Merge pull request #173 from COINtoolbox/172-add-malanchev-to-time-do…

…main

implement malanchev features for time_domain

resspect/feature_extractors/light_curve.py

@@ -236,7 +236,7 @@ def conv_flux_mag(flux, zpt: float = 27.5):
         return np.array(mag)
 
     def check_queryable(self, mjd: float, filter_lim: float, criteria: int =1,
-                        days_since_last_obs=2, feature_method='Bazin',
+                        days_since_last_obs=2, feature_method: str = 'Bazin',
                         filter_cut='r'):
         """Check if this object can be queried in a given day.
 
@@ -307,7 +307,7 @@ def check_queryable(self, mjd: float, filter_lim: float, criteria: int =1,
                 self.last_mag = self.conv_flux_mag([fitted_flux])[0]
 
             else:
-                raise ValueError('Only "Bazin" features are implemented!')
+                raise ValueError('Only "Bazin" and "malanchev" features are implemented!')
 
         elif sum(surv_flag):
             raise ValueError('Criteria needs to be "1" or "2". \n ' + \

resspect/feature_extractors/malanchev.py

@@ -71,8 +71,6 @@ def fit(self, band: str) -> np.ndarray:
 
         # build filter flag
         band_indices = self.photometry['band'] == band
-        if not sum(band_indices) > (len(self.features_names) - 1):
-            return np.array([])
 
         extractor = licu.Extractor(licu.AndersonDarlingNormal(),
                                    licu.InterPercentileRange(0.05),

resspect/time_domain_loop.py

@@ -70,7 +70,7 @@ def load_dataset(file_names_dict: dict, survey_name: str = 'DES',
         Currently only nclass == 2 is implemented.
     feature_extraction_method: str (optional)
         Feature extraction method. The current implementation only
-        accepts method=='bazin' or 'photometry'.
+        accepts method=='bazin', 'photometry', or 'malanchev'.
         Default is 'bazin'.
     is_save_samples: bool (optional)
         If True, save training and test samples to file.
@@ -137,10 +137,10 @@ def _load_first_loop_and_full_data(
     number_of_classes
         Number of classes to consider in the classification
         Currently only number_of_classes == 2 is implemented.
-    feature_extraction_method
-        Chosen classifier.
-        The current implementation accepts `RandomForest`,
-        'GradientBoostedTrees', 'KNN', 'MLP', 'SVM' and 'NB'.
+    feature_extraction_method: str (optional)
+        Feature extraction method. The current implementation only
+        accepts method=='bazin', 'photometry', or 'malanchev'.
+        Default is 'bazin'.
     is_save_samples
         If True, save training and test samples to file.
         Default is False.
@@ -151,14 +151,16 @@ def _load_first_loop_and_full_data(
             file_names_dict=first_loop_file_name,
             survey_name=survey_name, is_separate_files=is_separate_files,
             initial_training=0, ia_training_fraction=ia_training_fraction,
-            is_queryable=is_queryable)
+            is_queryable=is_queryable, 
+            feature_extraction_method=feature_extraction_method)
         light_curve_file_name = {None: initial_light_curve_file_name['train']}
         light_curve_data = load_dataset(
             file_names_dict=light_curve_file_name,
             survey_name=survey_name, is_separate_files=is_separate_files,
             initial_training=initial_training,
             ia_training_fraction=ia_training_fraction,
-            is_queryable=is_queryable)
+            is_queryable=is_queryable, 
+            feature_extraction_method=feature_extraction_method)
     else:
         first_loop_file_name = {'pool': first_loop_file_name}
         first_loop_data = load_dataset(
@@ -459,7 +461,7 @@ def _save_metrics_and_queried_sample(
 def _load_next_day_data(
         next_day_features_file_name: str, is_separate_files: bool,
         is_queryable: bool, survey_name: str, ia_training_fraction: float,
-        is_save_samples: bool):
+        is_save_samples: bool, feature_extraction_method: str='bazin'):
     """
     Loads features of next day
 
@@ -488,13 +490,14 @@ def _load_next_day_data(
         next_day_data = load_dataset(
             next_day_features_file_name, survey_name, samples_list=['pool'],
             is_separate_files=is_separate_files, is_queryable=is_queryable,
-            is_save_samples=is_save_samples
-        )
+            is_save_samples=is_save_samples,
+            feature_extraction_method=feature_extraction_method)
     else:
         next_day_features_file_name = {None: next_day_features_file_name}
         next_day_data = load_dataset(
             next_day_features_file_name, survey_name, is_queryable=is_queryable,
-            initial_training=0, ia_training_fraction=ia_training_fraction)
+            initial_training=0, ia_training_fraction=ia_training_fraction,
+            feature_extraction_method=feature_extraction_method)
     return next_day_data
 
 
@@ -724,7 +727,8 @@ def process_next_day_loop(
         is_separate_files: bool, is_queryable: bool, survey_name: str,
         ia_training_fraction: float, id_key_name: str,
         light_curve_train_ids: np.ndarray, is_save_samples: bool,
-        canonical_data: DataBase, strategy: str) -> DataBase:
+        canonical_data: DataBase, strategy: str, 
+        feature_extraction_method: str='bazin') -> DataBase:
     """
     Runs next day active learning loop
 
@@ -759,10 +763,15 @@ def process_next_day_loop(
         Query strategy. Options are (all can be run with budget):
         "UncSampling", "UncSamplingEntropy", "UncSamplingLeastConfident",
         "UncSamplingMargin", "QBDMI", "QBDEntropy", "RandomSampling"
+    feature_extraction_method: str (optional)
+        Feature extraction method. The current implementation only
+        accepts method=='bazin' or 'photometry'.
+        Default is 'bazin'.
     """
     next_day_data = _load_next_day_data(
         next_day_features_file_name, is_separate_files, is_queryable,
-        survey_name, ia_training_fraction, is_save_samples)
+        survey_name, ia_training_fraction, is_save_samples,
+        feature_extraction_method=feature_extraction_method)
     for metadata_value in light_curve_data.train_metadata[id_key_name].values:
         next_day_pool_metadata = next_day_data.pool_metadata[id_key_name].values
         if metadata_value in next_day_pool_metadata:
@@ -800,7 +809,7 @@ def run_time_domain_active_learning_loop(
         light_curve_train_ids: np.ndarray, canonical_data: DataBase,
         is_separate_files: bool, path_to_features_directory: str,
         fname_pattern: list, survey_name: str,  ia_training_fraction: float,
-        is_save_samples: bool, **kwargs: dict):
+        is_save_samples: bool, feature_extraction_method: str='bazin', **kwargs: dict):
     """
     Runs time domain active learning loop
 
@@ -861,6 +870,10 @@ def run_time_domain_active_learning_loop(
     is_save_samples
        If True, save training and test samples to file.
         Default is False.
+    feature_extraction_method: str (optional)
+        Feature extraction method. The current implementation only
+        accepts method=='bazin' or 'photometry'.
+        Default is 'bazin'.
     kwargs
        All keywords required by the classifier function.
 
@@ -900,7 +913,8 @@ def run_time_domain_active_learning_loop(
         light_curve_data = process_next_day_loop(
             light_curve_data, next_day_features_file_name, is_separate_files,
             is_queryable, survey_name,  ia_training_fraction, id_key_name,
-            light_curve_train_ids, is_save_samples, canonical_data, strategy)
+            light_curve_train_ids, is_save_samples, canonical_data, strategy,
+            feature_extraction_method=feature_extraction_method)
 
 
 # TODO: Too many arguments. Refactor and update docs
@@ -915,6 +929,7 @@ def time_domain_loop(days: list, output_metrics_file: str,
                      query_thre: float = 1.0, save_samples: bool = False,
                      sep_files: bool = False, survey: str = 'LSST',
                      initial_training: str = 'original',
+                     feature_extraction_method: str = 'bazin',
                      save_full_query: bool = False, **kwargs):
     """
     Perform the active learning loop. All results are saved to file.
@@ -992,6 +1007,10 @@ def time_domain_loop(days: list, output_metrics_file: str,
         If int: choose the required number of samples at random,
         ensuring that at least half are SN Ia
         Default is 'original'.
+    feature_extraction_method: str (optional)
+        Feature extraction method. The current implementation only
+        accepts method=='bazin' or 'photometry'.
+        Default is 'bazin'.
     """
 
     # load features for the first obs day
@@ -1001,8 +1020,8 @@ def time_domain_loop(days: list, output_metrics_file: str,
 
     first_loop_data, light_curve_data = _load_first_loop_and_full_data(
         first_loop_file_name, path_to_ini_files, survey, initial_training,
-        ia_frac, queryable, sep_files, nclass, is_save_samples=save_samples
-    )
+        ia_frac, queryable, sep_files, nclass, is_save_samples=save_samples,
+        feature_extraction_method = feature_extraction_method)
 
     # get keyword for obj identification
     id_key_name = light_curve_data.identify_keywords()
@@ -1024,7 +1043,8 @@ def time_domain_loop(days: list, output_metrics_file: str,
         strategy, budgets, queryable, query_thre, batch, output_metrics_file,
         output_queried_file, save_full_query, id_key_name,
         light_curve_train_ids, canonical_data, sep_files, path_to_features_dir,
-        fname_pattern, survey, ia_frac, save_samples, **kwargs)
+        fname_pattern, survey, ia_frac, save_samples,
+        feature_extraction_method=feature_extraction_method, **kwargs)
 
 def main():
     return None

resspect/time_domain_snpcc.py

@@ -22,7 +22,9 @@
 
 from resspect.feature_extractors.bazin import BazinFeatureExtractor
 from resspect.feature_extractors.bump import BumpFeatureExtractor
+from resspect.feature_extractors.malanchev import MalanchevFeatureExtractor
 from resspect.lightcurves_utils import BAZIN_HEADERS
+from resspect.lightcurves_utils import MALANCHEV_HEADERS
 from resspect.lightcurves_utils import get_files_list
 from resspect.lightcurves_utils import get_query_flags
 from resspect.lightcurves_utils import maybe_create_directory
@@ -34,20 +36,22 @@
 
 FEATURE_EXTRACTOR_MAPPING = {
     "bazin": BazinFeatureExtractor,
-    "bump": BumpFeatureExtractor
+    "bump": BumpFeatureExtractor,
+    "malanchev": MalanchevFeatureExtractor
 }
 
 
 FEATURE_EXTRACTOR_HEADERS_MAPPING = {
-    "bazin": BAZIN_HEADERS
+    "bazin": BAZIN_HEADERS,
+    "malanchev": MALANCHEV_HEADERS
 }
 
 
 class SNPCCPhotometry:
     """
     Handles photometric information for entire SNPCC data.
 
-    This class only works for Bazin feature extraction method.
+    This class only works for Bazin and Malanchev feature extraction methods.
 
     Attributes
     ----------
@@ -105,15 +109,15 @@ def create_daily_file(self, output_dir: str,
             If True, calculate cost of taking a spectra in the last
             observed photometric point. Default is False.
         feature_extractor: str
-            Feature extraction method, only possibility is 'Bazin'.
+            Feature extraction method, only possibilities are 'Bazin' and 'malanchev'.
         """
         maybe_create_directory(output_dir)
         self._features_file_name = os.path.join(
             output_dir, 'day_' + str(day) + '.csv')
         logging.info('Creating features file')
         with open(self._features_file_name, 'w') as features_file:
             if feature_extractor not in FEATURE_EXTRACTOR_HEADERS_MAPPING:
-                raise ValueError('Only Bazin headers are supported')
+                raise ValueError('Only Bazin and Malanchev headers are supported')
             self._header = FEATURE_EXTRACTOR_HEADERS_MAPPING[
                 feature_extractor]['snpcc_header']
             if get_cost:
@@ -152,7 +156,7 @@ def _maybe_create_features_file(self, output_dir: str, day_of_survey: int,
         day_of_survey: int
             Day since the beginning of survey.
         feature_extractor: str
-            Feature extraction method, only possibility is 'Bazin'.
+            Feature extraction method, only possibilities are 'Bazin' and 'malanchev'.
         get_cost: bool
            if True, cost of taking a spectra is computed.
         """
@@ -171,13 +175,13 @@ def _verify_dataset_and_features_method(dataset_name: str,
         dataset_name: str
             name of the dataset used
         feature_extractor: str
-            Feature extraction method, only possibility is 'Bazin'.
+            Feature extraction method, only possibilities are 'Bazin' and 'malanchev'.
         """
         if dataset_name != 'SNPCC':
             raise ValueError('This class supports only SNPCC dataset!')
         # TODO: Update when bump headers are available
-        if feature_extractor != 'bazin':
-            raise ValueError('Only bazin features are implemented!!')
+        if feature_extractor != 'bazin' and feature_extractor!='malanchev':
+            raise ValueError('Only bazin and malanchev features are implemented!!')
 
     def _check_queryable(self, light_curve_data,
                          queryable_criteria: int,
@@ -384,7 +388,7 @@ def build_one_epoch(self, raw_data_dir: str, day_of_survey: int,
         """
         Fit features for all objects with enough points in a given day.
 
-        Generate 1 file containing best-fit Bazin parameters for a given
+        Generate 1 file containing best-fit Bazin or malanchev parameters for a given
         day of the survey.
 
         Parameters
@@ -404,7 +408,7 @@ def build_one_epoch(self, raw_data_dir: str, day_of_survey: int,
             Only used if "queryable_criteria == 2". Default is 2.
         feature_extractor: str (optional)
             Feature extraction method.
-            Only possibility is 'Bazin'.
+            Only possibilities are 'Bazin' or 'malanchev'.
         get_cost: bool (optional)
             If True, calculate cost of taking a spectra in the last
             observed photometric point. Default is False.