def __init__(self, name, mapset=None, mode='r', stack=None, position='last', **kwargs): super(MVector, self).__init__(name, mapset, **kwargs) assert self.exist(), "%r not found" % name if mapset: self._set_mapset(mapset) else: self._get_mapset() glist = g.list(type='vector', pattern=name, stderr_=PIPE, stdout_=PIPE) assert len(glist.outputs.stdout.split()) == 1,\ "%r found in multiple mapsets. Specify mapset argument." % name path = os.path.join('$GISDBASE','$LOCATION_NAME', self.mapset,'sqlite','sqlite.db') self._con = sql.connect(gtable.get_path(path)) self.region = Region() self.open(mode) #TODO: add reset method to stack self.stack = dict() if stack is None: self.stack = ListDict({0: {'type': 'v', 'map': self.full_name}}) elif isinstance(stack, dict): stack = ListDict(stack) if isinstance(position, int): if position in stack.keys(): stack.insert_before(position, (max(stack.keys())+1, {'type': 'v', 'map': self.full_name})) else: pass elif position is 'last': stack.insert_after(stack.keys()[-1], (max(stack.keys())+1, {'type': 'v', 'map' : self.full_name})) elif position is 'first': stack.insert_before(stack.keys()[0], (max(stack.keys())+1, {'type': 'v', 'map' : self.full_name})) self.stack = stack
def db_path(): ''' :returns path to grass sqlite database ''' # get this path from gproject path = '$GISDBASE/$LOCATION_NAME/$MAPSET/sqlite/sqlite.db' return table.get_path(path)
def extendLine(map, map_out, maxlen=200, scale=0.5, debug=False, verbose=1): # # map=Input map name # map_out=Output map with extensions # maxlen=Max length in map units that line can be extended (def=200) # scale=Maximum length of extension as proportion of original line, disabled if 0 (def=0.5) # vlen=number of verticies to look back in calculating line end direction (def=1) # Not sure if it is worth putting this in as parameter. # allowOverwrite = os.getenv('GRASS_OVERWRITE', '0') == '1' grass.info("map={}, map_out={}, maxlen={}, scale={}, debug={}".format( map, map_out, maxlen, scale, debug)) vlen = 1 # not sure if this is worth putting in as parameter cols = [(u'cat', 'INTEGER PRIMARY KEY'), (u'parent', 'INTEGER'), (u'dend', 'TEXT'), (u'orgx', 'DOUBLE PRECISION'), (u'orgy', 'DOUBLE PRECISION'), (u'search_len', 'DOUBLE PRECISION'), (u'search_az', 'DOUBLE PRECISION'), (u'best_xid', 'INTEGER'), (u'near_x', 'DOUBLE PRECISION'), (u'near_y', 'DOUBLE PRECISION'), (u'other_cat', 'INTEGER'), (u'xtype', 'TEXT'), (u'x_len', 'DOUBLE PRECISION')] extend = VectorTopo('extend') if extend.exist(): extend.remove() extend.open('w', tab_name='extend', tab_cols=cols) # # Go through input map, looking at each line and it's two nodes to find nodes # with only a single line starting/ending there - i.e. a dangle. # For each found, generate an extension line in the new map "extend" # inMap = VectorTopo(map) inMap.open('r') dangleCnt = 0 tickLen = len(inMap) grass.info("Searching {} features for dangles".format(tickLen)) ticker = 0 grass.message("Percent complete...") for ln in inMap: ticker = (ticker + 1) grass.percent(ticker, tickLen, 5) if ln.gtype == 2: # Only process lines for nd in ln.nodes(): if nd.nlines == 1: # We have a dangle dangleCnt = dangleCnt + 1 vtx = min(len(ln) - 1, vlen) if len([1 for _ in nd.lines(only_out=True) ]) == 1: # Dangle starting at node dend = "head" sx = ln[0].x sy = ln[0].y dx = sx - ln[vtx].x dy = sy - ln[vtx].y else: # Dangle ending at node dend = "tail" sx = ln[-1].x sy = ln[-1].y dx = sx - ln[-(vtx + 1)].x dy = sy - ln[-(vtx + 1)].y endaz = math.atan2(dy, dx) if scale > 0: extLen = min(ln.length() * scale, maxlen) else: extLen = maxlen ex = extLen * math.cos(endaz) + sx ey = extLen * math.sin(endaz) + sy extLine = geo.Line([(sx, sy), (ex, ey)]) quiet = extend.write(extLine, (ln.cat, dend, sx, sy, extLen, endaz, 0, 0, 0, 0, 'null', extLen)) grass.info( "{} dangle nodes found, committing table extend".format(dangleCnt)) extend.table.conn.commit() extend.close(build=True, release=True) inMap.close() # # Create two tables where extensions intersect; # 1. intersect with original lines # 2. intersect with self - to extract intersects between extensions # # First the intersects with original lines grass.info( "Searching for intersects between potential extensions and original lines" ) table_isectIn = Table('isectIn', connection=sqlite3.connect(get_path(path))) if table_isectIn.exist(): table_isectIn.drop(force=True) run_command("v.distance", flags='a', overwrite=True, quiet=True, from_="extend", from_type="line", to=map, to_type="line", dmax="0", upload="cat,dist,to_x,to_y", column="near_cat,dist,nx,ny", table="isectIn") # Will have touched the dangle it comes from, so remove those touches run_command( "db.execute", sql= "DELETE FROM isectIn WHERE rowid IN (SELECT isectIn.rowid FROM isectIn INNER JOIN extend ON from_cat=cat WHERE near_cat=parent)", driver="sqlite", database="$GISDBASE/$LOCATION_NAME/$MAPSET/sqlite/sqlite.db") run_command("db.execute", sql="ALTER TABLE isectIn ADD ntype VARCHAR", driver="sqlite", database="$GISDBASE/$LOCATION_NAME/$MAPSET/sqlite/sqlite.db") run_command("db.execute", sql="UPDATE isectIn SET ntype = 'orig' ", driver="sqlite", database="$GISDBASE/$LOCATION_NAME/$MAPSET/sqlite/sqlite.db") # # Now second self intersect table # grass.info("Searching for intersects of potential extensions") table_isectX = Table('isectX', connection=sqlite3.connect(get_path(path))) if table_isectX.exist(): table_isectX.drop(force=True) run_command("v.distance", flags='a', overwrite=True, quiet=True, from_="extend", from_type="line", to="extend", to_type="line", dmax="0", upload="cat,dist,to_x,to_y", column="near_cat,dist,nx,ny", table="isectX") # Obviously all extensions will intersect with themself, so remove those "intersects" run_command("db.execute", sql="DELETE FROM isectX WHERE from_cat = near_cat", driver="sqlite", database="$GISDBASE/$LOCATION_NAME/$MAPSET/sqlite/sqlite.db") run_command("db.execute", sql="ALTER TABLE isectX ADD ntype VARCHAR", driver="sqlite", database="$GISDBASE/$LOCATION_NAME/$MAPSET/sqlite/sqlite.db") run_command("db.execute", sql="UPDATE isectX SET ntype = 'ext' ", driver="sqlite", database="$GISDBASE/$LOCATION_NAME/$MAPSET/sqlite/sqlite.db") # # Combine the two tables and add a few more attributes # run_command("db.execute", sql="INSERT INTO isectIn SELECT * FROM isectX", driver="sqlite", database="$GISDBASE/$LOCATION_NAME/$MAPSET/sqlite/sqlite.db") cols_isectIn = Columns('isectIn', connection=sqlite3.connect(get_path(path))) cols_isectIn.add(['from_x'], ['DOUBLE PRECISION']) cols_isectIn.add(['from_y'], ['DOUBLE PRECISION']) cols_isectIn.add(['ext_len'], ['DOUBLE PRECISION']) # Get starting coordinate at the end of the dangle run_command( "db.execute", sql= "UPDATE isectIn SET from_x = (SELECT extend.orgx FROM extend WHERE from_cat=extend.cat)", driver="sqlite", database="$GISDBASE/$LOCATION_NAME/$MAPSET/sqlite/sqlite.db") run_command( "db.execute", sql= "UPDATE isectIn SET from_y = (SELECT extend.orgy FROM extend WHERE from_cat=extend.cat)", driver="sqlite", database="$GISDBASE/$LOCATION_NAME/$MAPSET/sqlite/sqlite.db") table_isectIn.conn.commit() # For each intersect point, calculate the distance along extension line from end of dangle # Would be nicer to do this in the database but SQLite dosen't support sqrt or exponents grass.info( "Calculating distances of intersects along potential extensions") cur = table_isectIn.execute( sql_code="SELECT rowid, from_x, from_y, nx, ny FROM isectIn") for row in cur.fetchall(): rowid, fx, fy, nx, ny = row x_len = math.sqrt((fx - nx)**2 + (fy - ny)**2) sqlStr = "UPDATE isectIn SET ext_len={:.8f} WHERE rowid={:d}".format( x_len, rowid) table_isectIn.execute(sql_code=sqlStr) grass.verbose("Ready to commit isectIn changes") table_isectIn.conn.commit() # Remove any zero distance from end of their dangle. # This happens when another extension intersects exactly at that point run_command("db.execute", sql="DELETE FROM isectIn WHERE ext_len = 0.0", driver="sqlite", database="$GISDBASE/$LOCATION_NAME/$MAPSET/sqlite/sqlite.db") table_isectIn.conn.commit() # Go through the extensions and find the intersect closest to each origin. grass.info("Searching for closest intersect for each potential extension") # db.execute sql="ALTER TABLE extend_t1 ADD COLUMN bst INTEGER" # db.execute sql="ALTER TABLE extend_t1 ADD COLUMN nrx DOUBLE PRECISION" # db.execute sql="ALTER TABLE extend_t1 ADD COLUMN nry DOUBLE PRECISION" # db.execute sql="ALTER TABLE extend_t1 ADD COLUMN ocat TEXT" # run_command("db.execute", # sql = "INSERT OR REPLACE INTO extend_t1 (bst, nrx, nry, ocat) VALUES ((SELECT isectIn.rowid, ext_len, nx, ny, near_cat, ntype FROM isectIn WHERE from_cat=extend_t1.cat ORDER BY ext_len ASC LIMIT 1))", # driver = "sqlite", # database = "$GISDBASE/$LOCATION_NAME/$MAPSET/sqlite/sqlite.db") grass.verbose("CREATE index") run_command("db.execute", sql="CREATE INDEX idx_from_cat ON isectIn (from_cat)", driver="sqlite", database="$GISDBASE/$LOCATION_NAME/$MAPSET/sqlite/sqlite.db") grass.verbose("UPDATE best_xid") run_command( "db.execute", sql= "UPDATE extend SET best_xid = (SELECT isectIn.rowid FROM isectIn WHERE from_cat=extend.cat ORDER BY ext_len ASC LIMIT 1)", driver="sqlite", database="$GISDBASE/$LOCATION_NAME/$MAPSET/sqlite/sqlite.db") grass.verbose("UPDATE x_len") run_command( "db.execute", sql= "UPDATE extend SET x_len = (SELECT ext_len FROM isectIn WHERE from_cat=extend.cat ORDER BY ext_len ASC LIMIT 1)", driver="sqlite", database="$GISDBASE/$LOCATION_NAME/$MAPSET/sqlite/sqlite.db") grass.verbose("UPDATE near_x") run_command( "db.execute", sql= "UPDATE extend SET near_x = (SELECT nx FROM isectIn WHERE from_cat=extend.cat ORDER BY ext_len ASC LIMIT 1)", driver="sqlite", database="$GISDBASE/$LOCATION_NAME/$MAPSET/sqlite/sqlite.db") grass.verbose("UPDATE near_y") run_command( "db.execute", sql= "UPDATE extend SET near_y = (SELECT ny FROM isectIn WHERE from_cat=extend.cat ORDER BY ext_len ASC LIMIT 1)", driver="sqlite", database="$GISDBASE/$LOCATION_NAME/$MAPSET/sqlite/sqlite.db") grass.verbose("UPDATE other_cat") run_command( "db.execute", sql= "UPDATE extend SET other_cat = (SELECT near_cat FROM isectIn WHERE from_cat=extend.cat ORDER BY ext_len ASC LIMIT 1)", driver="sqlite", database="$GISDBASE/$LOCATION_NAME/$MAPSET/sqlite/sqlite.db") grass.verbose("UPDATE xtype") run_command( "db.execute", sql= "UPDATE extend SET xtype = (SELECT ntype FROM isectIn WHERE from_cat=extend.cat ORDER BY ext_len ASC LIMIT 1)", driver="sqlite", database="$GISDBASE/$LOCATION_NAME/$MAPSET/sqlite/sqlite.db") grass.verbose("DROP index") run_command("db.execute", sql="DROP INDEX idx_from_cat", driver="sqlite", database="$GISDBASE/$LOCATION_NAME/$MAPSET/sqlite/sqlite.db") grass.verbose("CREATE index on near_cat") run_command("db.execute", sql="CREATE INDEX idx_near_cat ON isectIn (near_cat)", driver="sqlite", database="$GISDBASE/$LOCATION_NAME/$MAPSET/sqlite/sqlite.db") quiet = table_isectIn.filters.select('rowid', 'ext_len', 'nx', 'ny', 'near_cat', 'ntype') # quiet=table_isectIn.filters.order_by(['ext_len ASC']) quiet = table_isectIn.filters.order_by('ext_len ASC') quiet = table_isectIn.filters.limit(1) table_extend = Table('extend', connection=sqlite3.connect(get_path(path))) # Code below was relplaced by commands above untill memory problem can be sorted # table_extend.filters.select('cat') # cur=table_extend.execute() # updateCnt = 0 # for row in cur.fetchall(): # cat, = row # quiet=table_isectIn.filters.where('from_cat={:d}'.format(cat)) ##SELECT rowid, ext_len, nx, ny, near_cat, ntype FROM isectIn WHERE from_cat=32734 ORDER BY ext_len ASC LIMIT 1 # x_sect=table_isectIn.execute().fetchone() # if x_sect is not None: # x_rowid, ext_len, nx, ny, other_cat, ntype = x_sect # sqlStr="UPDATE extend SET best_xid={:d}, x_len={:.8f}, near_x={:.8f}, near_y={:.8f}, other_cat={:d}, xtype='{}' WHERE cat={:d}".format(x_rowid, ext_len, nx, ny, other_cat, ntype, cat) # table_extend.execute(sql_code=sqlStr) ## Try periodic commit to avoide crash! # updateCnt = (updateCnt + 1) % 10000 # if updateCnt == 0: # table_extend.conn.commit() grass.verbose("Ready to commit extend changes") table_extend.conn.commit() # # There may be extensions that crossed, and that intersection chosen by one but # not "recripricated" by the other. # Need to remove those possibilities and allow the jilted extension to re-search. # grass.verbose("Deleting intersects already resolved") run_command( "db.execute", sql= "DELETE FROM isectIn WHERE rowid IN (SELECT isectIn.rowid FROM isectIn JOIN extend ON near_cat=cat WHERE ntype='ext' AND xtype!='null')", #"AND from_cat!=other_cat" no second chance! driver="sqlite", database="$GISDBASE/$LOCATION_NAME/$MAPSET/sqlite/sqlite.db") table_isectIn.conn.commit() grass.verbose("Deleting complete") # To find the jilted - need a copy of extensions that have found an # intersection (won't overwrite so drop first) grass.verbose( "Re-searching for mis-matched intersects between potential extensions") table_imatch = Table('imatch', connection=sqlite3.connect(get_path(path))) if table_imatch.exist(): table_imatch.drop(force=True) wvar = "xtype!='null'" run_command( "db.copy", overwrite=True, quiet=True, from_driver="sqlite", from_database="$GISDBASE/$LOCATION_NAME/$MAPSET/sqlite/sqlite.db", from_table="extend", to_driver="sqlite", to_database="$GISDBASE/$LOCATION_NAME/$MAPSET/sqlite/sqlite.db", to_table="imatch", where=wvar) # Memory problems? if gc.isenabled(): grass.verbose("Garbage collection enabled - forcing gc cycle") gc.collect() else: grass.verbose("Garbage collection not enabled") # Ensure tables are commited table_extend.conn.commit() table_imatch.conn.commit() table_isectIn.conn.commit() # Identify the jilted sqlStr = "SELECT extend.cat FROM extend JOIN imatch ON extend.other_cat=imatch.cat WHERE extend.xtype='ext' and extend.cat!=imatch.other_cat" cur = table_extend.execute(sql_code=sqlStr) updateCnt = 0 for row in cur.fetchall(): cat, = row grass.verbose("Reworking extend.cat={}".format(cat)) quiet = table_isectIn.filters.where('from_cat={:d}'.format(cat)) #print("SQL: {}".format(table_isectIn.filters.get_sql())) x_sect = table_isectIn.execute().fetchone( ) ## Problem here under modules if x_sect is None: sqlStr = "UPDATE extend SET best_xid=0, x_len=search_len, near_x=0, near_y=0, other_cat=0, xtype='null' WHERE cat={:d}".format( cat) else: x_rowid, ext_len, nx, ny, other_cat, ntype = x_sect sqlStr = "UPDATE extend SET best_xid={:d}, x_len={:.8f}, near_x={:.8f}, near_y={:.8f}, other_cat={:d}, xtype='{}' WHERE cat={:d}".format( x_rowid, ext_len, nx, ny, other_cat, ntype, cat) table_extend.execute(sql_code=sqlStr) ## Try periodic commit to avoide crash! updateCnt = (updateCnt + 1) % 100 if (updateCnt == 0): # or (cat == 750483): grass.verbose( "XXXXXXXXXXX Committing table_extend XXXXXXXXXXXXXXXXXXXXXX") table_extend.conn.commit() grass.verbose("Committing adjustments to table extend") table_extend.conn.commit() # # For debugging, create a map with the chosen intersect points # if debug: wvar = "xtype!='null' AND x_len!=0" # print(wvar) run_command( "v.in.db", overwrite=True, quiet=True, table="extend", driver="sqlite", database="$GISDBASE/$LOCATION_NAME/$MAPSET/sqlite/sqlite.db", x="near_x", y="near_y", key="cat", where=wvar, output="chosen") # # Finally adjust the dangle lines in input map - use a copy (map_out) if requested # if map_out: run_command("g.copy", overwrite=allowOverwrite, quiet=True, vector=map + "," + map_out) else: # Otherwise just modify the original dataset (map) if allowOverwrite: grass.warning("Modifying vector map ({})".format(map)) map_out = map else: grass.error( "Use switch --o to modifying input vector map ({})".format( map)) return 1 # # Get info for lines that need extending table_extend.filters.select( 'parent, dend, near_x, near_y, search_az, xtype') table_extend.filters.where("xtype!='null'") extLines = table_extend.execute().fetchall() cat_mods = [ext[0] for ext in extLines] tickLen = len(cat_mods) grass.info("Extending {} dangles".format(tickLen)) ticker = 0 grass.message("Percent complete...") # Open up the map_out copy (or the original) and work through looking for lines that need modifying inMap = VectorTopo(map_out) inMap.open('rw', tab_name=map_out) for ln_idx in range(len(inMap)): ln = inMap.read(ln_idx + 1) if ln.gtype == 2: # Only process lines while ln.cat in cat_mods: # Note: could be 'head' and 'tail' ticker = (ticker + 1) grass.percent(ticker, tickLen, 5) cat_idx = cat_mods.index(ln.cat) cat, dend, nx, ny, endaz, xtype = extLines.pop(cat_idx) dump = cat_mods.pop(cat_idx) if xtype == 'orig': # Overshoot by 0.1 as break lines is unreliable nx = nx + 0.1 * math.cos(endaz) ny = ny + 0.1 * math.sin(endaz) newEnd = geo.Point(x=nx, y=ny, z=None) if dend == 'head': ln.insert(0, newEnd) else: # 'tail' ln.append(newEnd) quiet = inMap.rewrite(ln_idx + 1, ln) else: quite = inMap.delete(ln_idx + 1) ## Try periodic commit and garbage collection to avoide crash! if (ln_idx % 1000) == 0: # inMap.table.conn.commit() - no such thing - Why?? if gc.isenabled(): quiet = gc.collect() inMap.close(build=True, release=True) grass.message("v.extendlines completing") # # Clean up temporary tables and maps # if not debug: table_isectIn.drop(force=True) table_isectX.drop(force=True) table_imatch.drop(force=True) extend.remove() chosen = VectorTopo('chosen') if chosen.exist(): chosen.remove() return 0
class DBconnection(object): """Define a class to share common methods between TestCase.""" path = os.path.join(tmp.gettempdir(), randstr(prefix='temp', suffix='.db')) connection = sqlite3.connect(get_path(path)) for t in (np.int8, np.int16, np.int32, np.int64, np.uint8, np.uint16, np.uint32, np.uint64): sqlite3.register_adapter(t, int) columns = [('cat', 'INTEGER PRIMARY KEY'), ('cint', 'INT'), ('creal', 'REAL'), ('ctxt', 'TEXT')] def create_table_instance(self, **kw): """Return a Table class instance :param **kw: keyword arguments of Table class without name and connection. :type **kw: key-word arguments :returns: Table instance """ self.tname = randstr(prefix='temp') return Table(name=self.tname, connection=self.connection, **kw) def create_empty_table(self, columns=None, **kw): """Create an empty table in the database and return Table class instance. :param columns: list of tuple containing the column names and types. :type columns: list of tuple :param **kw: keyword arguments of Table class without name and connection. :type **kw: key-word arguments :returns: Table instance """ columns = self.columns if columns is None else columns table = self.create_table_instance(**kw) table.create(columns) return table def create_not_empty_table(self, nrows=None, values=None, columns=None, **kw): """Create a not empty table in the database and return Table class instance. :param nrows: number of rows. :type nrows: list of tuple :param values: list of tuple containing the values for each row. :type values: list of tuple :param columns: list of tuple containing the column names and types. :type columns: list of tuple :param **kw: keyword arguments of Table class without name and connection. :type **kw: key-word arguments :returns: Table instance """ if nrows is None and values is None: msg = "Both parameters ``nrows`` ``values`` are empty" raise RuntimeError(msg) columns = self.columns if columns is None else columns values = (get_table_random_values(nrows, columns) if values is None else values) table = self.create_empty_table(columns=columns, **kw) table.insert(values, many=True) return table def setUp(self): """Create a not empty table instance""" self.table = self.create_not_empty_table(10) self.cols = self.table.columns def tearDown(self): """Remove the generated vector map, if exist""" self.table.drop(force=True) self.table = None self.cols = None
def add_color_table(self, cbrew=False, color=None, column=None, attr=None, layer='1', nk=5, s=None, cramp=('Diverging','RdYlBu'), method='Equal_Interval', invert_ramp=False, *args, **kwargs): """ Add/replace color table Keyword Description ---------- ---------------------------------------------------------------------- nk number of classes column layer column to color classify, If s is not provided, column is needed. Default=None s pandas series or numpy array. If column is not provided, s is needed. Default=None ccramp Boolean. If true, color scheme and color ramp use Color Brewer (http://colorbrewer2.org/) method pysal.mapclassify methods. Options: Map_Classifier, Box_Plot, Equal_Interval, Fisher_Jenks, Jenks_Caspall, Jenks_Caspall_Forced, Jenks_Caspall_Sampled, Max_P_Classifier, Maximum_Breaks, Natural_Breaks, Quantiles, Percentiles, Std_Mean, User_Defined. Default User_Defined. See PySal documentation for valid parameters in each case (http://pysal.org/1.2/library/esda/mapclassify.html). invert_ramp Boolean. Default False. layer Grass layer where rgb column will be added *args Additional arguments for v.colors **kwargs Additional arguments for rgb_classify. For example, if method='User_Defined' then the arguments bins is required. ---------- ---------------------------------------------------------------------- """ #todo: allow editing if current mapset and mapset are PERMANENT #assert self.mapset is not 'PERMANENT',\ # '%r is in PERMANENT and is not writable by user' % self.name assert int(layer) <= self.dblinks.num_dblinks(), 'layer %r does not exist' % layer if cbrew: table_name = self.dblinks.by_layer(int(layer)).name sqlfilter = gtable.Table(name=table_name, connection=self._con) assert sqlfilter.exist(), 'table %r not found in mapset %r. Verify dblinks.'\ % (sqlfilter.name, self.mapset) if hasattr(s, '__len__'): dtype = s.dtype.type assert (np.issubdtype(dtype, np.integer) or np.issubdtype(dtype, np.float)),\ 'Numpy array dtype must be integer or float' sqlfilter.filters.select(sqlfilter.key) cur = sqlfilter.execute() cat = cur.fetchall() cat = np.array(cat) cat = cat.flatten() assert s.shape == cat.shape, 'Series does not match map key shape' if column: assert column in sqlfilter.columns.names(),\ '%r has not column %r' % (table_name, column) sqlfilter.filters.select(sqlfilter.key, column) cur = sqlfilter.execute() col = cur.fetchall() col = np.array(col).T s = col[1] cat = col[0] dtype = s.dtype.type if not (np.issubdtype(dtype, np.integer) or np.issubdtype(dtype, np.float)): nulls = np.where(s == '')[0] if nulls.any: s[s == ''] = np.nan s = s.astype(float) s = pd.Series(s) s_rgb = rgb_classify(s, nk, cramp=cramp, method=method, invert_ramp=invert_ramp, **kwargs) #TODO convert this into a save_colr_file function # Write color table mapset = Mapset() mapset.current() if self.mapset is mapset.name: filename = os.path.join('$GISDBASE','$LOCATION_NAME', self.mapset, 'vector', self.name, 'colr') else: filename = os.path.join('$GISDBASE','$LOCATION_NAME', mapset.name, 'vcolr2', self.mapset, self.name) filename = gtable.get_path(filename) directory = os.path.dirname(filename) if not os.path.exists(directory): os.makedirs(directory) with open(filename,'w') as f: f.write('% ' + '%r %r' % (int(min(cat)), self.table.n_rows()) + '\n') for i,item in enumerate(s_rgb): f.write(str(cat[i]) + ':' + item + '\n') else: assert color, 'color attribute is required' if column: v.colors(map=self.full_name, layer=layer, column=column, color=color, *args) else: v.colors(map=self.full_name, color=color, *args)