# -*- tab-width: 4; indent-tabs-mode: nil; py-indent-offset: 4 -*- # # Copyright the Collabora Office contributors. # # SPDX-License-Identifier: MPL-2.0 # # This Source Code Form is subject to the terms of the Mozilla Public # License, v. 2.0. If a copy of the MPL was not distributed with this # file, You can obtain one at http://mozilla.org/MPL/2.0/. # from uitest.framework import UITestCase from uitest.uihelper.calc import enter_text_to_cell from uitest.uihelper.common import get_state_as_dict from libreoffice.calc.document import get_cell_by_position from libreoffice.uno.propertyvalue import mkPropertyValues # Three columns of five observations each. Height and Weight both rise in equal # steps, so they standardize to the same column and the block has rank two. aColumns = [ ("Height", [10, 12, 14, 16, 18]), ("Weight", [20, 25, 30, 35, 40]), ("Age", [30, 20, 40, 10, 50]), ] # The mean row, the standard deviation row, two blank rows and the header row # come before the standardized values. nHeaderRow = 4 nFirstValueRow = 5 class principalComponentAnalysis(UITestCase): def enterData(self, gridwin, bWithLabels): for nColumn, (sLabel, aValues) in enumerate(aColumns): sColumn = chr(ord("A") + nColumn) nRow = 1 if bWithLabels: enter_text_to_cell(gridwin, sColumn + "1", sLabel) nRow = 2 for fValue in aValues: enter_text_to_cell(gridwin, sColumn + str(nRow), str(fValue)) nRow += 1 def runAnalysis(self, gridwin, sRange, bWithLabels, close_button="ok", aClearBoxes=()): gridwin.executeAction("SELECT", mkPropertyValues({"RANGE": sRange})) with self.ui_test.execute_modeless_dialog_through_command( ".uno:PrincipalComponentAnalysisDialog", close_button=close_button) as xDialog: xInputRangeEdit = xDialog.getChild("input-range-edit") xInputRangeEdit.executeAction("TYPE", mkPropertyValues({"KEYCODE": "CTRL+A"})) xInputRangeEdit.executeAction("TYPE", mkPropertyValues({"KEYCODE": "BACKSPACE"})) xInputRangeEdit.executeAction("TYPE", mkPropertyValues({"TEXT": "$Sheet1." + sRange})) # Every box starts ticked, so one is only clicked to clear it. for sBox in aClearBoxes: xDialog.getChild(sBox).executeAction("CLICK", tuple()) if not bWithLabels: xDialog.getChild("withlabels-check").executeAction("CLICK", tuple()) if close_button != "ok": xDialog.getChild("ok").executeAction("CLICK", tuple()) self.sErrorMessage = get_state_as_dict(xDialog.getChild("error-message"))["Text"] def test_the_new_sheet_holds_the_standardized_data_and_its_decomposition(self): with self.ui_test.create_doc_in_start_center("calc") as document: xCalcDoc = self.xUITest.getTopFocusWindow() gridwin = xCalcDoc.getChild("grid_window") self.enterData(gridwin, True) self.runAnalysis(gridwin, "$A$1:$C$6", True) self.assertEqual(document.Sheets.getByIndex(1).Name, "Sheet1_PCA") # The feature columns take the labels of the source columns, and # every column after them says what it holds. aExpectedHeader = [sLabel for sLabel, _ in aColumns] \ + ["Score_1", "Score_2", "Score_3"] \ + ["Singular Value"] \ + ["Loading_1", "Loading_2", "Loading_3"] \ + ["Variance Share", "Cumulative Share"] aHeader = [ get_cell_by_position(document, 1, nColumn, nHeaderRow).getString() for nColumn in range(len(aExpectedHeader))] self.assertEqual(aHeader, aExpectedHeader) # The mean and the standard deviation of every column sit above the # standardized values, each row named just right of its values. self.assertEqual(get_cell_by_position(document, 1, 0, 0).getValue(), 14) self.assertEqual(get_cell_by_position(document, 1, 1, 0).getValue(), 30) self.assertEqual(get_cell_by_position(document, 1, 2, 0).getValue(), 30) self.assertEqual( round(get_cell_by_position(document, 1, 0, 1).getValue(), 10), 3.1622776602) self.assertEqual( get_cell_by_position(document, 1, len(aColumns), 0).getString(), "Mean") self.assertEqual( get_cell_by_position(document, 1, len(aColumns), 1).getString(), "Standard Deviation") # Two rows of nothing hold those two apart from the table below. for nRow in (2, 3): for nColumn in range(len(aColumns) + 1): self.assertEqual( get_cell_by_position(document, 1, nColumn, nRow).getString(), "") # Every column is wide enough for its longest entry, so none of them # is left at the width a new sheet starts with. oSheet = document.Sheets.getByIndex(1) nDefaultWidth = document.Sheets.getByIndex(0).Columns.getByIndex(0).Width for nColumn in range(len(aColumns) + 9): self.assertNotEqual(oSheet.Columns.getByIndex(nColumn).Width, nDefaultWidth) # The three parts of the decomposition all read the whole block of # standardized values and start on its first row. The left vectors # take up one column per component, so the singular values land # three columns further right and the right vectors one beyond # those. sValueRange = "$A$%d:$C$%d" % (nFirstValueRow + 1, nFirstValueRow + 5) for nColumn, nPart in ((3, "1"), (6, "2"), (7, "3")): self.assertEqual( get_cell_by_position(document, 1, nColumn, nFirstValueRow).getFormula(), "{=MSVD(" + sValueRange + ";" + nPart + ")}") # Standardizing leaves every column with a mean of zero and a # standard deviation of one. for nColumn in range(len(aColumns)): aStandardized = [ get_cell_by_position(document, 1, nColumn, nFirstValueRow + nRow).getValue() for nRow in range(5)] self.assertEqual(round(sum(aStandardized), 10), 0) self.assertEqual(round(sum(x * x for x in aStandardized), 10), 4) # The squared singular values add up to the sum of the squares of # the standardized values, which is five observations less one for # each of the three columns. aSingularValues = [ get_cell_by_position(document, 1, 6, nFirstValueRow + nRow).getValue() for nRow in range(3)] self.assertEqual(round(sum(x * x for x in aSingularValues), 8), 12) self.assertEqual(aSingularValues, sorted(aSingularValues, reverse=True)) # Two of the three columns are a straight line in the row number, # so the third singular value is zero. self.assertEqual(round(aSingularValues[2], 10), 0) # Each component carries the share of the variance its squared # singular value is of the total, and the shares add up to one. fTotal = sum(x * x for x in aSingularValues) aShares = [ get_cell_by_position(document, 1, 10, nFirstValueRow + nRow).getValue() for nRow in range(3)] for nRow in range(3): self.assertEqual( round(aShares[nRow], 10), round(aSingularValues[nRow] ** 2 / fTotal, 10)) self.assertEqual(round(sum(aShares), 10), 1) # The column beside it is the running total of the shares, so it # reaches one on the last component. aCumulative = [ get_cell_by_position(document, 1, 11, nFirstValueRow + nRow).getValue() for nRow in range(3)] for nRow in range(3): self.assertEqual( round(aCumulative[nRow], 10), round(sum(aShares[:nRow + 1]), 10)) self.assertEqual(round(aCumulative[2], 10), 1) # Both columns are shown as percentages. for nColumn in (10, 11): nFormatKey = get_cell_by_position( document, 1, nColumn, nFirstValueRow).NumberFormat sFormat = document.getNumberFormats().getByKey(nFormatKey).FormatString self.assertIn("%", sFormat) # A chart on the sheet draws the two share columns, the per # component share as bars and the running total as a line over # them. oCharts = document.Sheets.getByIndex(1).Charts self.assertEqual(oCharts.getCount(), 2) oChart = oCharts.getByIndex(0) aRanges = [ (oRange.StartColumn, oRange.StartRow, oRange.EndColumn, oRange.EndRow) for oRange in oChart.getRanges()] self.assertEqual(aRanges, [(10, nHeaderRow, 11, nFirstValueRow + 2)]) oChartDocument = oChart.getEmbeddedObject() def sTitleOf(oTitled): return "".join(oString.getString() for oString in oTitled.getTitleObject().getText()) self.assertEqual(sTitleOf(oChartDocument), "Variance share by component") oDiagram = oChartDocument.getFirstDiagram() self.assertEqual(oDiagram.Wall.FillColor, 0xFFFFFF) aChartTypes = [ oChartType.getChartType() for oSystem in oDiagram.getCoordinateSystems() for oChartType in oSystem.getChartTypes()] self.assertEqual(aChartTypes, ["com.sun.star.chart2.ColumnChartType", "com.sun.star.chart2.LineChartType"]) # The bars read the share column and the line the running total # beside it. aSeriesRanges = [] for oSystem in oDiagram.getCoordinateSystems(): for oChartType in oSystem.getChartTypes(): for oSeries in oChartType.getDataSeries(): for oLabeled in oSeries.getDataSequences(): oValues = oLabeled.getValues() if oValues.Role == "values-y": aSeriesRanges.append( oValues.getSourceRangeRepresentation()) self.assertEqual(len(aSeriesRanges), 2) self.assertIn("$K$", aSeriesRanges[0]) self.assertIn("$L$", aSeriesRanges[1]) # Both axes say what they carry. oSystem = oDiagram.getCoordinateSystems()[0] self.assertEqual( sTitleOf(oSystem.getAxisByDimension(0, 0)), "Principal Components") self.assertEqual( sTitleOf(oSystem.getAxisByDimension(1, 0)), "Variance share") # Both series are a share of a whole, so the axis they share counts # in percent of its own accord. oAxis = oSystem.getAxisByDimension(1, 0) self.assertFalse(oAxis.LinkNumberFormatToSource) sAxisFormat = document.getNumberFormats().getByKey( oAxis.NumberFormat).FormatString self.assertIn("%", sAxisFormat) # The second chart is a correlation circle over one range that runs # from the standardized features straight into the first two score # columns. oCircleChart = oCharts.getByIndex(1) aCircleRanges = [ (oRange.StartColumn, oRange.StartRow, oRange.EndColumn, oRange.EndRow) for oRange in oCircleChart.getRanges()] self.assertEqual( aCircleRanges, [(0, nHeaderRow, len(aColumns) + 1, nFirstValueRow + 4)]) oCircleDocument = oCircleChart.getEmbeddedObject() self.assertEqual(sTitleOf(oCircleDocument), "Correlation circle") oCircleDiagram = oCircleDocument.getFirstDiagram() self.assertEqual(oCircleDiagram.Wall.FillColor, 0xFFFFFF) # Each direction is named after which of the pair of components it # is. oCircleSystem = oCircleDiagram.getCoordinateSystems()[0] self.assertEqual( sTitleOf(oCircleSystem.getAxisByDimension(0, 0)), "Principal Component 1") self.assertEqual( sTitleOf(oCircleSystem.getAxisByDimension(1, 0)), "Principal Component 2") self.assertEqual( [oChartType.getChartType() for oSystem in oCircleDiagram.getCoordinateSystems() for oChartType in oSystem.getChartTypes()], ["com.sun.star.chart2.CorrelationCircleChartType"]) # One series per feature, each holding its own feature column beside # the two component columns they all share. aFeatureRoles = [] for oSystem in oCircleDiagram.getCoordinateSystems(): for oChartType in oSystem.getChartTypes(): for oSeries in oChartType.getDataSeries(): aFeatureRoles.append(sorted( oLabeled.getValues().Role for oLabeled in oSeries.getDataSequences())) self.assertEqual(len(aFeatureRoles), len(aColumns)) for aRoles in aFeatureRoles: self.assertEqual(aRoles, ["values-feature", "values-x", "values-y"]) # A second run cannot have the sheet it needs, so it reports that # and adds nothing. self.runAnalysis(gridwin, "$A$1:$C$6", True, close_button="cancel") self.assertIn("Sheet1_PCA", self.sErrorMessage) self.assertEqual(document.Sheets.getCount(), 2) # Undo takes the whole sheet away again. self.xUITest.executeCommand(".uno:Undo") self.assertEqual(document.Sheets.getCount(), 1) def test_features_are_named_after_their_place_when_the_source_has_no_labels(self): with self.ui_test.create_doc_in_start_center("calc") as document: xCalcDoc = self.xUITest.getTopFocusWindow() gridwin = xCalcDoc.getChild("grid_window") self.enterData(gridwin, False) self.runAnalysis(gridwin, "$A$1:$C$5", False) # With nothing to take a name from, each feature is named after its # place in the range, and the header row is there all the same. for nColumn in range(len(aColumns)): self.assertEqual( get_cell_by_position(document, 1, nColumn, nHeaderRow).getString(), "Feature_" + str(nColumn + 1)) # Every row of the source counts as data, so the mean covers all # five and the standardized values start below the header row. self.assertEqual(get_cell_by_position(document, 1, 0, 0).getValue(), 14) self.assertEqual( round(get_cell_by_position(document, 1, 0, nFirstValueRow).getValue(), 10), -1.2649110641) def test_a_labelled_source_column_left_blank_is_named_after_its_place(self): with self.ui_test.create_doc_in_start_center("calc") as document: xCalcDoc = self.xUITest.getTopFocusWindow() gridwin = xCalcDoc.getChild("grid_window") self.enterData(gridwin, True) gridwin.executeAction("SELECT", mkPropertyValues({"CELL": "B1"})) self.xUITest.executeCommand(".uno:ClearContents") self.runAnalysis(gridwin, "$A$1:$C$6", True) # The labelled columns keep their names and the blank one falls back # to its place in the range. self.assertEqual( get_cell_by_position(document, 1, 0, nHeaderRow).getString(), "Height") self.assertEqual( get_cell_by_position(document, 1, 1, nHeaderRow).getString(), "Feature_2") self.assertEqual( get_cell_by_position(document, 1, 2, nHeaderRow).getString(), "Age") def test_a_cell_that_is_not_a_number_stops_the_analysis(self): with self.ui_test.create_doc_in_start_center("calc") as document: xCalcDoc = self.xUITest.getTopFocusWindow() gridwin = xCalcDoc.getChild("grid_window") self.enterData(gridwin, True) enter_text_to_cell(gridwin, "B4", "no reading") self.runAnalysis(gridwin, "$A$1:$C$6", True, close_button="cancel") self.assertIn("$B$4", self.sErrorMessage) self.assertEqual(document.Sheets.getCount(), 1) def test_an_empty_cell_stops_the_analysis(self): with self.ui_test.create_doc_in_start_center("calc") as document: xCalcDoc = self.xUITest.getTopFocusWindow() gridwin = xCalcDoc.getChild("grid_window") self.enterData(gridwin, True) gridwin.executeAction("SELECT", mkPropertyValues({"CELL": "C5"})) self.xUITest.executeCommand(".uno:ClearContents") self.runAnalysis(gridwin, "$A$1:$C$6", True, close_button="cancel") self.assertIn("$C$5", self.sErrorMessage) self.assertEqual(document.Sheets.getCount(), 1) def test_every_box_starts_ticked(self): with self.ui_test.create_doc_in_start_center("calc"): xCalcDoc = self.xUITest.getTopFocusWindow() gridwin = xCalcDoc.getChild("grid_window") self.enterData(gridwin, True) gridwin.executeAction("SELECT", mkPropertyValues({"RANGE": "$A$1:$C$6"})) with self.ui_test.execute_modeless_dialog_through_command( ".uno:PrincipalComponentAnalysisDialog", close_button="cancel") as xDialog: for sBox in ("withlabels-check", "variance-chart-check", "correlation-chart-check"): self.assertEqual( get_state_as_dict(xDialog.getChild(sBox))["Selected"], "true") def test_clearing_a_chart_box_leaves_that_chart_out(self): with self.ui_test.create_doc_in_start_center("calc") as document: xCalcDoc = self.xUITest.getTopFocusWindow() gridwin = xCalcDoc.getChild("grid_window") self.enterData(gridwin, True) # Only the correlation circle is asked for, so it is the one chart # on the sheet. self.runAnalysis(gridwin, "$A$1:$C$6", True, aClearBoxes=("variance-chart-check",)) oCharts = document.Sheets.getByIndex(1).Charts self.assertEqual(oCharts.getCount(), 1) self.assertEqual( oCharts.getByIndex(0).getEmbeddedObject().getFirstDiagram() .getCoordinateSystems()[0].getChartTypes()[0].getChartType(), "com.sun.star.chart2.CorrelationCircleChartType") # Clearing both leaves the numbers on their own. self.xUITest.executeCommand(".uno:Undo") self.runAnalysis(gridwin, "$A$1:$C$6", True, aClearBoxes=("variance-chart-check", "correlation-chart-check")) self.assertEqual(document.Sheets.getByIndex(1).Charts.getCount(), 0) def test_a_single_column_cannot_be_analysed(self): with self.ui_test.create_doc_in_start_center("calc") as document: xCalcDoc = self.xUITest.getTopFocusWindow() gridwin = xCalcDoc.getChild("grid_window") self.enterData(gridwin, True) self.runAnalysis(gridwin, "$A$1:$A$6", True, close_button="cancel") self.assertIn("two columns", self.sErrorMessage) self.assertEqual(document.Sheets.getCount(), 1) # vim: set shiftwidth=4 softtabstop=4 expandtab: