New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[ENH] Linear Projection (LDA, PCA) #2445

Merged

lanzagar merged 6 commits into biolab:master from jerneju:new-linearprojection

Dec 11, 2017

Contributor

jerneju commented Jun 30, 2017 •

edited

Loading

Issue

Description of changes

Merged Linear Projection from Orange3, Linear Projection from Prototypes
LDA projection
PCA projection (NEW)
Input projections
Rank for LDA and Circular projections.
Now use Scatter Plot Graph instead of Plotutils
Added some new tests
Updated documentation

Includes

Code changes
Tests
Documentation

jerneju mentioned this pull request

Remove obsolete widgets biolab/orange3-prototypes#77

Closed

codecov-io commented Jul 7, 2017 •

edited

Loading

Codecov Report

Merging #2445 into master will decrease coverage by 5.12%.
The diff coverage is 88.38%.

@@            Coverage Diff             @@
##           master    #2445      +/-   ##
==========================================
- Coverage   81.68%   76.55%   -5.13%     
==========================================
  Files         323      337      +14     
  Lines       55450    59664    +4214     
==========================================
+ Hits        45293    45678     +385     
- Misses      10157    13986    +3829

kernc reviewed

View reviewed changes

Orange/projection/freeviz.py Outdated

		@@ -0,0 +1,348 @@

		import numpy

Contributor

kernc Jul 7, 2017

Import numpy as np.

Contributor Author

jerneju Jul 7, 2017

Sure, thanks. I always do replace numpy with np. Mostly before changing anything else. However this file is just a copy from Prototypes.

Contributor

janezd commented Aug 22, 2017

What is the status of this PR? Since we later decided for separate widgets, is this PR still relevant?

Contributor Author

jerneju commented Aug 23, 2017

Yes, this PR is still relevant. I have edited the first comment, see above.

jerneju changed the title ~~[WIP] Linear Projection~~ [WIP] Linear Projection (LDA, PCA)

Contributor Author

jerneju commented Sep 29, 2017

@janezd
When GUI is approved, I will updated the documentation.

jerneju changed the title ~~[WIP] Linear Projection (LDA, PCA)~~ Linear Projection (LDA, PCA)

Contributor Author

jerneju commented Oct 9, 2017

janezd previously requested changes

View reviewed changes

Contributor

janezd left a comment

I am reviewing this for some time. I am going up and down through the code, trying to get some sense of what is actually needed and what are scattered remains of previous reincarnations and designs of the widget. The comments are unfortunately ordered by lines and not chronologically, which would in this case make more sense (or at least illustrate my struggle here).

My reviewing stops here. I'm sorry, but this code is such a mess that I can' t provide a useful review. I'm marking it as WIP. Please remove this flag when the code is ready for a proper review.

Orange/widgets/utils/plot/owplotgui.py Outdated

		SIZE_POLICY_FIXED = (QSizePolicy.Minimum, QSizePolicy.Maximum)


		class Hidden(QDialog):

Contributor

janezd Oct 13, 2017

This name is uninformative. This is essentially and AddVariablesDialog or sth like that.

Orange/widgets/utils/plot/owplotgui.py Outdated

+                      )
+                      self.btn_select = QPushButton(
+                          "Select", autoDefault=False, sizePolicy=QSizePolicy(*SIZE_POLICY_FIXED)
+                      )

Contributor

janezd Oct 13, 2017

Rename this button to Add.

Contributor

janezd Oct 13, 2017

This is not a modal dialog. It is modal in the sense that is blocks other dialogs, but it does not function as one from the usage perspective. As it is now, the user must select attributes, click "Select" (after actually selecting them!) to add them. Then (s)he can add more. After (s)he is done, she has to click Cancel to essentially confirm what (s) has done.

Implement it like this: the widget calls this dialog and receives a list of features to add. That is, the dialog has two buttons, Add and Cancel. The user selects the features and clicks Add to confirm or Cancel to cancel. After this, the dialog closes and the features are added. See if you can use some standard Qt stuff so that placement of buttons is correct wrt the platform; I'm not sure this is possible, though.

Make Add a default button.

Orange/widgets/utils/plot/owplotgui.py Outdated

+                      self.btn_select = QPushButton(
+                          "Select", autoDefault=False, sizePolicy=QSizePolicy(*SIZE_POLICY_FIXED)
+                      )
+                      self.btn_select.clicked.connect(master.display_select)

Contributor

janezd Oct 13, 2017

... and this magic will no longer be needed.

Orange/widgets/utils/plot/owplotgui.py Outdated

+                      self.layout().addWidget(master.other_view)
+                      self.layout().addWidget(btns_area)
+                  def __call__(self, master):

Contributor

janezd Oct 13, 2017

Do it more in a style of Qt's standard dialogs. If you do it properly, it may even be a function, I guess.

Orange/widgets/utils/plot/owplotgui.py Outdated

+                      self.master.master.setEnabled(True)
+                      super().closeEvent(QCloseEvent)
+              class VariableSelection(QObject):

Contributor

janezd Oct 13, 2017

How much of this code is still needed? Accepting drops looks like functionality that we only needed when the widget still had two list views.

This looks like a huge class for what is essentially a decorated listview with ordering.

Orange/widgets/utils/plot/owplotgui.py Outdated

+                      view.setModel(model)
+                      self.filter_edit, self.other_view = variables_filter(
+                          model=model)

Contributor

janezd Oct 13, 2017

self.other_view is set here ... but also a few lines earlier. What is self.other_view?!

Orange/widgets/utils/plot/owplotgui.py Outdated

+                      view.setModel(model)
+                      self.filter_edit, self.other_view = variables_filter(
+                          model=model)
+                      self.hidden = Hidden(self)

Contributor

janezd Oct 13, 2017

I believe that a huge part of the mess related to models comes from having Hidden as a persistent dialog. Make it a normal modal dialog that is created when needed, and simply initialized with all attributes appearing in the domain that are currently hidden.

Orange/widgets/utils/plot/owplotgui.py Outdated

+                          "Move up", view,
+                          shortcut=QKeySequence(Qt.Key_Delete),
+                          triggered=self.__activate_selection
+                      )

Contributor

janezd Oct 13, 2017

The delete key invoked action "Move up", which calls method __activate_selection, which deletes a variable? Should something be renamed here, or am I getting something wrong?

...

Ah, OK, after checking the action attached to the minus button, I see. Move up refers to moving to a list view that used to be above some dialog. And activation refers to showing the selected variables in the plot.

Orange/widgets/utils/plot/owplotgui.py Outdated

+                      for i in sorted((ind.row() for ind in indices), reverse=True):
+                          del model[i]
+                      self.varmodel_other.extend(variables)

Contributor

janezd Oct 13, 2017

And this is needed just because the dialog with hidden attributes is persistent. The reason why it is persistent is that the list view used to be persistent in the previous GUI.

Orange/widgets/utils/plot/owplotgui.py Outdated

+                      self.hidden(self)
+                  @staticmethod
+                  def encode_var_state(lists):

Contributor

janezd Oct 13, 2017

Are these encodes and decodes really needed? Cannot settings take care of lists of variables? Is this again some legacy from the previous form of the widget?

Contributor Author

jerneju Nov 7, 2017

Yes, they are needed.

janezd changed the title ~~Linear Projection (LDA, PCA)~~ [WIP] Linear Projection (LDA, PCA)

Contributor

janezd commented Oct 13, 2017

One more thing: the layout. Rank button is currently above the list view, while the +- buttons have empty space on the right. They are also below the box instead of in it.

Since ranking is closely related to +-, it should be there. Like this:

I'm attaching a patch that will do some of the job.

jerneju mentioned this pull request

[FIX] FreeViz: work when Linear Projection is merged biolab/orange3-prototypes#96

Merged

3 tasks

jerneju changed the title ~~[WIP] Linear Projection (LDA, PCA)~~ [WIP] [ENH] Linear Projection (LDA, PCA)


          Distributions: move LegendItem and ScatterPlotItem from Linear Projec…

7dee9c9

…tion

jerneju changed the title ~~[WIP] [ENH] Linear Projection (LDA, PCA)~~ [ENH] Linear Projection (LDA, PCA)


          Plot GUI: Common tool for variables selection

b4d48df

kernc reviewed

View reviewed changes

Orange/widgets/visualize/owlinearprojection.py Outdated

+                                        PCA=2,
+                                        Projection=3),
+                                   type=int,
+                                   qualname="Placement")

Contributor

kernc Nov 21, 2017

pickle.PicklingError: Can't pickle <enum 'Placement'>: attribute lookup Placement on Orange.widgets.visualize.owlinearprojection failed

Perhaps should be qualname+'OWLinearProjection.Placement'.

jerneju modified the milestones: 3.8, 3.9

kernc requested a review from ajdapretnar

November 30, 2017 16:40

kernc approved these changes

View reviewed changes

jerneju assigned lanzagar

ajdapretnar reviewed

View reviewed changes

Contributor

ajdapretnar left a comment

I vote for updating the Example part, too. Now the workflow has three different outputs that are not explained anywhere. Add a line commenting on what these outputs mean/do/work in the analysis.
It would also be nice to add an example on how to work with a custom projection.

doc/visual-programming/source/widgets/visualize/linearprojection.rst Outdated

@@ @@ -18,17 +18,37 @@ Signals @@
                  A subset of data instances
+              -  **Projection**
+                  Custom projection vectors

Contributor

ajdapretnar Dec 1, 2017

De-indent.

doc/visual-programming/source/widgets/visualize/linearprojection.rst Outdated

+              of class-labeled data. It supports various types of projections such as circular,
+              `linear discriminant analysis <https://en.wikipedia.org/wiki/Linear_discriminant_analysis>`_,
+              `principal component analysis <https://en.wikipedia.org/wiki/Principal_component_analysis>`_,
+              and custom projection

Contributor

ajdapretnar Dec 1, 2017

a custom projection. Add a full stop.

doc/visual-programming/source/widgets/visualize/linearprojection.rst Outdated

+. Optimize your projection by using **Suggest Features**. This feature
+                 scores attributes by average classification accuracy and returns the
+                 top scoring attributes with a simultaneous visualization update.
+. Choose the type of projection.

Contributor

ajdapretnar Dec 1, 2017

Indenting is all off. Please fix. Never forget to check how the docs look like when built with make html. Sphinx will report errors in docs. You also need to pip install CommonMark==0.5.5. Weird stuff. 🤷‍♀️

doc/visual-programming/source/widgets/visualize/linearprojection.rst Outdated


		- Label only selected points allows you to select individual data instances and label them.

		7. Select, zoom, pan and zoom to fit are the options for exploring the graph.

Contributor

ajdapretnar Dec 1, 2017

and should not be in italics

doc/visual-programming/source/widgets/visualize/linearprojection.rst Outdated

+                 -  *Label only selected points* allows you to select individual data instances and label them.
+. *Select, zoom, pan and zoom to fit* are the options for exploring the graph.
+                 The manual selection of data instances works as an angular/square

Contributor

ajdapretnar Dec 1, 2017

Manual selection (omit the)

jerneju dismissed janezd’s stale review

December 7, 2017 12:40

Fixed

lanzagar requested changes

View reviewed changes

Contributor

lanzagar left a comment

Use iris, enable class density in Linear projection and switch between different projections => only part of the picture is colored.
Show axes has a radius slider. When I increase it, the axes start disappearing which is counterintuitive for a "Show axes" control. Maybe rename to "Hide axes"? (@janezd any better suggestions?)

jerneju added 3 commits

December 11, 2017 12:37


          Linear Projection: major changes

76b96b4

Use ScatterPlotGraph
add VizRank (Suggest Features)
LDA, PCA, and Input Projection
New Warnings
Migrate settings and context
Always commit as metas
new Scatter Plot Graph compatibility


          Linear Projection: Suggest Features based on a chosen color

9a417af

fix PCA
selection fix
fix vizrank
lint


          Linear Projection: documentation update

c18ce0c


          Linear Projection: new tests

6b932f9

lanzagar approved these changes

View reviewed changes

lanzagar merged commit 2c54035 into biolab:master

jerneju deleted the new-linearprojection branch

December 11, 2017 14:48

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet