augmentrum 0.0.1__py3-none-any.whl

augmentrum/sampling/subject_splitter.py

@@ -0,0 +1,131 @@
+####################################################################################################
+#                                      subject_splitter.py                                         #
+####################################################################################################
+#                                                                                                  #
+# Authors: J. P. Merkofer (j.p.merkofer@tue.nl)                                                    #
+#                                                                                                  #
+# Created: 2025-10-07                                                                              #
+#                                                                                                  #
+# Purpose: Implements SubjectSplitter for reproducible subject-wise train/val/test splits.         #
+#          Supports stratified or random splitting modes.                                          #
+#                                                                                                  #
+####################################################################################################
+
+
+#*************#
+#   imports   #
+#*************#
+import torch
+
+
+#**************************************************************************************************#
+#                                          Class SubjectSampler                                    #
+#**************************************************************************************************#
+#                                                                                                  #
+# Simplistic subject splitter for train/val/test splits. Supports random splitting.                #
+#                                                                                                  #
+#**************************************************************************************************#
+class SubjectSplitter:
+    """
+    Splits subjects into train/val/test sets.
+
+    Works with any backend (NIfTI_MRS_Plus, lists, or raw data).
+    Returns splits in the same format as input.
+    """
+
+    def __init__(self, data, water=None, seed=0, val_frac=0.1, test_frac=0.1):
+        """
+        Initialization.
+
+        Args:
+            data: Input data (NIfTI_MRS_Plus, list, or other).
+            water: Water reference data (same format as data), optional.
+            seed (int): Random seed for reproducibility.
+            val_frac (float): Fraction of data for validation set.
+            test_frac (float): Fraction of data for test set.
+        """
+        self.data = data
+        self.water = water
+        self.val_frac = val_frac
+        self.test_frac = test_frac
+        self.seed = seed
+        torch.manual_seed(seed)
+
+    def __call__(self, **kwargs):
+        """
+        Perform the split and return the dictionary of splits.
+        """
+        return self.split()
+
+    def split(self):
+        """
+        Splits subjects into train/val/test sets.
+
+        Handles NIfTI_MRS_Plus with any backend by converting to list,
+        splitting, and optionally converting back.
+
+        Returns:
+            dict: Dictionary with 'train', 'val', and 'test' keys.
+                  Each value is a tuple of (data, water).
+        """
+        # Import here to avoid circular imports
+        from augmentrum.core import NIfTI_MRS_Plus
+
+        # Store original type and backend for later reconstruction
+        is_nifti_plus = isinstance(self.data, NIfTI_MRS_Plus)
+        original_backend = self.data.backend if is_nifti_plus else None
+        original_volatile = self.data.volatile if is_nifti_plus else False
+
+        # Convert to list for splitting
+        if is_nifti_plus:
+            data_list = self.data.to_nifti_list()
+        else:
+            data_list = self.data if isinstance(self.data, list) else list(self.data)
+
+        if self.water is not None:
+            if isinstance(self.water, NIfTI_MRS_Plus):
+                water_list = self.water.to_nifti_list()
+            else:
+                water_list = self.water if isinstance(self.water, list) else list(self.water)
+            
+            # Validate that water_list and data_list have the same length
+            if len(water_list) != len(data_list):
+                raise ValueError(
+                    f"Mismatch between data and water lists: "
+                    f"data has {len(data_list)} subjects but water has {len(water_list)} subjects. "
+                    f"They must have the same length."
+                )
+        else:
+            water_list = None
+
+        # Perform random split
+        idxs = torch.randperm(len(data_list)).tolist()
+        n_total = len(idxs)
+        n_test = int(n_total * self.test_frac)
+        n_val = int(n_total * self.val_frac)
+
+        def get_lists(sel):
+            data_sel = [data_list[i] for i in sel]
+            water_sel = [water_list[i] for i in sel] if water_list is not None else None
+
+            # Optionally convert back to NIfTI_MRS_Plus with original backend
+            if is_nifti_plus and len(data_sel) > 0:
+                data_sel = NIfTI_MRS_Plus(
+                    nifti_list=data_sel,
+                    backend=original_backend,
+                    volatile=original_volatile
+                )
+                if water_sel is not None:
+                    water_sel = NIfTI_MRS_Plus(
+                        nifti_list=water_sel,
+                        backend=original_backend,
+                        volatile=original_volatile
+                    )
+
+            return data_sel, water_sel
+
+        return {
+            'train': get_lists(idxs[n_test + n_val:]),
+            'val': get_lists(idxs[n_test:n_test + n_val]),
+            'test': get_lists(idxs[:n_test]),
+        }

augmentrum/utils/__init__.py

@@ -0,0 +1,10 @@
+"""
+Utility modules for Augmentrum.
+"""
+
+from augmentrum.utils.tensor_ops import (  # noqa: F401
+    fft, ifft, fftshift, ifftshift, to_numpy, match_backend,
+)
+
+__all__ = ['fft', 'ifft', 'fftshift', 'ifftshift', 'to_numpy', 'match_backend']
+

augmentrum/utils/philips.py

@@ -0,0 +1,90 @@
+####################################################################################################
+#                                            philips.py                                            #
+####################################################################################################
+#                                                                                                  #
+# Authors: J. P. Merkofer (j.p.merkofer@tue.nl)                                                    #
+#                                                                                                  #
+# Created: 20/06/22                                                                                #
+#                                                                                                  #
+# Purpose: Contains various Philips specific utils.                                                #
+#                                                                                                  #
+####################################################################################################
+
+
+#*************#
+#   imports   #
+#*************#
+import numpy as np
+
+
+#****************#
+#   definition   #
+#****************#
+index_options = ['STD', 'REJ', 'PHC', 'FRC', 'NOI', 'NAV']
+
+index_headers = ['typ', 'mix', 'dyn', 'card', 'echo', 'loca', 'chan',
+                 'extr1', 'extr2', 'kx', 'ky', 'kz', 'aver', 'sign',
+                 'rf', 'grad', 'enc', 'rtop', 'rr', 'size', 'offset']
+
+indices = ['chan', 'aver', 'dyn', 'card', 'echo', 'loca', 'extr1', 'extr2']
+
+defaults = {'chan': 'DIM_COIL',
+            'aver': 'DIM_DYN',
+            'dyn': 'DIM_DYN',
+            'echo': 'DIM_INDIRECT_0'}
+
+
+#***********************#
+#   read a .data file   #
+#***********************#
+def read_Philips_data(data_file, df):
+    """
+    Reads a .data file with given .list data frame. Parts of the code are taken from:
+    https://github.com/wtclarke/spec2nii/blob/master/spec2nii/philips_data_list.py
+
+    @param data_file -- The path to the data file.
+    @param df -- The data frame obtained from the .list file.
+
+    @returns -- A dict containing the data.
+    """
+    with open(data_file, 'rb') as f:
+        raw = np.fromfile(f, dtype='<f4')
+        raw = raw[0::2] + 1j * raw[1::2]
+
+    data_types = df.typ.unique()
+    output_dict = {}
+    for tt in data_types:
+        curr_df = df.loc[df.typ == tt, :]
+        if tt == 'NOI':
+            # special simple case for NOI
+            spec_res = int(curr_df['size'].max() / 8)
+            ncha = curr_df['chan'].max() + 1
+            nloca = curr_df['loca'].max() + 1
+            output_dict[tt] = np.zeros((spec_res, ncha, nloca), dtype=np.complex128)
+        else:
+            n_mix = curr_df['mix'].max() + 1
+            # other types might use all the loops
+            for mix in range(n_mix):
+                curr_mix_df = curr_df.loc[curr_df.mix == mix, :]
+                shape = []
+                shape.append(int(curr_mix_df['size'].max() / 8))
+                for ind in indices:
+                    shape.append(curr_mix_df[ind].max() + 1)
+                output_dict[f'{tt}_{mix}'] = np.zeros(shape, dtype=np.complex128)
+
+    # now extract data
+    offset = 0
+    for ind in range(df.shape[0]):
+        cdf = df.iloc[ind]
+        tt = cdf.typ
+        dsize = int(cdf['size'].max() / 8)
+        if tt == 'NOI':
+            output_dict[tt][:, cdf.chan, cdf.loca] = raw[offset:(offset + dsize)]
+        else:
+            mix = cdf.mix
+            ind = [cdf[ii] for ii in indices]
+            ind = tuple([slice(None), ] + ind)
+            output_dict[f'{tt}_{mix}'][ind] = raw[offset:(offset + dsize)]
+        offset += dsize
+
+    return output_dict