data-manipulation-utilities 0.2.6__py3-none-any.whl → 0.2.7__py3-none-any.whl

dmu/ml/train_mva.py

@@ -3,8 +3,10 @@ Module with TrainMva class
 '''
 # pylint: disable = too-many-locals, no-name-in-module
 # pylint: disable = too-many-arguments, too-many-positional-arguments
+# pylint: disable = too-many-instance-attributes
 
 import os
+import copy
 
 import joblib
 import pandas as pnd
@@ -20,12 +22,13 @@ import dmu.ml.utilities         as ut
 import dmu.pdataframe.utilities as put
 import dmu.plotting.utilities   as plu
 
+from dmu.ml.cv_diagnostics   import CVDiagnostics
 from dmu.ml.cv_classifier    import CVClassifier as cls
 from dmu.plotting.plotter_1d import Plotter1D    as Plotter
 from dmu.plotting.matrix     import MatrixPlotter
 from dmu.logging.log_store   import LogStore
 
-npa = numpy.ndarray
+NPA = numpy.ndarray
 log = LogStore.add_logger('dmu:ml:train_mva')
 # ---------------------------------------------
 class TrainMva:
@@ -42,30 +45,62 @@ class TrainMva:
         self._cfg       = cfg
         self._l_ft_name = self._cfg['training']['features']
 
-        df_ft_sig, l_lab_sig = self._get_sample_inputs(rdf = sig, label = 1)
-        df_ft_bkg, l_lab_bkg = self._get_sample_inputs(rdf = bkg, label = 0)
+        self._rdf_sig_org = sig
+        self._rdf_bkg_org = bkg 
+
+        rdf_bkg = self._preprocess_rdf(bkg)
+        rdf_sig = self._preprocess_rdf(sig)
+
+        df_ft_sig, l_lab_sig = self._get_sample_inputs(rdf = rdf_sig, label = 1)
+        df_ft_bkg, l_lab_bkg = self._get_sample_inputs(rdf = rdf_bkg, label = 0)
 
         self._df_ft = pnd.concat([df_ft_sig, df_ft_bkg], axis=0)
         self._l_lab = numpy.array(l_lab_sig + l_lab_bkg)
 
-        self._rdf_bkg = self._get_rdf(rdf = bkg, df=df_ft_bkg)
-        self._rdf_sig = self._get_rdf(rdf = sig, df=df_ft_sig)
+        self._rdf_bkg = self._get_rdf(rdf = rdf_bkg, df_feat=df_ft_bkg)
+        self._rdf_sig = self._get_rdf(rdf = rdf_sig, df_feat=df_ft_sig)
     # ---------------------------------------------
-    def _get_rdf(self, rdf : RDataFrame, df : pnd.DataFrame) -> RDataFrame:
+    def _get_extra_columns(self, rdf : RDataFrame, df : pnd.DataFrame) -> list[str]:
+        d_plot = self._cfg['plotting']['features']['plots']
+        l_expr = list(d_plot)
+        l_rdf  = [ name.c_str() for name in rdf.GetColumnNames() ]
+
+        l_extr = []
+        for expr in l_expr:
+            if expr not in l_rdf:
+                continue
+
+            if expr in df.columns:
+                continue
+
+            l_extr.append(expr)
+
+        return l_extr
+    # ---------------------------------------------
+    def _get_rdf(self, rdf : RDataFrame, df_feat : pnd.DataFrame) -> RDataFrame:
         '''
         Takes original ROOT dataframe and pre-processed features dataframe
         Adds missing branches to latter and returns expanded ROOT dataframe
+        Need to make plots
         '''
 
-        l_pnd_col = df.columns.tolist()
-        l_rdf_col = [ name.c_str() for name in rdf.GetColumnNames() ]
-        l_mis_col = [ col for col in l_rdf_col if col not in l_pnd_col ]
+        l_extr_col = self._get_extra_columns(rdf, df_feat)
+        if len(l_extr_col) > 20:
+            for name in l_extr_col:
+                log.debug(name)
+            raise ValueError('Found more than 20 extra columns')
+
+        d_data = rdf.AsNumpy(l_extr_col)
+        log.debug(f'Adding extra-nonfeature columns: {l_extr_col}')
+        df_extr = pnd.DataFrame(d_data)
+
+        nmain = len(df_feat.columns)
+        nextr = len(df_extr.columns)
 
-        log.debug(f'Adding extra-nonfeature columns: {l_mis_col}')
+        log.debug(f'Main  DF size: {nmain}')
+        log.debug(f'Extra DF size: {nextr}')
 
-        d_data = rdf.AsNumpy(l_mis_col)
-        df_ext = pnd.DataFrame(d_data)
-        df_all = pnd.concat([df, df_ext], axis=1)
+        df_all = pnd.concat([df_feat, df_extr], axis=1)
 
         return RDF.FromPandas(df_all)
     # ---------------------------------------------
@@ -90,6 +125,19 @@ class TrainMva:
 
         return df
     # ---------------------------------------------
+    def _preprocess_rdf(self, rdf : RDataFrame) -> RDataFrame:
+        if 'define' not in self._cfg['dataset']:
+            log.debug('No definitions found')
+            return rdf
+
+        log.debug('Definitions found')
+        d_def = self._cfg['dataset']['define']
+        for name, expr in d_def.items():
+            log.debug(f'{name:<20}{expr}')
+            rdf = rdf.Define(name, expr)
+
+        return rdf
+    # ---------------------------------------------
     def _get_sample_inputs(self, rdf : RDataFrame, label : int) -> tuple[pnd.DataFrame, list[int]]:
         d_ft = rdf.AsNumpy(self._l_ft_name)
         df   = pnd.DataFrame(d_ft)
@@ -99,7 +147,7 @@ class TrainMva:
 
         return df, l_lab
     # ---------------------------------------------
-    def _get_model(self, arr_index : npa) -> cls:
+    def _get_model(self, arr_index : NPA) -> cls:
         model = cls(cfg = self._cfg)
         df_ft = self._df_ft.iloc[arr_index]
         l_lab = self._l_lab[arr_index]
@@ -111,10 +159,14 @@ class TrainMva:
 
         return model
     # ---------------------------------------------
-    def _get_models(self):
+    def _get_models(self, load_trained : bool):
         '''
         Will create models, train them and return them
         '''
+        if load_trained:
+            log.warning('Not retraining, but loading trained models')
+            return self._load_trained_models()
+
         nfold = self._cfg['training']['nfold']
         rdmst = self._cfg['training']['rdm_stat']
 
@@ -141,6 +193,22 @@ class TrainMva:
 
         return l_model
     # ---------------------------------------------
+    def _load_trained_models(self) -> list[cls]:
+        model_path = self._cfg['saving']['path']
+        nfold      = self._cfg['training']['nfold']
+        l_model    = []
+        for ifold in range(nfold):
+            fold_path = model_path.replace('.pkl', f'_{ifold:03}.pkl')
+
+            if not os.path.isfile(fold_path):
+                raise FileNotFoundError(f'Missing trained model: {fold_path}')
+
+            log.debug(f'Loading model from: {fold_path}')
+            model = joblib.load(fold_path)
+            l_model.append(model)
+
+        return l_model
+    # ---------------------------------------------
     def _labels_from_varnames(self, l_var_name : list[str]) -> list[str]:
         try:
             d_plot = self._cfg['plotting']['features']['plots']
@@ -180,7 +248,7 @@ class TrainMva:
         d_form = {'Variable' : '{}', 'Importance' : '{:.1f}'}
         put.df_to_tex(df, table_path, d_format = d_form)
     # ---------------------------------------------
-    def _get_scores(self, model : cls, arr_index : npa, on_training_ok : bool) -> tuple[npa, npa, npa, npa]:
+    def _get_scores(self, model : cls, arr_index : NPA, on_training_ok : bool) -> tuple[NPA, NPA, NPA, NPA]:
         '''
         Returns a tuple of four arrays
 
@@ -203,7 +271,7 @@ class TrainMva:
 
         return arr_sig, arr_bkg, arr_all, arr_lab
     # ---------------------------------------------
-    def _split_scores(self, arr_prob : npa, arr_label : npa) -> tuple[npa, npa]:
+    def _split_scores(self, arr_prob : NPA, arr_label : NPA) -> tuple[NPA, NPA]:
         '''
         Will split the testing scores (predictions) based on the training scores
 
@@ -259,7 +327,7 @@ class TrainMva:
 
         return cfg
     # ---------------------------------------------
-    def _plot_correlation(self, arr_index : npa, ifold : int) -> None:
+    def _plot_correlation(self, arr_index : NPA, ifold : int) -> None:
         df_ft = self._df_ft.iloc[arr_index]
         cfg = self._get_correlation_cfg(df_ft, ifold)
         cov = df_ft.corr()
@@ -276,7 +344,7 @@ class TrainMva:
         plt.savefig(f'{val_dir}/covariance.png')
         plt.close()
     # ---------------------------------------------
-    def _get_nentries(self, arr_val : npa) -> str:
+    def _get_nentries(self, arr_val : NPA) -> str:
         size = len(arr_val)
         size = size / 1000.
 
@@ -311,10 +379,10 @@ class TrainMva:
         plt.close()
     # ---------------------------------------------
     def _plot_roc(self,
-                  l_lab_ts : npa,
-                  l_prb_ts : npa,
-                  l_lab_tr : npa,
-                  l_prb_tr : npa,
+                  l_lab_ts : NPA,
+                  l_prb_ts : NPA,
+                  l_lab_tr : NPA,
+                  l_prb_tr : NPA,
                   ifold    : int):
         '''
         Takes the labels and the probabilities and plots ROC
@@ -359,10 +427,10 @@ class TrainMva:
         plt.close()
     # ---------------------------------------------
     def _plot_probabilities(self,
-                            arr_seff: npa,
-                            arr_brej: npa,
-                            arr_sprb: npa,
-                            arr_labl: npa) -> None:
+                            arr_seff: NPA,
+                            arr_brej: NPA,
+                            arr_sprb: NPA,
+                            arr_labl: NPA) -> None:
 
         roc_cfg = self._cfg['plotting']['roc']
         if 'annotate' not in roc_cfg:
@@ -447,11 +515,32 @@ class TrainMva:
         os.makedirs(val_dir, exist_ok=True)
         put.df_to_tex(df, f'{val_dir}/hyperparameters.tex')
     # ---------------------------------------------
-    def run(self, skip_fit : bool = False) -> None:
+    def _run_diagnostics(self, models : list[cls], rdf : RDataFrame, name : str) -> None:
+        if 'diagnostics' not in self._cfg:
+            log.warning('Diagnostics section not found, not running diagnostics')
+            return
+
+        cfg_diag = self._cfg['diagnostics']
+        out_dir  = cfg_diag['output']
+        plt_dir  = None
+
+        if 'overlay' in cfg_diag['correlations']['target']:
+            plt_dir  = cfg_diag['correlations']['target']['overlay']['saving']['plt_dir']
+
+        cfg_diag = copy.deepcopy(cfg_diag)
+        cfg_diag['output'] = f'{out_dir}/{name}'
+        if plt_dir is not None:
+            cfg_diag['correlations']['target']['overlay']['saving']['plt_dir'] = f'{plt_dir}/{name}'
+
+        cvd = CVDiagnostics(models=models, rdf=rdf, cfg=cfg_diag)
+        cvd.run()
+    # ---------------------------------------------
+    def run(self, skip_fit : bool = False, load_trained : bool = False) -> None:
         '''
         Will do the training
 
         skip_fit: By default false, if True, it will only do the plots of features and save tables
+        load_trained: If true, it will load the models instead of training, by default false
         '''
         self._save_settings_to_tex()
         self._plot_features()
@@ -459,7 +548,11 @@ class TrainMva:
         if skip_fit:
             return
 
-        l_mod = self._get_models()
-        for ifold, mod in enumerate(l_mod):
-            self._save_model(mod, ifold)
+        l_mod = self._get_models(load_trained = load_trained)
+        if not load_trained:
+            for ifold, mod in enumerate(l_mod):
+                self._save_model(mod, ifold)
+
+        self._run_diagnostics(models = l_mod, rdf = self._rdf_sig_org, name='Signal'    )
+        self._run_diagnostics(models = l_mod, rdf = self._rdf_bkg_org, name='Background')
 # ---------------------------------------------

dmu/pdataframe/utilities.py

@@ -2,20 +2,28 @@
 Module containing utilities for pandas dataframes
 '''
 import os
+import yaml
 import pandas as pnd
 
 from dmu.logging.log_store import LogStore
 
 log=LogStore.add_logger('dmu:pdataframe:utilities')
-
 # -------------------------------------
-def df_to_tex(df : pnd.DataFrame, path : str, hide_index : bool = True, d_format : dict[str,str]=None, caption : str =None) -> None:
+def df_to_tex(df         : pnd.DataFrame,
+              path       : str,
+              hide_index : bool         = True,
+              d_format   : dict[str,str]= None,
+              **kwargs   : str       ) -> None:
     '''
     Saves pandas dataframe to latex
 
     Parameters
     -------------
+    df              : Dataframe with data
+    path     (str)  : Path to latex file
+    hide_index      : If true (default), index of dataframe won't appear in table
     d_format (dict) : Dictionary specifying the formattinng of the table, e.g. `{'col1': '{}', 'col2': '{:.3f}', 'col3' : '{:.3f}'}`
+    kwargs          : Arguments needed in `to_latex`
     '''
 
     if path is not None:
@@ -30,7 +38,32 @@ def df_to_tex(df : pnd.DataFrame, path : str, hide_index : bool = True, d_format
         st=st.format(formatter=d_format)
 
     log.info(f'Saving to: {path}')
-    buf = st.to_latex(buf=path, caption=caption, hrules=True)
+    buf = st.to_latex(buf=path, hrules=True, **kwargs)
 
     return buf
 # -------------------------------------
+def to_yaml(df : pnd.DataFrame, path : str):
+    '''
+    Takes a dataframe and the path to a yaml file
+    Makes the directory path if not found and saves data in YAML file
+    '''
+    dir_path = os.path.dirname(path)
+    os.makedirs(dir_path, exist_ok=True)
+
+    data = df.to_dict()
+
+    with open(path, 'w', encoding='utf-8') as ofile:
+        yaml.safe_dump(data, ofile)
+# -------------------------------------
+def from_yaml(path : str) -> pnd.DataFrame:
+    '''
+    Takes path to a yaml file
+    Makes dataframe from it and returns it
+    '''
+    with open(path, encoding='utf-8') as ifile:
+        data = yaml.safe_load(ifile)
+
+    df = pnd.DataFrame(data)
+
+    return df
+# -------------------------------------

dmu/plotting/fwhm.py

@@ -0,0 +1,64 @@
+'''
+Module with FWHM plugin class
+'''
+import zfit
+import numpy
+import matplotlib.pyplot as plt
+
+from dmu.logging.log_store import LogStore
+
+log = LogStore.add_logger('dmu:plotting:fwhm')
+# --------------------------------------------
+class FWHM:
+    '''
+    Class meant to be used to calculate Full Width at Half Maximum
+    as a Plotter1d plugin
+    '''
+    # -------------------------
+    def __init__(self, cfg : dict, val : numpy.ndarray, wgt : numpy.ndarray, maxy : float):
+        self._cfg     = cfg
+        self._arr_val = val
+        self._arr_wgt = wgt
+        self._maxy    = maxy
+    # -------------------------
+    def _normalize_yval(self, arr_pdf_val : numpy.ndarray) -> None:
+        max_pdf_val = numpy.max(arr_pdf_val)
+        arr_pdf_val*= self._maxy / max_pdf_val
+
+        return arr_pdf_val
+    # -------------------------
+    def _get_fwhm(self, arr_x : numpy.ndarray, arr_y : numpy.ndarray) -> float:
+        maxy = numpy.max(arr_y)
+        arry = numpy.where(arr_y > maxy/2.)[0]
+        imax = arry[ 0]
+        imin = arry[-1]
+
+        x1 = arr_x[imax]
+        x2 = arr_x[imin]
+
+        if self._cfg['plot']:
+            plt.plot([x1, x2], [maxy/2, maxy/2], linestyle=':', linewidth=1, color='k')
+
+        return x2 - x1
+    # -------------------------
+    def run(self) -> float:
+        '''
+        Runs plugin and return FWHM
+        '''
+        [minx, maxx] = self._cfg['obs']
+
+        log.info('Running FWHM pluggin')
+        obs = zfit.Space('mass', limits=(minx, maxx))
+        pdf= zfit.pdf.KDE1DimExact(obs=obs, data=self._arr_val, weights=self._arr_wgt)
+
+        xval = numpy.linspace(minx, maxx, 200)
+        yval = pdf.pdf(xval)
+        yval = self._normalize_yval(yval)
+
+        if self._cfg['plot']:
+            plt.plot(xval, yval, linestyle='-', linewidth=2, color='gray')
+
+        fwhm = self._get_fwhm(xval, yval)
+
+        return fwhm
+# --------------------------------------------

dmu/plotting/plotter.py

@@ -29,6 +29,8 @@ class Plotter:
         self._d_cfg = cfg
         self._d_rdf : dict[str, RDataFrame]    = { name : self._preprocess_rdf(rdf) for name, rdf in d_rdf.items()}
         self._d_wgt : Union[dict[str, Union[numpy.ndarray, None]], None]
+
+        self._title : str = ''
     #-------------------------------------
     def _check_quantile(self, qnt : float):
         '''

dmu/plotting/plotter_1d.py

@@ -1,7 +1,7 @@
 '''
 Module containing plotter class
 '''
-
+import copy
 from hist import Hist
 
 import numpy
@@ -9,6 +9,7 @@ import matplotlib.pyplot as plt
 
 from dmu.logging.log_store import LogStore
 from dmu.plotting.plotter  import Plotter
+from dmu.plotting.fwhm     import FWHM
 
 log = LogStore.add_logger('dmu:plotting:Plotter1D')
 # --------------------------------------------
@@ -55,6 +56,72 @@ class Plotter1D(Plotter):
 
         return minx, maxx, bins
     #-------------------------------------
+    def _run_plugins(self,
+                     arr_val : numpy.ndarray,
+                     arr_wgt : numpy.ndarray,
+                     hst,
+                     name    : str,
+                     varname : str) -> None:
+        if 'plugin' not in self._d_cfg:
+            log.debug('No plugins found')
+            return
+
+        if 'fwhm' in self._d_cfg['plugin']:
+            if varname not in self._d_cfg['plugin']['fwhm']:
+                log.debug(f'No FWHM plugin found for variable {varname}')
+                return
+
+            log.debug(f'FWHM plugin found for variable {varname}')
+            cfg = self._d_cfg['plugin']['fwhm'][varname]
+            self._run_fwhm(arr_val = arr_val, arr_wgt=arr_wgt, hst=hst, name=name, cfg = cfg)
+
+        if 'stats' in self._d_cfg['plugin']:
+            if varname not in self._d_cfg['plugin']['stats']:
+                log.debug(f'No stats plugin found for variable {varname}')
+                return
+
+            log.debug(f'stats plugin found for variable {varname}')
+            cfg = self._d_cfg['plugin']['stats'][varname]
+            self._run_stats(arr_val = arr_val, arr_wgt=arr_wgt, name=name, cfg = cfg)
+    #-------------------------------------
+    def _run_stats(self, arr_val : numpy.ndarray, arr_wgt : numpy.ndarray, name : str, cfg : dict[str:str]) -> None:
+        this_title = ''
+        if 'sum' in cfg:
+            form = cfg['sum']
+            sumv = numpy.sum(arr_wgt)
+            this_title += form.format(sumv) + '; '
+
+        if 'mean' in cfg:
+            form = cfg['mean']
+            mean = numpy.average(arr_val, weights=arr_wgt)
+            this_title += form.format(mean) + '; '
+
+        if 'rms'  in cfg:
+            form = cfg['rms']
+            mean = numpy.average(arr_val, weights=arr_wgt)
+            rms  = numpy.sqrt(numpy.average((arr_val - mean) ** 2, weights=arr_wgt))
+            this_title += form.format(rms ) + '; '
+
+        self._title+= f'\n{name}: {this_title}'
+    #-------------------------------------
+    def _run_fwhm(self, arr_val : numpy.ndarray, arr_wgt : numpy.ndarray, hst, name : str, cfg : dict) -> None:
+        arr_bin_cnt = hst.values()
+        maxy = numpy.max(arr_bin_cnt)
+        obj  = FWHM(cfg=cfg, val=arr_val, wgt=arr_wgt, maxy=maxy)
+        fwhm = obj.run()
+
+        form        = cfg['format']
+        this_title  = form.format(fwhm)
+
+        if 'add_std' in cfg and cfg['add_std']:
+            mu         = numpy.average(arr_val            , weights=arr_wgt)
+            avg        = numpy.average((arr_val - mu) ** 2, weights=arr_wgt)
+            std        = numpy.sqrt(avg)
+            form       = form.replace('FWHM', 'STD')
+            this_title+= '; ' + form.format(std)
+
+        self._title+= f'\n{name}: {this_title}'
+    #-------------------------------------
     def _plot_var(self, var : str) -> float:
         '''
         Will plot a variable from a dictionary of dataframes
@@ -70,6 +137,7 @@ class Plotter1D(Plotter):
 
         d_data = {}
         for name, rdf in self._d_rdf.items():
+            log.debug(f'Plotting: {var}/{name}')
             d_data[name] = rdf.AsNumpy([var])[var]
 
         minx, maxx, bins = self._get_binning(var, d_data)
@@ -82,7 +150,18 @@ class Plotter1D(Plotter):
             arr_wgt      = self._normalize_weights(arr_wgt, var)
             hst          = Hist.new.Reg(bins=bins, start=minx, stop=maxx, name='x').Weight()
             hst.fill(x=arr_val, weight=arr_wgt)
-            hst.plot(label=label)
+            self._run_plugins(arr_val, arr_wgt, hst, name, var)
+
+            if 'styling' in self._d_cfg['plots'][var]:
+                style = self._d_cfg['plots'][var]['styling']
+                style = copy.deepcopy(style)
+            else:
+                style = {'label' : label, 'histtype' : 'errorbar', 'marker' : '.', 'linestyle' : 'none'}
+
+            if 'label' not in style:
+                style['label'] = label
+
+            hst.plot(**style)
             l_bc_all    += hst.values().tolist()
 
         max_y = max(l_bc_all)
@@ -131,9 +210,12 @@ class Plotter1D(Plotter):
         if yscale == 'linear':
             plt.ylim(bottom=0)
 
-        title = ''
+        title = self._title
         if 'title'      in d_cfg:
-            title = d_cfg['title']
+            this_title = d_cfg['title']
+            title += f'\n {this_title}'
+
+        title = title.lstrip('\n')
 
         plt.ylim(top=1.2 * max_y)
         plt.legend()
@@ -160,8 +242,7 @@ class Plotter1D(Plotter):
 
         fig_size = self._get_fig_size()
         for var in self._d_cfg['plots']:
-            log.debug(f'Plotting: {var}')
-
+            self._title = ''
             plt.figure(var, figsize=fig_size)
             max_y = self._plot_var(var)
             self._style_plot(var, max_y)

dmu/stats/fitter.py

@@ -231,7 +231,7 @@ class Fitter:
                 continue
 
             const = zfit.constraint.GaussianConstraint(params=par, observation=float(par_mu), uncertainty=float(par_sg))
-            log.info(f'{"":<4}{par_name:<25}{par_mu:<15.3e}{par_sg:<15.3e}')
+            log.info(f'{"":<4}{par_name:<45}{par_mu:<15.3e}{par_sg:<15.3e}')
             l_const.append(const)
 
         return l_const