version 0.2.22

.github/workflows/test-package-conda.yml

@@ -1,6 +1,6 @@
 name: Tests
 
-on: [push]
+on: [push, pull_request]
 
 jobs:
   build-linux:
@@ -9,29 +9,31 @@ jobs:
       max-parallel: 5
       matrix:
         python-version:
-        - "3.7"
+        # - "3.7"  # conda takes forever to install the dependencies
         - "3.8"
         - "3.9"
-
+        - "3.10"
+        - "3.11"
     steps:
     - uses: actions/checkout@v2
-    - name: Set up Python
-      uses: actions/setup-python@v2
+    - uses: conda-incubator/setup-miniconda@v2
       with:
         python-version: ${{ matrix.python-version }}
-    - name: Add conda to system path
-      run: |
-        # $CONDA is an environment variable pointing to the root of the miniconda directory
-        echo $CONDA/bin >> $GITHUB_PATH
+        channels: conda-forge
+    - name: Conda info
+      # the shell setting is necessary for loading profile etc which activates the conda environment
+      shell: bash -el {0}
+      run: conda info
     - name: Install dependencies
+      shell: bash -el {0}
       run: |
-        # conda env update --file environment.yml --name base
-        conda config --add channels conda-forge
         conda install \
           boltons \
           cython \
           deprecation \
           dill \
+          diskcache \
+          h5py \
           imageio \
           natsort \
           numba \
@@ -40,6 +42,7 @@ jobs:
           pandas \
           piexif \
           pillow \
+          pip \
           pkg-config \
           pytorch \
           requests \
@@ -48,23 +51,19 @@ jobs:
           scipy \
           setuptools \
           sphinx \
-          theano \
+          torchvision \
           tqdm
-        pip install h5py  # https://github.com/h5py/h5py/issues/1880
-#    - name: Lint with flake8
-#      run: |
-#        conda install flake8
-#        # stop the build if there are Python syntax errors or undefined names
-#        flake8 . --count --select=E9,F63,F7,F82 --show-source --statistics
-#        # exit-zero treats all errors as warnings. The GitHub editor is 127 chars wide
-#        flake8 . --count --exit-zero --max-complexity=10 --max-line-length=127 --statistics
+    - name: Conda list
+      shell: bash -el {0}
+      run: conda list
     - name: Test with pytest
+      shell: bash -el {0}
       run: |
         conda install pytest hypothesis
-
         python setup.py build_ext --inplace
-        python -m pytest --nomatlab tests
+        python -m pytest --nomatlab --notheano --nodownload tests
     - name: test build and install
+      shell: bash -el {0}
       run: |
         python setup.py sdist
         pip install dist/*.tar.gz

CHANGELOG.md

@@ -1,6 +1,32 @@
 # Changelog
 
-* 0.2.22 (dev):
+* 0.2.22:
+  * Enhancement: New [Tutorial](notebooks/Tutorial.ipynb).
+  * Bugfix: `SaliencyMapModel.AUC` failed if some images didn't have any fixations.
+  * Feature: `StimulusDependentSaliencyMapModel`
+  * Bugfix: The NUSEF dataset scaled some fixations not correctly to image coordinates. Also, we now account for some typos in the
+    dataset source data.
+  * Feature: CrossvalMultipleRegularizations and GeneralMixtureKernelDensityEstimator in baseline utils (names might change!)
+  * Feature: DVAAwareScanpathModel
+  * Feature: ShuffledBaselineModel is now much more efficient and able to handle large numbers of stimuli.
+    hence, ShuffledSimpleBaselineModel is not necessary anymore and a deprecated alias to ShuffledBaselineModel
+  * Feature: ShuffledBaselineModel can now compute predictions for very large numbers of stimuli without needing
+    to have all individual predictions in memory due to a recursive reduce logsumexp implementation.
+  * Feature: `plotting.plot_scanpath` to visualize scanpaths and saccades. WIP, expect the API to change!
+  * Feature: DeepGaze I and DeepGazeIIE models
+  * Feature: COCO Freeview dataset
+  * Feature: `optimize_for_information_gain(framework='torch', ...) now supports a `cache_directory`,
+    where intermediate steps are cached. This supports resuming crashed optimization runs.
+  * Bugfix: fixed some edge cases in `optimize_for_information_gain(framework='torch')`
+  * Feature: COCO Seach18 dataset
+  * Feature: `FixationTrains.train_lengths`
+  * Feature: `FixationTrains.scanpath_fixation_attributes` allows handling of per-fixation attributes on scanpath level,
+    e.g. fixation durations. According attributes as in a Fixations instance are automatically created,
+    e.g. for durations there will be an attribute `durations` and an attribute `duration_hist`. Also
+    for scanpath_attributes (e.g., attributes applying to a whole scanpath, such as task) will also generate
+    an attribute for each fixation to make this information available in Fixations instance.
+  * Feature: `scanpaths_from_fixations` reconstructs a FixationTrains object from a Fixations instance
+  * Bugfix: `t_hist` got replaced with `y_hist` in Fixations instances (but luckily not in FixationTrains instances)
   * Bugfix: torch code was broken due to changes in torch 1.11
   * Bugfix: SALICON dataset download did not work anymore
   * Bugfix: NUSEF datast links changed

README.md

@@ -11,37 +11,6 @@ Pysaliency can evaluate most commonly used saliency metrics, including AUC, sAUC
 image-based KL divergence, fixation based KL divergence and SIM for saliency map models and
 log likelihoods and information gain for probabilistic models.
 
-Pysaliency provides several important datasets:
-
-* MIT1003
-* MIT300
-* CAT2000
-* Toronto
-* Koehler
-* iSUN
-* SALICON (both the 2015 and the 2017 edition and each with both the original mouse traces and the inferred fixations)
-* FIGRIM
-* OSIE
-* NUSEF (the part with public images)
-
-and some influential models:
-* AIM
-* SUN
-* ContextAwareSaliency
-* BMS
-* GBVS
-* GBVSIttiKoch
-* Judd
-* IttiKoch
-* RARE2012
-* CovSal
-
-
-These models are using the original code which is often matlab.
-Therefore, a matlab licence is required to make use of these models, although quite some of them
-work with octave, too (see below).
-
-
 Installation
 ------------
 
@@ -54,7 +23,7 @@ Quickstart
 ----------
 
     import pysaliency
-    
+
     dataset_location = 'datasets'
     model_location = 'models'
 
@@ -72,6 +41,40 @@ If you already have saliency maps for some dataset, you can import them into pys
     my_model = pysaliency.SaliencyMapModelFromDirectory(mit_stimuli, '/path/to/my/saliency_maps')
     auc = my_model.AUC(mit_stimuli, mit_fixations)
 
+Check out the [Tutorial](notebooks/Tutorial.ipynb) for a more detailed introduction!
+
+Included datasets and models
+----------------------------
+
+Pysaliency provides several important datasets:
+
+* MIT1003
+* MIT300
+* CAT2000
+* Toronto
+* Koehler
+* iSUN
+* SALICON (both the 2015 and the 2017 edition and each with both the original mouse traces and the inferred fixations)
+* FIGRIM
+* OSIE
+* NUSEF (the part with public images)
+
+and some influential models:
+* AIM
+* SUN
+* ContextAwareSaliency
+* BMS
+* GBVS
+* GBVSIttiKoch
+* Judd
+* IttiKoch
+* RARE2012
+* CovSal
+
+These models are using the original code which is often matlab.
+Therefore, a matlab licence is required to make use of these models, although quite some of them
+work with octave, too (see below).
+
 
 Using Octave
 ------------