SGI MineSet 2.0.1

home *** CD-ROM | disk | FTP | other *** search

/ SGI MineSet 2.0.1 / SGI MineSet 2.0.1.iso / docs6.3 / relnotes / MineSet / ch6.z / ch6

Wrap

Text File | 1998-01-06 | 26.1 KB | 727 lines

- 1 - 6. _K_n_o_w_n__P_r_o_b_l_e_m_s__a_n_d__W_o_r_k_a_r_o_u_n_d_s 6.1 _G_e_n_e_r_a_l__p_r_o_b_l_e_m_s +o In IRIX 6.3 and later, the system graphics libraries have an unenforced prerequisite on dmedia_eoe.sw.base. If the graphics tools fail to run, giving an error that libvice.so could not be loaded, this should be installed. Note that the error message will go to the terminal window from which mineset was launched, and if it was launched from the desktop, it might not be seen at all. +o Some graphics tools will display font problems when running over the network from a machine running IRIX 6.4. To solve this, install patch 1946 on an Origin, or Patch 1953 on an Octane. See chapter 2 for details. +o If you tear off menus in the visual tools, and then reopen the file, the tools may crash. To fix this, you should install patch 1840 on IRIX 6.2, patch 2037 on IRIX 6.3, or patch 2107 on IRIX 6.4. See chapter 2 for details. +o If a drill-through involves a floating point value, the drill-through expression will include a small margin of error around that floating point value to avoid round off error. (For example, instead of searching for x=1.0, it may search for x>=0.99999999999999 && x<=1.00000000000001.) This may result in drill through returning some data that doesn't exactly match the selection criteria. If this is a problem, use "Edit History" in the tool manager to edit the filter criteria. +o Most of the visual tools have a File pulldown menu option to save or print images of themselves. However, in Release 2.01, due to a limitation in the implementation, this functionality is only available when displaying on a Silicon Graphics workstation. When displaying to other hardware, use save or print functions provided by that hardware. For example, when displaying to a PC running Hummingbird Exceed 3D, you can print or save using the pulldown menu in the upper left corner of the main window. +o If the visual tools are displaying a table with several hundred columns, they may crash when one of the objects is highlighted or selected. - 2 - +o Some of the tools may fail ungracefully if there is insufficient memory. +o When entering an expression where a string is assigned a null value, you must use the string "?" rather than a non quoted question mark (?). Failure to do so will result in a type mismatch. +o The graphical tools may run out of colors, especially if there are many colors in a legend. While this will not break the tool, it may degrade the appearance, the legend will not show the colors of the objects, and it may affect other tools. +o In the graphics tools, the F1 key does not work for providing help when clicked over the graphics window. +o In many of the file selection boxes, if you open a file, and then reopen the file selection box, the previously opened file will be highlighted, but will not be in the text field below. Pressing OK at that point will attempt to open the directory containing the file, rather than the file itself. It is necessary to specifically select the file to cause it to be opened. +o If you complement the drill-through in a visual tool with sliders, the drill-through will select everything except that which matches the current slider setting and the current selection. (For example, you might get everything except California in 1986. You will get California in 1987, and Nebraska in 1986.) This is probably not what is desired. To work around this, use Send to Tool Manager, and edit the resulting filter expression. +o When the tools display floating point values, they only display them to a few significant decimal places. This is intended to be easier to read, but could result in problems if you need exact values. 6.2 _P_r_o_b_l_e_m_s__w_h_e_n__r_u_n_n_i_n_g__u_n_d_e_r__H_u_m_m_i_n_g_b_i_r_d__E_x_c_e_e_d Several problems occur when displaying MineSet to a PC running Hummingbird Exceed. Hummingbird Communications has indicated that these will be fixed in the next release of Exceed. +o The Statistics Visualizer does not work. Only the first pane will display graphics. Also scrolling may leave the window corrupted. A dialog will be posted if you attempt to run the Statistics Visualizer against - 3 - Hummingbird warning of these problems. To disable the warning add the line *minesetWarnHummingbirdStatviz:FALSE to your .Xdefaults file. +o In the Tree Visualizer, if you use the open or reopen menus to open a second visualization, some or all of the letters may be missing from labels. Also, spotlights may appear opaque or might not point to the correct locations. To work around this problem, exit and restart the Tree Visualizer. 6.3 _K_n_o_w_n _p_r_o_b_l_e_m_s _w_i_t_h _I_n_d_u_c_e_r_s, _C_l_a_s_s_i_f_i_e_r_s, _a_n_d _C_o_l_u_m_n _I_m_p_o_r_t_a_n_c_e +o When a classifier is applied (apply-classifier from Tool Manager), the resulting column is always a string, even if the original label column was of another type. If the type needs to be an int, a type conversion can be applied from the data transformation panel in tool manager. +o The MIndUtil options in .mineset-classopt are not part of the session mechanism that Tool Manager supports. If Tool Manager files (.mineset suffix) are transferred between users without passing the .mineset-classopt file, then results may differ. 6.4 _K_n_o_w_n__p_r_o_b_l_e_m_s__w_i_t_h__t_h_e__T_r_e_e__V_i_s_u_a_l_i_z_e_r +o If you are running IRIX 6.2, you must install patch 1535. Otherwise, the Tree Visualizer may fail with a core dump while navigating through the scene. See chapter 2 for details on installing this patch. +o If you are running IRIX 6.2, you must install patch 1840. Otherwise, the Tree Visualizer may fail with a core dump when trying to display the color chooser from the marks panel. See chapter 2 for details on installing this patch. +o If you average data in the aggregation panel of the tool manager, and then have the Tree Visualizer average it again, this will result in an average of averages, which is generally not correct. Tool manager will give a warning if it detects this case. To work around this, don't aggregate in tool manager, and let the Tree Visualizer do all of the averaging. +o When a second view is opened in place of the first, not all of the memory used by the first view is released. - 4 - If an error occurs in a configuration file and the file is corrected and reopened, some memory will also be lost. After reopening the file several times or opening several configuration files, it is wise to exit the program and restart it. +o If a node has more than 480 children, and the menu listing children is brought up, and a child after the 480th is selected, the Tree Visualizer will fail with a core dump. +o For very large hierarchies, navigation and selection may not work correctly near the leaves of the hierarchy. +o If filtering by depth is on in a very large hierarchy, and some nodes have a large number of children, navigation may be difficult. +o There is no easy way to set everything to the same color, other than the default grey. The only way to force everything to a single color is to explicitly map a column to the color, and enter a single scale value, assigning it to that color. For example, in the configuration file, if var were the name of a variable, you could enter color var, scale 0, colors "red"; +o On some hardware platforms, the ground and sky are a single solid color rather than shaded between two colors. +o There is no way to save marks when operation on a configuration file in a read-only directory. +o Occasionally, a graphical object will not unhighlight correctly when the pointer is moved off of it. Navigating through the scene slightly, or forcing a redraw by iconifying and deiconifying the window, will clear the problem. +o When the decision tree generates a view for the Tree Visualizer, sometimes the colors used for mapping the different bars will be similar to the colors used for mapping purity, although there is no relationship between them. - 5 - 6.5 _K_n_o_w_n__p_r_o_b_l_e_m_s__w_i_t_h__t_h_e__M_a_p__V_i_s_u_a_l_i_z_e_r +o On rare occasion, Mapviz may appear to hang at startup, never even showing the initial window. This problem has been traced to a bug in the lower-level ToolTalk library. An effective workaround is to go to a shell window and execute this command: xprop -root -remove _SUN_TT_SESSION +o Every named graphical object keyword in the Hierarchy file must be found in the Gfx File associated with that graphical object's level in the hierarchy. +o When heights and/or colors are derived from values from two-dimensional arrays, the sliders must be declared in the View Map Section in the order they are used in the array declaration in the Input Section. That is, if the data is in array xyz[keyA][keyB], then the "slider keyA;" statement must precede the "slider keyB;" statement. +o Heights and/or colors which are derived from array values are computed using the precise data values from the input data file. As the slider(s) move between these precise data points, the heights and/or colors are computed using values which are linear interpolations between these known data values, and hence the resulting graphics is only a "best guess" of what the true intermediate data values might be. +o The above comments hold true for any array expressions which are derived from array input data. This means that if an array C[] is defined as arrays A[]/B[], then the value of C[] will match this expression at the precise datapoints, but not necessarily at the intermediate interpolated points. +o The Gfx File /_u_s_r/_l_i_b/_M_i_n_e_S_e_t/_m_a_p_v_i_z/_g_f_x__f_i_l_e_s/_u_s_a._z_i_p_c_o_d_e._g_f_x purports to contain U.S. zipcode graphics, but there are flaws in its contents: many zipcodes are missing, which means that data which refers to a missing zipcode will produce a generic rectangle off to the right of the rest of the zipcodes in the main display window; zipcode graphical objects are merely suggestive, rather than precisely accurate in size and shape. Zipcodes are represented by equally sized rectangles which are roughly centered on the geographical center of the true zipcode area. The "zip100" aggregations (consisting of zipcode groups which share the first three digits) are larger rectangles which are also generally centered at - 6 - the aggregated geographical center of the constituent zipcodes. 6.6 _K_n_o_w_n__p_r_o_b_l_e_m_s__w_i_t_h__t_h_e__S_c_a_t_t_e_r__V_i_s_u_a_l_i_z_e_r +o For the cube and diamond shapes, entity sizes are based on absolute values. This makes it impossible to interpret the sizes unambiguously if the variable mapped to size includes both negative and positive data values. You can work around this by using the bar entity shape. +o It is possible for entities to overlap, or for one entity to be contained in another (and thus hidden from view). If the entities are the same size, this can lead to visual artifacts, such as an entity being drawn in two different colors. To minimize this problem, use the diamond entity shape. +o Highlighting entities in large datasets can be slow. When this happens, you can filter to a subset of the data to improve the speed. +o Highlighting does not always work when 3D perspective is off. +o Display attributes (size, color, position) which are derived from array values are computed using the precise data values from the input data file. As the animation moves between these precise data points, the display attributes are computed using values which are linear interpolations between these known data values, and hence the resulting graphics is only a "best guess" of what the true intermediate data values might be. +o The above comments hold true for any array expressions which are derived from array input data. This means that if an array C[] is defined as arrays A[]/B[], then the value of C[] will match this expression at the precise datapoints, but not necessarily at the intermediate interpolated points. 6.7 _K_n_o_w_n__p_r_o_b_l_e_m_s__w_i_t_h__t_h_e__S_p_l_a_t__V_i_s_u_a_l_i_z_e_r +o In rare cases where a column having only a single value is mapped to an axis requirement, the Splat Visualizer can hang. - 7 - 6.8 _K_n_o_w_n__P_r_o_b_l_e_m_s__w_i_t_h__t_h_e__D_a_t_a__M_o_v_e_r +o The ascii file format produced by the Data Mover uses tabs to delimit the boundaries between attribute values, and newlines to delimit records. If a database table contains attribute values with embedded tabs or newlines, the file produced by the Data Mover will not be readable by the Data Mover or any of the MineSet tools. The symptom in this case is a reader error complaining of too much or too little data on a line. The workaround is to use MineSet binary files, rather than ascii, if the user suspects a database contains values with embedded tab or newline characters. +o Even if the Tool Manager is killed with a control-C, the Data Mover will continue processing the current query or mining command until it finishes. +o Since the Data Mover locks the files directory in order to prevent conflicts and incorrect results, a hung datamove process will prevent further access to the server by that user. The MineSet Tool Manager will offer to kill such processes. +o The algorithm used by the Data Mover to perform aggregation clusters, rather than sorts, the output records. The output order of the records is not meaningful. 6.9 _K_n_o_w_n__P_r_o_b_l_e_m_s__w_i_t_h__t_h_e__T_o_o_l__M_a_n_a_g_e_r +o On the classifier destination panel if the discrete label is changed while the loss matrix is turned on, the loss matrix is automatically turned off and a warning is shown. When this occurs the loss matrix values can be restored by setting the discrete label back to the label for which the loss matrix values were set and reselect the loss matrix. However, if the loss matrix is turned on for another label the previous values will be erased. +o The Tool Manager does not support drag and drop. If you try to drag and drop items from dialog boxes, fatal errors will sometimes result. +o When creating your own ".schema" files, there are a number of reserved names in the file syntax. To avoid conflict with these names, enclose the column names in backward single quotes, i.e., `column name`. - 8 - +o Drill-through behaves differently when a client data file has been mapped directly to a visual tool with no transformations. The history used for drill-through is stored in the client schema file, thus drill-through on a visualization of that file will show the history that created that file. As a result, if you map a client file directly to a visual tool and drill-through, the results may be quite different than if you, for example, removed a column, then mapped it to the visual tool and did drill-through. +o During the "Show Original Data" aspect of drill- through, if the current Tool Manager is connected to the necessary server, it uses that connection to retrieve the data. If the drill-through requires a different server, or server login, it will create a new connection. If the Tool Manager is connected to the same server but a different DBMS than that used in the drill-through, it is possible to kill the server process currently in use by the Tool Manager. This will void the current transformation history. +o Also during the "Show Original Data" process, when specifying a login and password for a new server, if you change the server name to that of the server currently in use by the Tool Manager, it will ask if you want to kill the other datamove process. Killing it will cause Tool Manager to lose its main connection, voiding the current transformation history. +o Also during the "Show Original Data" process, the Tool Manager may ask you for the server/DBMS login and password each time. +o The mineset_batch program sometimes reports incorrect names for the visualization files created as the result of running the .mineset file. +o The mineset_batch program will not run correctly on a history whose destination is the Statistics Visualizer tool. +o If you try to run a visual tool that is not installed on your system, you get the message: "Cannot find the tool <null>". +o If you do drill-through, and the currently running Tool Manager is iconified, it does not raise itself. +o Tool Manager cannot support database tables whose columns require quoting (e.g., those columns with - 9 - spaces or special characters in their names). +o Trying to change the types of columns used for associations will cause errors when running AssocConvert. Similarly, adding new columns of string type will cause errors when running AssocConvert. +o When working with client files as the data source, the file name is shown in the top panel. If the file has a long name or path, the name may be cut off. Increase the window size to see more of the name. +o Visualization tool files sometimes contain absolute path names when referring to their schema/data files, making it impossible to copy them to new machines. +o If you are running IRIX 6.2, you must install patch 1840. Otherwise, the Tree Visualizer may fail with a core dump when trying to display the color chooser from the marks panel. +o If you create a new column using an expression and that expression causes an error for Datamove, deleting the column using the "remove column" button does not work. The only way to fix the error is to fix the expression or remove the operation that uses it through "Edit History". +o When entering integer numeric values in any dialog box, if you enter a number larger than 2^31 (slightly more than 2 billion), the number you typed will be replaced by a random value. There are no cases where such a large number would be appropriate, however. +o When creating classifiers using the Classifiers mining panel, the name of the classifier created has the same prefix as the current session. Thus, if your current session is called "cars.mineset", a decision tree classifier will be called "cars-dt.class", and an evidence classifier will be called "cars-evi.class". If you create a classifier, apply the classifier to the current table, and create the classifier again, the original classifier is overwritten. The new classifier requires columns not present in the original table, thus operations produce the error message: InstanceReader::construct_assim_map: attributes in original schema were unaccounted for. +o When working with data files containing arrays, the Tool Manager lets you change the type of the array - 10 - using the Change Types dialog. This causes an error for Datamove, however. +o In the Change Types dialog, if you select several columns with different types, the type menu only shows the type of the first column. If you wish to change all selected columns to that type, you must reselect the menu item to effect the change. The list of columns is updated to reflect the change. +o Intensive random clicking and dragging in the Apply Classifier dialog box can occasionally cause the Tool Manager to core dump. +o User changes to a .schema file can cause errors and rarely core dumps when loading .mineset files that refer to those .schema files. +o The Remove Column button requires a column to be selected in the Current Columns list. It is normally grayed out when no columns are selected, but sometimes it is active even when no columns are selected. Pressing it at this time has no effect. +o In the Binning dialog, in the panel for automatic binning, it lets you specify 0 as the number of bins to create. This should be disallowed; currently it results in the use of "Automatically choose number of bins" mode. +o The Tool Manager sometimes incorrectly thinks you have made changes that would require the current session file to be saved. 6.10 _K_n_o_w_n__P_r_o_b_l_e_m_s__w_i_t_h__m_i_n_e_s_e_t___b_a_t_c_h +o If the passwords are entered via the command line arguments, the arguments will be briefly visible to a user running ps(1). Mineset_batch will remove them from its argument list as quickly as possible, but there is a brief amount of time during which they will be visible. If this is of concern, the passwords should be read from standard input instead. 6.11 _K_n_o_w_n__P_r_o_b_l_e_m_s__w_i_t_h__t_h_e__R_e_c_o_r_d__V_i_e_w_e_r_. +o Attempting to expand the last column to the right will not always work correctly. - 11 - 6.12 _K_n_o_w_n__P_r_o_b_l_e_m_s__w_i_t_h__P_r_i_n_t_i_n_g_. +o Image files sent to printer may get mistaken for VRML 2.0 data file if CosmoWorld is installed. To fix this comment out the following line "CONVERT CosmoWorlds1.0Data AsciiTextFile COST 500" in "/usr/lib/filetype/install/cosmo_worlds.ftr"