Skip to content
Permalink
Branch: master
Commits on Apr 15, 2019
  1. Move criterion to compute device automatically (#455)

    ottonemo authored and BenjaminBossan committed Apr 15, 2019
    Move criterion to compute device automatically
    
    Previously the user had to make sure that criterion parameters
    such as class weights are on the correct computing device.
    Now the criterion is moved to the compute device using `.to`
    which moves the parameters as well.
    
    Non nn.Module classes still work as criterion.
Commits on Apr 9, 2019
  1. Add button logo

    ottonemo committed Apr 9, 2019
Commits on Apr 5, 2019
Commits on Mar 19, 2019
Commits on Feb 26, 2019
  1. Workaround for misused cuda dependent attributes (#435)

    ottonemo authored and BenjaminBossan committed Feb 26, 2019
    Workaround for misused cuda dependent attributes
    
    the `cuda_dependent_attributes_` member was misused as storage in
    `__getstate__` which results in errors after re-pickling the unpickled
    model since `__getstate__` expects strings in
    `cuda_dependent_attributes_`.
Commits on Feb 21, 2019
  1. Update requirements.txt of docs (pytorch 1.0.1)

    ottonemo committed Feb 21, 2019
Commits on Feb 7, 2019
  1. Mark master as 0.5.1a0

    ottonemo committed Feb 7, 2019
Commits on Feb 5, 2019
  1. Fix unpickling issues with CUDA sub-attributes (#431)

    ottonemo authored and BenjaminBossan committed Feb 5, 2019
    * Fix unpickling issues with CUDA sub-attributes
    
    Solves issues with, for example,
    * `criterion__weight = <cuda tensor>`
    * `optimizer__state`
    * `module__some_weird_parameter = <cuda_tensor>`
    
    Before `criterion` was never considered a CUDA dependent
    attribute, this is also fixed.
    
    The old format is going to be deprecated with the next non-bugfix release.
Commits on Dec 17, 2018
  1. Satisfy pypi version linter

    ottonemo committed Dec 17, 2018
  2. Post-release p0 to fix #412

    ottonemo committed Dec 17, 2018
  3. Do not fix PyTorch version in deploy script

    ottonemo committed Dec 17, 2018
Commits on Dec 13, 2018
  1. Bump version

    ottonemo committed Dec 13, 2018
  2. Highlight some features in README (#405)

    ottonemo committed Dec 13, 2018
    * Highlight some features in README
    
    * Add CLI example
    
    * Less colloquial
  3. Fix dict ordering issue in test with py 3.5

    ottonemo committed Dec 13, 2018
  4. Attempt to merge setup dependencies (#408)

    ottonemo authored and benjamin-work committed Dec 13, 2018
    Merge setup dependencies
    
    * Use pip to install all requirements except pytorch
    * Remove strict pin of tqdm and tabulate
    * Add pip caching to travis.yml
Commits on Dec 11, 2018
  1. RFC: runtime parameter setting (e.g. LR) (#369)

    ottonemo committed Dec 11, 2018
    * RFC: runtime parameter setting (e.g. LR)
    
    * Re-model runtime parameters as virtual parameters
    
    Separation of concerns: application of virtual parameters
    is now separated from applying them.
    
    There is a prototypical implementation for a virtual
    learning rate parameter but this needs to be extended to
    match all possible optimizer parameters (and others as well).
    
    * Remove unused partial
    
    * Test comment statement explicitly
    
    * Rename initialize_params to initialize_virtual_params
    
    * Move setter to own module, add tests
    
    * Remove unused import
    
    * Document and refactor parameter name matching
    
    * Move set_optimizer_param to setter module
    
    * Satisfy pylint
    
    * Test optimizer_setter error message explicitly
    
    * Document intention of optimizer parameters in setter
    
    * Document changes
    
    * Rephrase changes
    
    * Limit number of epochs to save time
    
    * Rename function to reflect affiliation
    
    * Make dummy functions fixtures and move to class
Commits on Nov 12, 2018
  1. Move changes to appropriate section

    ottonemo committed Nov 12, 2018
Commits on Oct 26, 2018
  1. Document advanced usage notebook change

    ottonemo committed Oct 26, 2018
Commits on Oct 25, 2018
  1. Fix typo

    ottonemo committed Oct 25, 2018
  2. Fix links in CHANGES.md

    ottonemo committed Oct 25, 2018
Commits on Oct 24, 2018
  1. Bump version to 0.4.0

    ottonemo committed Oct 24, 2018
  2. Refresh coverage badge

    ottonemo committed Oct 24, 2018
  3. Unify device handling in __setstate__ and load_params (#367)

    ottonemo authored and benjamin-work committed Oct 24, 2018
    * Unify device handling in __setstate__ and load_params
    
    Before each of these functions had their own logic of
    determining a fallback device if special devices such
    as CUDA are not available. Now there is one common
    way of handling this case.
Commits on Oct 20, 2018
  1. deprecate filtered optimizer (#346)

    ottonemo committed Oct 20, 2018
    * Update transfer learning example to use Freezer
    
    Also fix a bug in data loading.
    
    * Deprecate filtered_optimizer et al.
    
    This feature was introduced for the sole purpose of filtering parameters
    with `requires_grad == False`. These don't need to be filtered anymore
    and there is `skorch.callbacks.Freezer` for parameter freezing.
    
    Therefore helpers can be removed in the future.
Commits on Oct 16, 2018
  1. Update MyDataset example in Advanced Usage

    ottonemo committed Oct 16, 2018
Commits on Sep 18, 2018
  1. Implement ParamMapper from #342 (#343)

    ottonemo authored and benjamin-work committed Sep 18, 2018
    Implements callbacks to modify module parameters at specific points during training. Main uses include initializing weights at the start of training, freezing and unfreezing of parameters.
    
    The old way required a couple of steps which weren't always intuitive and could silently fail. With the new callbacks, tasks such as transfer learning should become considerably easier.
  2. Add missing toy section in API and order topics

    ottonemo committed Sep 18, 2018
Commits on Aug 6, 2018
  1. Bump unstable version to 0.4.0

    ottonemo committed Aug 6, 2018
Commits on Aug 2, 2018
  1. On name conflict assign unique callback names (#301)

    ottonemo authored and benjamin-work committed Aug 2, 2018
    On callback name conflict assign unique callback names
    
    Callback names will be appended with a count suffix to
    make the name unique if it appears more than once.
    The suffix is of format `_{i}` with `i` starting at 1
    with underscore as a separator to be addressable using
    keywords arguments.
    
    Example:
    
        e1 = EpochScoring(name='a')
        e2 = EpochScoring(name='b')
        net = Net(callbacks=[e1, e2]).initialize()
        cbs = dict(net.callbacks_)
    
    Results in:
    
        cbs['EpochScoring_1'] == e1
        cbs['EpochScoring_2'] == e2
    
    This addresses #294.
    
    Makes sure that user-set names remain untouched.
    
    Also avoids conflicts with unique names that already exist for
    some reason (assuming user-set names).
Commits on Jul 26, 2018
  1. Release 0.3.0

    ottonemo committed Jul 26, 2018
  2. Update links in README

    ottonemo committed Jul 26, 2018
  3. Fix doc string for grad/filter helper

    ottonemo committed Jul 26, 2018
Commits on Jun 13, 2018
  1. Merge branch 'master' of https://github.com/dnouri/skorch

    ottonemo committed Jun 13, 2018
Older
You can’t perform that action at this time.