tonic 1.4.4.dev21__tar.gz → 1.4.4.dev40__tar.gz

{tonic-1.4.4.dev21 → tonic-1.4.4.dev40}/AUTHORS

@@ -2,6 +2,7 @@ Anindya Ghosh <anindya128@yahoo.com>
 Antoine Grimaldi <antoine.grimaldi@univ-amu.fr>
 AntoineGrimaldi <antoine.grimaldi@univ-amu.fr>
 Carsten Nielsen <carstenlau.nielsen@uzh.ch>
+Dhia Zerrari <dhia.zerrari@outlook.com>
 Elias Martinsson <131291250+e-martinsson@users.noreply.github.com>
 Fabrizio Ottati <45739782+fabhertz95@users.noreply.github.com>
 Fabrizio Ottati <fabriziottati@gmail.com>

{tonic-1.4.4.dev21 → tonic-1.4.4.dev40}/ChangeLog

@@ -1,6 +1,15 @@
 CHANGES
 =======
 
+* pin numpy version <2
+* adding numpy to test requirements
+* change np.product to np.prod
+* fix issue in time surface generation where pixels without events would still result in activity
+* Fix file loading order of N-MNIST samples
+* DataSet class import was removed, the mini dataset does not inherent from that
+* moving torchvision import to inside function
+* moving typing-extensions to the root requirement
+* moving torch import to inside function, where it is used
 * pin torchvision version to something compatible with torch 2.1
 * add torch requirements to documentation github action
 * shorten GH actions pipeline to three Python versions
@@ -12,6 +21,10 @@ CHANGES
 * Revert "fix path format on windows for test case for ThreeET\_Eyetracking"
 * fix path format on windows for test case for ThreeET\_Eyetracking
 * add test case for ThreeET\_Eyetracking
+* notebook added to elaborate teh function of AugDiskCachedDataset
+* test added for aug\_diskcached
+* including Aug\_DiskCachedDataset in init imports
+* bug fixed in Aug\_DiskCach
 * add ThreeET\_Eyetracking to the documentation
 * fix: check for empty array in DropPixel transform
 * added: tests for empty frames
@@ -32,6 +45,9 @@ v1.4.3
 * sample\_length was removed from some transforms (when not needed)
 * Fix: silent frames is now supported in uniform noise
 * Fix: silent frame support
+* importing TypedDict from typing\_extensions for python 3.7
+* adding typing-extensions to requirments
+* small fixes and cleanup
 * noise related augmentations removed
 * Fix: wrong range of histogram in drop\_pixel
 * Testing with python>=3.8 and python<=3.11
@@ -51,11 +67,13 @@ v1.4.3
 * adding torchaudio
 * removing conflicting requirements
 * updated the required version of torch
+* dict keys updated to more generic
 * RIR test updated
 * removing hard coded room impulse audio from RIR transform
 * requirments updated with torch and torchaudio versions
 * removing torchaudio dependency temporarily to run the tests
 * torchaudio added to requirements
+* AugDiskCachedDataset added
 * tests\_passed
 * tests for audio augmentations
 * tests for added transforms

{tonic-1.4.4.dev21 → tonic-1.4.4.dev40}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: tonic
-Version: 1.4.4.dev21
+Version: 1.4.4.dev40
 Summary: Neuromorphic datasets and transformations.
 Home-page: UNKNOWN
 Author: The Neuromorphs of Telluride

tonic-1.4.4.dev40/docs/tutorials/Aug_DiskCachDataset.ipynb

@@ -0,0 +1,179 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Using Aug_DiskCachedDataset for efficient caching of augmented copies\n",
+    "-  `Aug_DiskCachedDataset` is a modified version of `DiskCachedDataset` that is useful while applying deterministic augmentations on data samples. \n",
+    "\n",
+    "- This is the case when the parameter space of augmentation is desceret, for instance applying     `pitchshift` on audio data in which shift parameter (semitone) can only take N values.\n",
+    "\n",
+    "- Using `DiskCachedDataset` and setting `num_copies` to N is likely to cause 2 issues:\n",
+    "\n",
+    "    - Copies might not be unique, as copy_index is not linked to the augmentation parameter \n",
+    "    - And there is no guarantee that copies cover the desired augmentation space\n",
+    "    \n",
+    "\n",
+    "\n",
+    "- `Aug_DiskCachedDataset` resolves this limitation by mapping and linking copy index to augmentation parameter. Following considerations need to be takes into account:\n",
+    "\n",
+    "    - The user needs to pass `all_transforms` dict as input with seperated transforms `pre_aug`,  `aug`,  `post_aug` (spesifying transforms that are applied before and after augmentations, also augmentation transforms). \n",
+    "    \n",
+    "    - The augmentation class receives `aug_index` (aug_index = copy) as initialization parameter also `caching=True` needs to be set (please see `tonic.audio_augmentations`)\n",
+    "\n",
+    "- Follwing is a simple example to show function of `Aug_DiskCachedDataset`"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### A simple dataset "
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# %%writefile mini_dataset.py\n",
+    "import warnings\n",
+    "warnings.filterwarnings('ignore')\n",
+    "from torch.utils.data import Dataset\n",
+    "import numpy as np\n",
+    "\n",
+    "class mini_dataset(Dataset):\n",
+    "    def __init__(self) -> None:\n",
+    "        super().__init__()\n",
+    "        np.random.seed(0)\n",
+    "        self.data = np.random.rand(10, 16000)\n",
+    "        self.transform = None\n",
+    "        self.target_transform = None\n",
+    "\n",
+    "    def __getitem__(self, index):\n",
+    "        sample = self.data[index]\n",
+    "        label =  1\n",
+    "        if sample.ndim==1:\n",
+    "            sample = sample[None,...]\n",
+    "        if self.transform is not None:\n",
+    "            sample = self.transform(sample)\n",
+    "        if self.target_transform is not None:\n",
+    "            label = self.target_transform(label)  \n",
+    "\n",
+    "        return sample, label      "
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Initializing `Aug_DiskCachedDataset` with transforms"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from tonic.cached_dataset import Aug_DiskCachedDataset, load_from_disk_cache\n",
+    "from tonic.audio_transforms import AmplitudeScale, FixLength\n",
+    "from tonic.audio_augmentations import RandomPitchShift\n",
+    "\n",
+    "all_transforms = {}\n",
+    "all_transforms[\"pre_aug\"] = [AmplitudeScale(max_amplitude = 0.150)]\n",
+    "all_transforms[\"augmentations\"] = [RandomPitchShift(samplerate=16000, caching=True)]\n",
+    "all_transforms[\"post_aug\"] = [FixLength(16000)]\n",
+    "\n",
+    "# number of copies is set to number of augmentation params (factors)\n",
+    "n = len(RandomPitchShift(samplerate=16000, caching=True).factors)\n",
+    "Aug_cach = Aug_DiskCachedDataset(dataset=mini_dataset(), cache_path='cache/', all_transforms = all_transforms, num_copies=n)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Generating all copies of a data sample\n",
+    "    - 10 augmented versions of data sample with index = 0 are generated"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "sample_index = 0\n",
+    "Aug_cach.generate_all(sample_index)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### To verify\n",
+    "    - loading the saved copies \n",
+    "    - and comparing them with the ones generated out of `Aug_DiskCacheDataset` with the same transforms and matching augmentation parameter \n",
+    "    - they are equal\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 7,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "True\n",
+      "True\n",
+      "True\n",
+      "True\n",
+      "True\n",
+      "True\n",
+      "True\n",
+      "True\n",
+      "True\n",
+      "True\n"
+     ]
+    }
+   ],
+   "source": [
+    "from torchvision.transforms import Compose\n",
+    "\n",
+    "for i in range(n):\n",
+    "    transform = Compose([AmplitudeScale(max_amplitude = 0.150),RandomPitchShift(samplerate=16000, caching=True, aug_index=i), FixLength(16000)])\n",
+    "    ds = mini_dataset()\n",
+    "    ds.transform = transform\n",
+    "    sample = ds[sample_index][0]\n",
+    "    data, targets = load_from_disk_cache('cache/' + '0_' + str(i) + '.hdf5' )\n",
+    "    print((sample==data).all())\n"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "py_310",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.10.2"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}

{tonic-1.4.4.dev21 → tonic-1.4.4.dev40}/requirements.txt

@@ -1,4 +1,4 @@
-numpy
+numpy<2.0.0
 h5py
 importRosbag>=1.0.4
 scipy
@@ -7,3 +7,4 @@ typing_extensions
 librosa
 pbr
 expelliarmus
+typing-extensions

{tonic-1.4.4.dev21 → tonic-1.4.4.dev40}/test/requirements.txt

@@ -3,4 +3,4 @@ coverage
 matplotlib
 hdf5plugin
 imageio
-aedat
+aedat

tonic-1.4.4.dev40/test/test_aug_caching.py

@@ -0,0 +1,66 @@
+import os
+
+import numpy as np
+
+from tonic.audio_augmentations import RandomPitchShift
+from tonic.audio_transforms import AmplitudeScale, FixLength
+from tonic.cached_dataset import Aug_DiskCachedDataset, load_from_disk_cache
+
+
+class mini_dataset:
+    def __init__(self) -> None:
+        np.random.seed(0)
+        self.data = np.random.rand(10, 16000)
+        self.transform = None
+        self.target_transform = None
+
+    def __getitem__(self, index):
+        sample = self.data[index]
+        label = 1
+        if sample.ndim == 1:
+            sample = sample[None, ...]
+        if self.transform is not None:
+            sample = self.transform(sample)
+        if self.target_transform is not None:
+            label = self.target_transform(label)
+
+        return sample, label
+
+
+def test_aug_disk_caching():
+    from torchvision.transforms import Compose
+
+    all_transforms = {}
+    all_transforms["pre_aug"] = [AmplitudeScale(max_amplitude=0.150)]
+    all_transforms["augmentations"] = [RandomPitchShift(samplerate=16000, caching=True)]
+    all_transforms["post_aug"] = [FixLength(16000)]
+    # number of copies is set to number of augmentation params (factors)
+    n = len(RandomPitchShift(samplerate=16000, caching=True).factors)
+    Aug_cach = Aug_DiskCachedDataset(
+        dataset=mini_dataset(),
+        cache_path="cache/",
+        all_transforms=all_transforms,
+        num_copies=n,
+    )
+
+    if not os.path.isdir("cache/"):
+        os.mkdir("cache/")
+
+    sample_index = 0
+    Aug_cach.generate_all(sample_index)
+
+    for i in range(n):
+        transform = Compose(
+            [
+                AmplitudeScale(max_amplitude=0.150),
+                RandomPitchShift(samplerate=16000, caching=True, aug_index=i),
+                FixLength(16000),
+            ]
+        )
+        ds = mini_dataset()
+        ds.transform = transform
+        augmented_sample = ds[sample_index][0]
+        loaded_sample, targets = load_from_disk_cache(
+            "cache/" + "0_" + str(i) + ".hdf5"
+        )
+        assert (augmented_sample == loaded_sample).all()

{tonic-1.4.4.dev21 → tonic-1.4.4.dev40}/test/test_tonic_utils.py

@@ -47,13 +47,13 @@ def test_pytorch_batch_collation_dense_tensor():
     assert batch.shape[2] == sensor_size[2]
 
 
-@pytest.mark.skipif(not platform == "linux", reason="Test doesn't finish on Windows.")
+@pytest.mark.skipif(platform == "win32", reason="Test doesn't finish on Windows.")
 def test_plotting():
     events, sensor_size = create_random_input()
     tonic.utils.plot_event_grid(events)
 
 
-@pytest.mark.skipif(not platform == "linux", reason="Test doesn't finish on Windows.")
+@pytest.mark.skipif(platform == "win32", reason="Test doesn't finish on Windows.")
 def test_animation():
     events, sensor_size = create_random_input()
 

{tonic-1.4.4.dev21 → tonic-1.4.4.dev40}/tonic/__init__.py

@@ -1,7 +1,12 @@
 from pbr.version import VersionInfo
 
 from . import collation, datasets, io, slicers, transforms, utils
-from .cached_dataset import CachedDataset, DiskCachedDataset, MemoryCachedDataset
+from .cached_dataset import (
+    Aug_DiskCachedDataset,
+    CachedDataset,
+    DiskCachedDataset,
+    MemoryCachedDataset,
+)
 from .dataset import Dataset
 from .sliced_dataset import SlicedDataset
 

{tonic-1.4.4.dev21 → tonic-1.4.4.dev40}/tonic/cached_dataset.py

@@ -1,9 +1,17 @@
 import logging
 import os
+import sys
+
+if sys.version_info >= (3, 8):
+    from typing import Callable, Iterable, Optional, Tuple, TypedDict, Union
+else:
+    from typing import Callable, Iterable, Optional, Tuple, Union
+    from typing_extensions import TypedDict
+
+import random
 import shutil
 from dataclasses import dataclass, field
 from pathlib import Path
-from typing import Callable, Iterable, Optional, Tuple, Union
 from warnings import warn
 
 import h5py
@@ -227,6 +235,83 @@ def load_from_disk_cache(file_path: Union[str, Path]) -> Tuple:
     return data_list, target_list
 
 
+@dataclass
+class Aug_DiskCachedDataset(DiskCachedDataset):
+    """Aug_DiskCachedDataset is a child class from DiskCachedDataset with further customizations to
+    handle augmented copies of a sample. The goal of this customization is to map the indices of
+    cached files (copy) to augmentation parameters. This is useful in a category of augmentations
+    where the range of parameter is rather disceret and non probabilistic, for instance an audio
+    sample is being augmented with noise and SNR can take only N=5 values. Passing copy_index to
+    augmentation Class as an init argument ensures that each copy will be a a distinct augmented
+    sample with a trackable parameter.
+
+    'generate_all' method generates all augmented vesions of a sample.
+    'generate_copy' method generates the missing variant (augmented version)
+
+
+
+    Therefore all transforms applied to the dataset are categorized by the keys:  "pre_aug", "augmentations"
+     and "post_aug".
+
+     Args:
+         'all_transforms' is a dictionarty passed to this class containing information about all transforms.
+    """
+
+    all_transforms: Optional[TypedDict] = None
+
+    def __post_init__(self):
+        self.pre_aug = self.all_transforms["pre_aug"]
+        self.aug = self.all_transforms["augmentations"]
+        self.post_aug = self.all_transforms["post_aug"]
+
+    def generate_all(self, item):
+        for copy in range(0, self.num_copies):
+            file_path = os.path.join(self.cache_path, f"{item}_{copy}.hdf5")
+            try:
+                data, targets = load_from_disk_cache(file_path)
+            except (FileNotFoundError, OSError) as _:
+                self.generate_copy(item, copy)
+
+    def generate_copy(self, item, copy):
+        from torchvision.transforms import Compose
+
+        file_path = os.path.join(self.cache_path, f"{item}_{copy}.hdf5")
+        # copy index is passed to augmentation (callable)
+        self.aug[0].aug_index = copy
+        augmentation = self.aug
+        self.dataset.transform = Compose(self.pre_aug + augmentation + self.post_aug)
+        data, targets = self.dataset[item]
+        save_to_disk_cache(data, targets, file_path=file_path, compress=self.compress)
+
+    def __getitem__(self, item) -> Tuple[object, object]:
+        if self.dataset is None and item >= self.n_samples:
+            raise IndexError(f"This dataset only has {self.n_samples} items.")
+
+        copy = random.randint(0, self.num_copies - 1)
+        file_path = os.path.join(self.cache_path, f"{item}_{copy}.hdf5")
+        try:
+            data, targets = load_from_disk_cache(file_path)
+
+        except (FileNotFoundError, OSError) as _:
+            logging.info(
+                f"Data {item}: {file_path} not in cache, generating it now",
+                stacklevel=2,
+            )
+            self.generate_copy(item, copy)
+
+            # format might change during save to hdf5, i.e. tensors -> np arrays
+            # We load the sample here again to keep the output format consistent.
+            data, targets = load_from_disk_cache(file_path)
+
+        if self.transform is not None:
+            data = self.transform(data)
+        if self.target_transform is not None:
+            targets = self.target_transform(targets)
+        if self.transforms is not None:
+            data, targets = self.transforms(data, targets)
+        return data, targets
+
+
 class CachedDataset(DiskCachedDataset):
     """Deprecated class that points to DiskCachedDataset for now but will be removed in a future
     release.

{tonic-1.4.4.dev21 → tonic-1.4.4.dev40}/tonic/datasets/nmnist.py

@@ -97,6 +97,7 @@ class NMNIST(Dataset):
 
         file_path = os.path.join(self.location_on_system, self.folder_name)
         for path, dirs, files in os.walk(file_path):
+            dirs.sort()
             files.sort()
             for file in files:
                 if file.endswith("bin"):

{tonic-1.4.4.dev21 → tonic-1.4.4.dev40}/tonic/functional/to_timesurface.py

@@ -33,7 +33,7 @@ def to_timesurface_numpy(
     event_slices = slice_events_by_time(
         events, time_window=dt, overlap=overlap, include_incomplete=include_incomplete
     )
-    memory = np.zeros((sensor_size[::-1]), dtype=int)
+    memory = np.ones((sensor_size[::-1]), dtype=int) * -np.inf
     all_surfaces = []
     x_index = event_slices[0].dtype.names.index("x")
     y_index = event_slices[0].dtype.names.index("y")

{tonic-1.4.4.dev21 → tonic-1.4.4.dev40}/tonic/utils.py

@@ -42,7 +42,7 @@ def plot_event_grid(
         sensor_size = (sensor_size_x, sensor_size_y, sensor_size_p)
 
         transform = transforms.ToFrame(
-            sensor_size=sensor_size, n_time_bins=np.product(axis_array)
+            sensor_size=sensor_size, n_time_bins=np.prod(axis_array)
         )
 
         frames = transform(events)

{tonic-1.4.4.dev21 → tonic-1.4.4.dev40}/tonic.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: tonic
-Version: 1.4.4.dev21
+Version: 1.4.4.dev40
 Summary: Neuromorphic datasets and transformations.
 Home-page: UNKNOWN
 Author: The Neuromorphs of Telluride

{tonic-1.4.4.dev21 → tonic-1.4.4.dev40}/tonic.egg-info/SOURCES.txt

@@ -77,6 +77,7 @@ docs/reading_material/intro-event-cameras.rst
 docs/reading_material/intro-snns.rst
 docs/reading_material/reading_material.rst
 docs/reading_material/training-snns.rst
+docs/tutorials/Aug_DiskCachDataset.ipynb
 docs/tutorials/audio_transforms_tutorial.ipynb
 docs/tutorials/batching.ipynb
 docs/tutorials/davis_data.ipynb
@@ -89,6 +90,7 @@ test/prototype_dataset_utils.py
 test/requirements.txt
 test/test_audio_augmentations.py
 test/test_audio_transforms.py
+test/test_aug_caching.py
 test/test_caching.py
 test/test_chained_transforms.py
 test/test_datasets.py

tonic-1.4.4.dev40/tonic.egg-info/pbr.json

@@ -0,0 +1 @@
+{"git_version": "3326802", "is_release": false}

{tonic-1.4.4.dev21 → tonic-1.4.4.dev40}/tonic.egg-info/requires.txt

@@ -1,4 +1,4 @@
-numpy
+numpy<2.0.0
 h5py
 importRosbag>=1.0.4
 scipy
@@ -7,3 +7,4 @@ typing_extensions
 librosa
 pbr
 expelliarmus
+typing-extensions

tonic-1.4.4.dev21/tonic.egg-info/pbr.json

@@ -1 +0,0 @@
-{"git_version": "5a20a54", "is_release": false}