Refactor experiments

• rename baselines experiments
• merge all baseline model experiments into one file replicate_experiments.py
• move each model_dataset config to configurations.py
• create models.py with classes of each model to abstract instantiaion of each model.
• refactor evaluators.py so only single and competing events evaluator.
• clean up process_results.py
• add SurvTRACE and variants to replicate_experiments.py
• document

baselines/data_class.py

-import numpy as np
-from survtrace.dataset import load_data
-
-from baselines.utils import df_to_event_time_array
-
-class Data:
-
-    def __init__(self, config, censor_event=False):
-
-        # load data
-        # TODO: splitting the data into train, validation, test should be separate process. Each run of each experiment
-        # should use the same train, validation, test datasets so the results are comparable. currently each run, of each
-        # dataset uses its own train, validation, test.
-        self.df, self.df_train, self.df_y_train, self.df_test, self.df_y_test, self.df_val, self.df_y_val = load_data(config)
-
-        # censor event
-        if censor_event:
-            censor_event = lambda df: df.drop(config.event_to_censor, axis=1).rename(columns={config.event_to_keep: 'event'})
-
-            self.df = censor_event(self.df)
-            self.df_y_train = censor_event(self.df_y_train)
-            self.df_y_val = censor_event(self.df_y_val)
-            self.df_y_test = censor_event(self.df_y_test)
-
-        # competing events
-        if ((config.model == 'DeepHit' or config.model == 'DSM') and config.data == 'seer'):
-            # make event variable for pycox model
-            # 0: right censored
-            # 1: Heart Disease (event_0)
-            # 2: Breast Cancer (event_1)
-            def make_event_col(df):
-                df['event'] = np.where(df['event_0'] == 1,
-                                        1,
-                                        np.where(df['event_1'] == 1,
-                                                2,
-                                                0
-                                                )
-                                        )
-                return df
-
-            self.df, self.df_y_train, self.df_y_val, self.df_y_test = map(make_event_col, [self.df, self.df_y_train, self.df_y_val, self.df_y_test])
-    
-
-        # convet data to format necessary for pycox
-        self.x_train = np.array(self.df_train, dtype='float32')
-        self.x_val = np.array(self.df_val, dtype='float32')
-        self.x_test = np.array(self.df_test, dtype='float32')
-
-        if config.model == 'PCHazard':
-            y_df_to_tuple = lambda df: tuple([np.array(df['duration'], dtype='int64'), np.array(df['event'], dtype='int64'), np.array(df['proportion'], dtype='float32')])
-        else:
-            y_df_to_tuple = lambda df: tuple([np.array(df['duration'], dtype='int64'), np.array(df['event'], dtype='int64')])
-
-        self.y_train = y_df_to_tuple(self.df_y_train)
-        self.y_val = y_df_to_tuple(self.df_y_val)
-
-        # package in tuple to pass to pycox models
-        self.val_data = tuple([self.x_val, self.y_val])
-
-        # process data for DSM
-        if config.model == 'DSM':
-            _, self.train_outcomes = self.y_train
-            self.train_times = np.array(self.df.loc[self.df_train.index]['duration'])
-            _, self.val_outcomes = self.y_val
-            self.val_times = np.array(self.df.loc[self.df_val.index]['duration'])
-
-        if config.model == 'CPH' or config.model == 'RSF':
-            self.y_et_train = df_to_event_time_array(self.df_y_train)
-
-