1. 16 10月, 2021 1 次提交
    • N
      text: add checkbox to control server-side markdown conversion (#5378) · e34e9e4d
      Nick Felt 提交于
      * define plugin_util.safe_html() with no markdown interpretation
      
      * make plugin_util.safe_html() handle unicode vs bytes clearly
      
      * add markdown=false request parameter to disable markdown interpretation
      
      * fix bytes to str conversion bug in text plugin no-markdown codepath
      
      * text: add checkbox to control server-side markdown conversion
      
      * yarn fix-lint
      e34e9e4d
  2. 15 10月, 2021 2 次提交
  3. 14 10月, 2021 5 次提交
  4. 13 10月, 2021 2 次提交
  5. 09 10月, 2021 1 次提交
    • S
      line chart: fix fixedViewBox flow (#5367) · 65371efb
      Stephan Lee 提交于
      fixedViewBox is way for parent to set the viewBox so it is not computed
      from the data. This is approrpriate for some use cases like PR curve
      where the axis should always span [0, 1].
      
      Previously, it was broken when there was a resize on the window or when
      data series changed which was not an intended behavior. This change
      fixes that.
      65371efb
  6. 08 10月, 2021 2 次提交
    • B
      webapp: fix tab breakage (#5365) · 4d96664b
      Brandon Yanofsky 提交于
      Forward compatible change to preemptively fix breakage that will occur
      when https://github.com/angular/components/pull/23672 is submitted.
      Because `mat-tab-header` will no longer be a direct child of
      `mat-tab-group`, the `mat-tab-list` element must be given a height of
      100% to ensure the tab group height does not collapse.
      4d96664b
    • Y
      histogram: implement buckets_v3 (#5356) · 42368f5b
      Yating 提交于
      * rename functions to avoid confusion
      
        Use `single_value` here, `singular` has unrelated mathematical meaning:
        https://en.wikipedia.org/wiki/Singularity_(mathematics).
      
      * add v3 implementation for single value input
      
      * check if bucket_count <= 0 and fix tf ops
      
      * Make SummaryV3OpGraphTest inherit from V2 test case and add test for zero bucket count
      
      * use an alternative (tf.fill) to be consistent
      
      * distinguish zero bucket count case v.s. empty input data case
      
      * add SummaryV3OpGraphTest test case and update tests
      
      * make bucket_count a variable before tf.cond op
      
        Move tf.math.maximum() to the top and avoid compile time shape
        inference that fails the conditional branch that isn't supposed
        to be execute when bucket_count is 0.
      42368f5b
  7. 06 10月, 2021 2 次提交
  8. 05 10月, 2021 4 次提交
    • S
      refactor: move settings template in to html (#5360) · 5f2e8c20
      Stephan Lee 提交于
      We have plans to add more features to the dialog and having the template
      in the TypeScript file is hard to read and understand. This change
      simply moves the template out into its own file.
      5f2e8c20
    • S
      runs table: fix color-picker position (#5349) · 0f8dc75f
      Stephan Lee 提交于
      ngx-color-picker is capable of automatically position w.r.t. the view
      port. However, because:
      1. we hard coded "bottom-right" position, and
      2. we had flicker of color-picker positioning incorrectly in the
      viewport because of lack of `overflow: hidden`
      
      it miscalculated the dialog's position and free space w.r.t. to the view
      port. By putting the `overflow`, it knows to only take available pixels
      in the view port (instead of making the page scrollable and think that
      there exists more space below the view port) into account.
      0f8dc75f
    • S
      runs-table: do not match against exp name (#5351) · ef3feb7f
      Stephan Lee 提交于
      Runs table, when in comparison mode, is capable of rendering experiment
      alias and show the full experiment name in the hover text. However,
      because we are matching regex against experiment name that is not
      visible without a user interaction, the match is quite surprising and is
      surprising users.
      
      This change removes an ability to match against the experiment name.
      ef3feb7f
    • N
      hparams: treat no data as an empty experiment rather than an error (#5357) · 049f875d
      Nick Felt 提交于
      * change backend_context.experiment_from_metadata() to always return an Experiment proto
      
      * remove fallback code to handle experiment_from_metadata() returning None
      
      * add backend_context_test.py test for experiment with no hparams summaries
      049f875d
  9. 01 10月, 2021 3 次提交
  10. 30 9月, 2021 1 次提交
  11. 29 9月, 2021 4 次提交
    • Y
      histogram: add back the test for large counts (#5348) · 551e0a03
      Yating 提交于
      * Revert "remove test that causes segmentation fault (#5346)"
      
      This reverts commit 9529cf7d.
      
      * increase summary_test size
      
      * add a comment in BUILD
      551e0a03
    • Y
      cleanup: log as info when no hparams experiment data is found (#5341) · 6a73acc8
      Yating 提交于
      * raise NotFoundError and return full msg in 400 response
      
      * add TODOs
      6a73acc8
    • Y
      remove test that causes segmentation fault (#5346) · 9529cf7d
      Yating 提交于
      9529cf7d
    • S
      run selector: subdue violent exception when hparams are invalid (#5344) · 468eb75b
      Stephan Lee 提交于
      * run selector: subdue violent exception when hparams are invalid
      
      When hparams specs in comparison mode are not compatible (e.g., an
      experiment has Discrete one while for the same name another experiment
      has Interval one), we currently throw a `RangeError`. That error manifests
      as very violent error that breaks entire app and make the TimeSeries
      plugin not functional. Since TimeSeries does not really render hparams,
      it does not make sense for this error to so violently throw error. Now,
      with this change, we do not even read the hparams value when in
      TimeSeries.
      
      Also, as a bonus, we have relaxed the constraint a bit and try to
      reconcile with incompatible filters if any of the compatible ones are
      empty.
      
      * grammar
      468eb75b
  12. 28 9月, 2021 7 次提交
    • S
      sync: change data-id binding to attribute (#5345) · 858658ec
      Stephan Lee 提交于
      We added the `[data-id]` on a span to uniquely identify a DOM newly
      stamped when new pin gets pinned. Bucause AOT Angular compiler complains
      about binding to a property in `<span>`, we are instead explicitly
      binding to `data-id` attribute on the DOM instead.
      858658ec
    • S
      color: put ColorPalette as state and improve data structure (#5342) · bd46127a
      Stephan Lee 提交于
      This change makes a color palette more elaborate so you can have a light
      and dark variant of a color with an id and display name. This is so that
      user can use different sets of colors without modifying our source code.
      
      After this change, we will introduce a new setting in the dialog that
      let user choose and preview different color palettes.
      bd46127a
    • S
      timeseries: show indication when there is a new pin (#5314) · a24c687d
      Stephan Lee 提交于
      The reason why we decided to show pinned section during search is
      because we wanted to make pinning more intuitive while you are filter
      searching (previously, when you pinned a card, there was no visual
      indication of an action besides a small fill on the pin icon which is
      occluded by the cursor). This change makes the section a bit smaller but
      compensate for the small UI with an indicator for new pinned card.
      a24c687d
    • S
      timeseries: match selection using same logic (#5334) · 6fe84b0a
      Stephan Lee 提交于
      Previously, to allow regex string also match run selection, we have
      extend the utility selector to match run selection against the regex.
      The logic, however, was massively flawed--instead of matching against
      name of a run, experiment, alias, and legacy run name, the
      implementation matched against runId which coincidentally contained run
      name but was supposed to be an opaque id. This caused an obvious match
      like `/^foo/` not to match anything especially when an opaque id is not
      created to start with "foo".
      
      In the end, this change basically re-implements the run selection based
      on regex more correct.
      
      In the process, we refactored how a run is matched against regex so the
      behavior is more congruent in our app.
      6fe84b0a
    • Y
      histogram: cast tf.reduce_sum input to float64 (#5337) · 589b0144
      Yating 提交于
      589b0144
    • S
      feat: adds experiment id as one of the searchable (#5339) · 2cc65d51
      Stephan Lee 提交于
      This change adds an enum for searching by experiment id.
      2cc65d51
    • S
      9f8553fe
  13. 25 9月, 2021 3 次提交
  14. 24 9月, 2021 2 次提交
  15. 23 9月, 2021 1 次提交