature : 9     cdrec      with a score of 0.04       Trend              Wangs periodicity metric                                          PD_PeriodicityWang_th0_01
+Feature : 1     cdrec      with a score of 90.54      Geometry           10-bin histogram mode                                             DN_HistogramMode_10
+Feature : 12    cdrec      with a score of 3.99       Correlation        Change in autocorrelation timescale after incremental differencing FC_LocalSimple_mean1_tauresrat
+Feature : 5     cdrec      with a score of 3.83       Correlation        Time reversibility                                                CO_trev_1_num
+Feature : 18    cdrec      with a score of 0.57       Geometry           Rescaled range fluctuation analysis (low-scale scaling)           SC_FluctAnal_2_rsrangefit_50_1_logi_prop_r1
+Feature : 13    cdrec      with a score of 0.37       Geometry           Positive outlier timing                                           DN_OutlierInclude_p_001_mdrmd
+Feature : 3     cdrec      with a score of 0.33       Correlation        First minimum of the ACF                                          CO_FirstMin_ac
+Feature : 14    cdrec      with a score of 0.29       Geometry           Negative outlier timing                                           DN_OutlierInclude_n_001_mdrmd
+Feature : 6     cdrec      with a score of 0.09       Geometry           Proportion of high incremental changes in the series              MD_hrv_classic_pnn40
+Feature : 0     cdrec      with a score of 0.0        Geometry           5-bin histogram mode                                              DN_HistogramMode_5
+Feature : 2     cdrec      with a score of 0.0        Correlation        First 1/e crossing of the ACF                                     CO_f1ecac
+Feature : 4     cdrec      with a score of 0.0        Correlation        Histogram-based automutual information (lag 2, 5 bins)            CO_HistogramAMI_even_2_5
 Feature : 7     cdrec      with a score of 0.0        Geometry           Longest stretch of above-mean values                              SB_BinaryStats_mean_longstretch1
+Feature : 8     cdrec      with a score of 0.0        Geometry           Transition matrix column variance                                 SB_TransitionMatrix_3ac_sumdiagcov
+Feature : 9     cdrec      with a score of 0.0        Trend              Wangs periodicity metric                                          PD_PeriodicityWang_th0_01
+Feature : 10    cdrec      with a score of 0.0        Geometry           Goodness of exponential fit to embedding distance distribution    CO_Embed2_Dist_tau_d_expfit_meandiff
 Feature : 11    cdrec      with a score of 0.0        Correlation        First minimum of the AMI function                                 IN_AutoMutualInfoStats_40_gaussian_fmmi
+Feature : 15    cdrec      with a score of 0.0        Transformation     Power in the lowest 20% of frequencies                            SP_Summaries_welch_rect_area_5_1
 Feature : 16    cdrec      with a score of 0.0        Geometry           Longest stretch of decreasing values                              SB_BinaryStats_diff_longstretch0
-Feature : 18    cdrec      with a score of 0.0        Geometry           Rescaled range fluctuation analysis (low-scale scaling)           SC_FluctAnal_2_rsrangefit_50_1_logi_prop_r1
+Feature : 17    cdrec      with a score of 0.0        Trend              Entropy of successive pairs in symbolized series                  SB_MotifThree_quantile_hh
 Feature : 19    cdrec      with a score of 0.0        Geometry           Detrended fluctuation analysis (low-scale scaling)                SC_FluctAnal_2_dfa_50_1_2_logi_prop_r1
 Feature : 20    cdrec      with a score of 0.0        Transformation     Centroid frequency                                                SP_Summaries_welch_rect_centroid
+Feature : 21    cdrec      with a score of 0.0        Trend              Error of 3-point rolling mean forecast                            FC_LocalSimple_mean3_stderr
diff --git a/tests/assets/shap/chlorine_cdrec_shap_aggregate_plot.png b/tests/assets/shap/chlorine_cdrec_shap_aggregate_plot.png
diff --git a/tests/assets/shap/chlorine_cdrec_shap_aggregate_reverse_plot.png b/tests/assets/shap/chlorine_cdrec_shap_aggregate_reverse_plot.png
diff --git a/tests/assets/shap/chlorine_cdrec_shap_correlation_plot.png b/tests/assets/shap/chlorine_cdrec_shap_correlation_plot.png
diff --git a/tests/assets/shap/chlorine_cdrec_shap_geometry_plot.png b/tests/assets/shap/chlorine_cdrec_shap_geometry_plot.png
diff --git a/tests/assets/shap/chlorine_cdrec_shap_plot.png b/tests/assets/shap/chlorine_cdrec_shap_plot.png
diff --git a/tests/assets/shap/chlorine_cdrec_shap_reverse_plot.png b/tests/assets/shap/chlorine_cdrec_shap_reverse_plot.png
diff --git a/tests/assets/shap/chlorine_cdrec_shap_transformation_plot.png b/tests/assets/shap/chlorine_cdrec_shap_transformation_plot.png
diff --git a/tests/assets/shap/chlorine_cdrec_shap_trend_plot.png b/tests/assets/shap/chlorine_cdrec_shap_trend_plot.png
diff --git a/tests/reports/benchmarking_rmse.jpg b/tests/reports/benchmarking_rmse.jpg
diff --git a/tests/reports/run_0/eegalcohol_mcar_CORRELATION.jpg b/tests/reports/run_0/eegalcohol_mcar_CORRELATION.jpg
diff --git a/tests/reports/run_0/eegalcohol_mcar_MAE.jpg b/tests/reports/run_0/eegalcohol_mcar_MAE.jpg
diff --git a/tests/reports/run_0/eegalcohol_mcar_MI.jpg b/tests/reports/run_0/eegalcohol_mcar_MI.jpg
diff --git a/tests/reports/run_0/eegalcohol_mcar_RMSE.jpg b/tests/reports/run_0/eegalcohol_mcar_RMSE.jpg
diff --git a/tests/reports/run_0/eegalcohol_mcar_imputation_time.jpg b/tests/reports/run_0/eegalcohol_mcar_imputation_time.jpg
diff --git a/tests/reports/run_0/eegalcohol_mcar_log_imputation.jpg b/tests/reports/run_0/eegalcohol_mcar_log_imputation.jpg
diff --git a/tests/reports/run_0/eegalcohol_mcar_metrics_subplot.jpg b/tests/reports/run_0/eegalcohol_mcar_metrics_subplot.jpg