Ug~ddlmZddlZddlZddlmZddlZddlZ ddl Z ddl m Z ddlmZddlmZ ddlmZ ddlmZ ddlZ ddlmZ ddlmZGdd eZ ejBd d Z"ejBd d d Z#e$ejJd ejLjODcgc] }e(| c}Z)GddZ*GddZ+dZ,dZ-dZ.dZ/dZ0dZ1dZ2dFdZ3dGdZ4dHdZ5dZ6dIdZ7dZ8dZ9d Z:e d!"d#Z;d$Zd'Z?d(Z@dJd)ZAe"d*ZBe"d+ZCe#d,ZDe#d-ZEe#d.ZFe#d/ZGe"d0ZHe"d1ZIe#d2ZJe#d3ZKe"d4ZLe#d5ZMe#d6ZNe"d7ZOe#d8ZPe#d9ZQe"d:ZRe#d;ZSe#d<ZTe#d=ZUe"d>ZVe"d?ZWe"d@ZXe"dAZYe"dBZZe"dCZ[e"dDZ\e"dEZ]y#e$rdZYwxYw#e$r edZYwxYw#e$rdZYwxYw#e$r edZY wxYw#e$r edZYwxYwcc}w)K) annotationsN)getmro)memoize) DataArray) RaggedDtype) GeometryDtypeceZdZdZy)VisibleDeprecationWarningzVisible deprecation warning. By default, python will not show deprecation warnings, so this class can be used when a very visible warning is helpful, for example because the usage is most likely a user bug. N)__name__ __module__ __qualname____doc__0lib/python3.12/site-packages/datashader/utils.pyr r +srr T)nopythonnogil)rrparallelz([0-9]+)\.([0-9]+)\.([0-9]+)c(eZdZdZdZdZdZdZy)ExprzBase class for expression-like objects. Implements hashing and equality checks. Subclasses should implement an ``inputs`` attribute/property, containing a tuple of everything that fully defines that expression. cJtt||jfSN)hashtype_hashable_inputsselfs r__hash__z Expr.__hash__DsT$Z!6!6!89::rctt|t|uxr!|j|jk(Sr)rrrothers r__eq__z Expr.__eq__Gs8T d5k)D%%'5+A+A+CC Erc||k( Srrr s r__ne__z Expr.__ne__Ks5=  rc6g}|jD]~}t|ttfr|j t |4t|t jr |j |jn|j |t |S)zt Return a version of the inputs tuple that is suitable for hashing and equality comparisons ) inputs isinstancelistsetappendtuplenpndarraytobytes)rresultips rrzExpr._hashable_inputsNsm ++B"tSk* eBi(B + bjjl+ b! V}rN)r r r rrr"r$rrrrrr=s ;E!rrc$eZdZdZdZddZdZy) DispatcherzSimple single dispatch.ci|_yr)_lookuprs r__init__zDispatcher.__init__as  rNc|fdSttrD]}j|||S|j<|S)z6Register dispatch of `func` on arguments of type `typ`c(j|Sr)register)frtyps rz%Dispatcher.register..gsT]]32r)r'r+r8r4)rr:functs`` rr8zDispatcher.registerdsL <2 2 c5 ! a& !%DLL  rc|j}t|}||vr|||g|i|St|ddD]}||vs|||g|i|cStdj |)NzNo dispatch for {0} type)r4rr TypeErrorformat)rheadrestkwargslkr:clss r__call__zDispatcher.__call__os\\4j "92c741$1&1 1#;qr?Cbyr#wt5d5f55#299#>??rr)r r r rr5r8rGrrrr2r2_s!  @rr2ctjj|}t|tjxr|tj j vS)zCheck if a datashape is numeric and real. Example ------- >>> isrealfloat('int32') False >>> isrealfloat('float64') True >>> isrealfloat('string') False >>> isrealfloat('complex64') False ) datashape predicateslaunderr'Unittypesetsfloatingdts r isrealfloatrQ}sA    % %b )B b).. ) ObI4F4F4O4O.OOrctjj|}t|tjxr|tj j vS)zCheck if a datashape is numeric and real. Example ------- >>> isreal('int32') True >>> isreal('float64') True >>> isreal('string') False >>> isreal('complex64') False )rIrJrKr'rLrMrealrOs risrealrTsA    % %b )B b).. ) KbI4F4F4K4K.KKrcXtt|j}|j||j d||j |}t j|}t j|d}||z}t j|d|jdS)aWnansum where all-NaN values remain NaNs. Note: In NumPy <=1.9 NaN is returned for slices that are all NaN, while later versions return 0. This function emulates the older behavior, which allows using NaN as a missing value indicator. Parameters ---------- array: Array to sum over axis: Axis to sum over raxis) r(rangendimremoveinsert transposer,isnanallwheresum)arrayrWT missing_vals all_empty set_to_zeros rnansum_missingrfs U5:: AHHTNHHQ OOA E88E?L|!,I)+K 88KE * . .A . 66rc|jdd\}}|jdd\}}||j}||j}|d|dz |dz z }|d|dz |dz z }||fS)zCalculate the resolution of xarray.DataArray raster and return it as the two-tuple (xres, yres). yres is positive if it is decreasing. Nrr?)shapedimsvalues) rasterhwydimxdimxcoordsycoordsxresyress rcalc_resrvs << DAqRS!JD$Tl!!GTl!!G BK'!* $Q /D AJ $Q /D :rc |ddkr|jn|j}|ddkr|jn|j}tjx}}tj x}}tj|dd|gd|d |ggdg} ddt |ft |dft |t |ffD]R\} } tj | tj| | dg\} } }| |kr| }| |kDr| }| |kr| }| |kDsQ| }T|ddz |ddz }}||z ||z||z ||zfS)aCalculate the bounding box of a raster, and return it in a four-element tuple: (xmin, ymin, xmax, ymax). This calculation assumes the raster is uniformly sampled (equivalent to a flat-earth assumption, for geographic data) so that an affine transform (using the "Augmented Matrix" approach) suffices: https://en.wikipedia.org/wiki/Affine_transformation#Augmented_matrix Parameters ---------- xs : numpy.array 1D NumPy array of floats representing the x-values of a raster. This likely originated from an xarray.DataArray or xarray.Dataset object (xr.open_rasterio). ys : numpy.array 1D NumPy array of floats representing the y-values of a raster. This likely originated from an xarray.DataArray or xarray.Dataset object (xr.open_rasterio). res : tuple Two-tuple (int, int) which includes x and y resolutions (aka "grid/cell sizes"), respectively. rr?)rxrx?)rrry@)maxminr,infralendot)xsysresxboundyboundxminyminxmaxymaxAbx_y_xy_xpadypads r calc_bboxrsU,Q!RVVXFQ!RVVXF&&D466'D4 CFRf-c!fWf-)+ ,BAs2w<#b'1BR7IJB&&RXXr2rl341a t8D t8D t8D t8DKQCF2I$D 9d4idDI 55rct|}t|dz }|j|j}}||z }||z|z ||z |z }}t ||z |zt ||z |z} } | | kr| | fS| | fS)ai Transform continuous start and end coordinates into array indices. Parameters ---------- start : float coordinate of the lower bound. end : float coordinate of the upper bound. coords : numpy.ndarray coordinate values along the axis. res : tuple Resolution along an axis (aka "grid/cell sizes") rz)r~absr|r{int) startendcoordsrsizehalfvminvmaxspansidxeidxs r get_indicesrs v;D s8B;Dvzz|$D 9DtD#d(4-3EeDj$&'c$h_)=$D d{Tz :rcB|r |ddddddf}|r |ddddddf}|S)N.rir)raxflipyflips r _flip_arrayr s5 c4R4l# c1ddl# LrcZ| t|}|j}|||dz }t|dtjrtjddnd}t|dtjrtjddnd}|d|k}|d|kD}t |||}|S)a  Reorients the array to a canonical orientation depending on whether the x and y-resolution values are positive or negative. Parameters ---------- raster : DataArray xarray DataArray to be reoriented res : tuple Two-tuple (int, int) which includes x and y resolutions (aka "grid/cell sizes"), respectively. layer : int Index of the raster layer to be reoriented (optional) Returns ------- array : numpy.ndarray Reoriented 2d NumPy ndarray r?rns)rvdatar'r, timedelta64r)rmrlayerrar0zeror1zerorrs r orient_arrayrs( {v KKE eAg(23q62>>(JR^^At $PQF(23q62>>(JR^^At $PQF FVOE FVOE ue ,E Lrc|jdd\}}|d||t|zz d||t|zz f}tjt |Dcgc]&}t |Dcgc]}||d||d|fc}(c}}}|dk(rtj |dS|dk(rtj |dS|dk(rtj|dS|dk(rtj|dS|d k(rtj|dS|d k(rtj|dS|d k(rtj|dStd cc}wcc}}w) z3Create downsampled aggregate factor in pixels unitsNmeanrrVr`r{r|medianstdvarzNInvalid 'how' downsample method. Options mean, sum, max, min, median, std, var) rjrr, concatenaterXnanmeannansumnanmaxnanmin nanmediannanstdnanvar ValueError) aggregatefactorhowrrcrarrjiconcats rdownsample_aggregater5sr __Ra FB ,r2F +,,.Dr2F 3C/D.DD EE ^^%*6]4%2&+6]4%2$AIvIqy&y$89%24%245F f}zz&q)) yya(( yya(( yya(( ||F++ yya(( yya(($% %'44sE +E?E E ctj|j}tj|j}|dk(r*|j|jdkDj }|dk(rtj |||dddf}nRtjdtj||z tj||j|zdddf}t|||fS)zHelper function similar to np.linspace which return values from aggregate min value to aggregate max value in either linear or log space. rlinearN)basenumdtype) r,rrlrrr|linspacelogspacelog1perr)rrrmax_valmin_valvalss rsummarize_aggregate_valuesrPs ii (()Gii (()G!|..!!3488: h{{7GS1$': AHHWw%67!#3")--14;;=A!GE T?GW ,,rcgfd}|S)z& simple arg caching decorator c6rd|k7r ||fdddS)Nrr?r)argsr9lasts rrzhold.._ls*tAw$AtHnDGAwrr)r9rrs` @rholdrfs D Hrcddlm}tjj |stj ||r |||}|j jtjj|||z|r|SdS)zPGiven a datashader Image object, saves it to a disk file in the requested formatr)set_backgroundN) datashader.transfer_functionsrospathexistsmkdirto_pilsavejoin)imgfilenamefmt_return export_path backgroundrs r export_imagerssf= 77>>+ & S*-JJLbggll;3?@3#t#rct|ttfrtj|}t|ttfrtj|}tj dz}||zdz }tj tjd|ztj zdz |ztj z }||fS)a Projects the given (longitude, latitude) values into Web Mercator coordinates (meters East of Greenwich and meters North of the Equator). Longitude and latitude can be provided as scalars, Pandas columns, or Numpy arrays, and will be returned in the same form. Lists or tuples will be converted to Numpy arrays. Examples: easting, northing = lnglat_to_meters(-74,40.71) easting, northing = lnglat_to_meters(np.array([-74]),np.array([40.71])) df=pandas.DataFrame(dict(longitude=np.array([-74]),latitude=np.array([40.71]))) df.loc[:, 'longitude'], df.loc[:, 'latitude'] = lnglat_to_meters(df.longitude,df.latitude) iRagf@Zgv@)r'r(r+r,rapilogtan) longitudelatitude origin_shifteastingnorthings rlnglat_to_metersrs")dE]+HHY' (T5M*88H%557?L,&.Gvvbffb8mruu4u<=>MPRPUPUUH X rcDt|jttjjsxt|jtj j js@trft|jtjjjr-|jj}tr*t|tjr|j}tr*t|tjr|j!}t#j$|}|jj&dk(r|j)d}t+j,dj/t1|jj|j}t+j|||jj2S|jj&dk(rMt5|jdd}| t7|}t+j8t+j:|St|jt<t>fr |jSt@r&t|jt@r |jSt*jBjE|j}|t*jFk(rt*jHn|}|t*jHt*jJfvrt+j8|S|S) zeReturn an object from datashader.datashape.coretypes given a column from a pandas dataframe. Uobjectz{} * {})rorderedMtzN)r)&r'rrpd CategoricalapitypesCategoricalDtypecudfcoredtypescat categoriesddIndexcompute to_pandasr,rakindastyperIdshaperAr~rgetattrstrOptionDateTimerrgpd_GeometryDtypeCTypefrom_numpy_dtypeobject_string datetime_)col pd_categoriesr cat_dshaperrs rdshape_from_pandas_helperrs 399d2>>#7#789 syy"&&,,"?"? @ Z 499+;+;+L+LM** *]BHH5)113M J}djj9)335MXXm,    C '#**84J%%i&6&6 "" #   '  $$Z*4-0WW__> > 3  SYYd + >RB 2 2b 9:: CII ]; <yy z#))5FGyy __ - -cii 8F!'9+<+>+?+?&@ A r#w}}bffll&C&C D2c7;;6 z8 {% 0B ==9++HJ -HR1 %be&9&9!&< => -    8-s BD(&D c tjtjt|jt|j zDcgc]}|t ||fc}zScc}w)z;Return a datashape.DataShape object given a xarray Dataset.)rIrrr( data_varsrr)xr_dsrs rdshape_from_xarray_datasetr)sf ==9++eoo&ell);;-;A %eAh /0;- -s A, ctj|jddz}tj|d<||ddtj||jd}tj|jd|jddzf}tj|dddf<||ddddf<t j ||jdS)aU Converts a set of multiple sequences (eg: time series), stored as a 2 dimensional numpy array into a pandas dataframe that can be plotted by datashader. The pandas dataframe eventually contains two columns ('x' and 'y') with the data. Each time series is separated by a row of NaNs. Discussion at: https://github.com/bokeh/datashader/issues/286#issuecomment-334619499 x_values: 1D numpy array with the values to be plotted on the x axis (eg: time) y_values: 2D numpy array with the sequences to be plotted of shape (num sequences X length of each sequence) rr?riN)rr)r,zerosrjnantiler DataFrameflatten)x_valuesy_valuesrrs r!dataframe_from_multiple_sequencesr2s xxq!A%&1 661R5 1Sb6 wwq(..#$1 xx"HNN1$5$9:;1ff1QU81QV9 ,,QQYY[1 22rcgd}|j|jd|fjtjddf}|\}}}||z ||z }}|d|dz|d|dzz } | dk\rgd}|jdd|f} | jdk(s| jtj} tj |j| d} | j tj| jdd| jd} tj| |j } t|jdkD} | s4|jd }|jddd fjd | |<| S) zkHelper for ``datashader.utils.mesh()``. Both arguments are assumed to be Pandas DataFrame objects. )rr?rrNrr?)rrr?int64rV)r)rlrr,r4rtakereshapeprodrjrr.rr~repeat)vertices simpliceswinding first_triabcp1p2 cross_product vertex_idxsrrverts_have_weights weight_cols r_pd_meshrGsl G 0 0G < C CBHH MrPQr QRIGAq! UAEBqEBqEMBqEBqEM1M""1g:.K    '!((2 778??Ka 8D << 2A/A ?D ,,tX%5%5 6CX--.2 &&q) #**1a4077:J Jrc$t|j|j}t|j|j}t t j t|d|zz dz}tj||}|S)ziHelper for ``datashader.utils.mesh()``. Both arguments are assumed to be Dask DataFrame objects. r5) chunksize) rGrr{ npartitionsrr,ceilr~r from_pandas)r:r;rapprox_npartitionsrIs r_dd_meshrN%s{ 8##%y'8'8': ;CX1193H3HIBGGCH*<(<=>BCI .. 2C Jrc|jjddk\sJd|jjddj dj }|sJdt |jdkDs#|jjddkDsJdtr@t|tjr&t|tjr t||St||S) zMerge vertices and simplices into a triangular mesh, suitable to be passed into the ``Canvas.trimesh()`` method via the ``mesh`` keyword-argument. Both arguments are assumed to be Dask DataFrame objects. r?r5zFAt least three vertex columns are required for the triangle definitionNcJtj|tjSr)r, issubdtypeintegerrOs rr;zmesh..Bs2==RZZ0rz^Simplices must be integral. You may consider casting simplices to integers with ".astype(int)"rzMIf no vertex weight column is provided, a triangle weight column is required.) rlrjrilocmapr^r~rrr'r.rNrG)r:r;simplices_all_intss rmeshrV7s    ! !! $ ):-9: )#))..r2660 ce 7!67  x 1 $ (8(8(>(>q(AA(EXWX E j2<<0Z 2<<5X),, Hi ((rc |r||i|S||Srr)r<rrDs rapplyrXRs T$V$$T{rc|dkxs|dkD S)zQ Equivalent to isnan for floats, but also numba compatible with integers rrvals risnullr\Ys q#C!G $$rc |dk(S)zO Check for -1 which is equivalent to NaN for some integer aggregations rirrZs risminus1r^as "9rc|j}|j}tjt|D](}t ||st ||r!||||<*y)zPFirst of 2 arrays but taking nans into account. Return the first array. Nravelnbpranger~r\retr!rs rnanfirst_in_placerfisT ))+C KKME YYs3x  #a&>&q"21XCF!rc|j}|j}tjt|D]}t ||r||||<y)zOLast of 2 arrays but taking nans into account. Return the first array. Nr`rds rnanlast_in_placerhusJ ))+C KKME YYs3x eAh1XCF!rc|j}|j}tjt|D]K}t ||rt ||r ||||<)t ||r8||||kDsD||||<My)zMax of 2 arrays but taking nans into account. Could use np.nanmax but would need to replace zeros with nans where both arrays are nans. Return the first array. Nr`rds rnanmax_in_placerjs} ))+C KKME YYs3x  #a&>%(#qAa!eAhQ&71XCF !rc|j}|j}tjt|D]K}t ||rt ||r ||||<)t ||r8||||ksD||||<My)zMin of 2 arrays but taking nans into account. Could use np.nanmin but would need to replace zeros with nans where both arrays are nans. Accepts 3D (ny, nx, ncat) and 2D (ny, nx) arrays. Return the first array. Nr`rds rnanmin_in_placerls} ))+C KKME YYs3x  #a&>%(#qAa!eAhQ&71XCF !rclt|}t|dz |dD] }||dz ||<|||<|dzS)a(Insert a value into a 1D array at a particular index, but before doing that shift the previous values along one to make room. For use in ``FloatingNReduction`` classes such as ``max_n`` and ``first_n`` which store ``n`` values per pixel. Parameters ---------- target : 1d numpy array Target pixel array. value : float Value to insert into target pixel array. index : int Index to insert at. Returns ------- Index beyond insertion, i.e. where the first shifted value now sits. r?ri)r~rX)targetvaluer!nrs rshift_and_insertrqsH, F A 1Q3r "1Q3Kq #F5M 19rct|}d}|D]:}t|ryt||D]}t||s|||<|dz}:<y)afSingle pixel implementation of nanfirst_n_in_place. ret_pixel and other_pixel are both 1D arrays of the same length. Walk along other_pixel a value at a time, find insertion index in ret_pixel and shift values along to insert. Next other_pixel value is inserted at a higher index, so this walks the two pixel arrays just once each. rr?N)r~r\rX ret_pixel other_pixelrpistart other_valuers r_nanfirst_n_implrxsY IA F" +  61%)A,'#.IaLqSF & #rc |j\}}}}tj|D]:}t|D]*}t|D]}t ||||f||||f,<yzN3d version of nanfirst_n_in_place_4d, taking arrays of shape (ny, nx, n). NrjrbrcrXrx rer!nynxncat_nrrrs rnanfirst_n_in_place_4drscyyBD" YYr]rAT{ Q3Yq!Sy1AB#rc|j\}}}tj|D](}t|D]}t |||f|||f*yrzr{rer!r}r~rrrs rnanfirst_n_in_place_3drsMJBB YYr]rA SAYad 4rct|}d}|D].}t|ryt||D]}t|||}.0y)aeSingle pixel implementation of nanlast_n_in_place. ret_pixel and other_pixel are both 1D arrays of the same length. Walk along other_pixel a value at a time, find insertion index in ret_pixel and shift values along to insert. Next other_pixel value is inserted at a higher index, so this walks the two pixel arrays just once each. rNr~r\rXrqrss r_nanlast_n_implrsI IA F" +  61%))[&I& #rc |j\}}}}tj|D]:}t|D]*}t|D]}t ||||f||||f,<yrzrjrbrcrXrr|s rnanlast_n_in_place_4drscyyBD" YYr]rAT{Aq#IaCi0@A#rc|j\}}}tj|D](}t|D]}t |||f|||f*y)zM3d version of nanlast_n_in_place_4d, taking arrays of shape (ny, nx, n). Nrrs rnanlast_n_in_place_3dr sMJBB YYr]rA C1IuQT{ 3rct|}d}|D]E}t|ryt||D]'}t||s |||kDst|||}EGy)adSingle pixel implementation of nanmax_n_in_place. ret_pixel and other_pixel are both 1D arrays of the same length. Walk along other_pixel a value at a time, find insertion index in ret_pixel and shift values along to insert. Next other_pixel value is inserted at a higher index, so this walks the two pixel arrays just once each. rNrrss r_nanmax_n_implr` IA F" +  61%)A,';1+E-iaHF& #rc |j\}}}}tj|D]:}t|D]*}t|D]}t ||||f||||f,<y)aCombine two max-n arrays, taking nans into account. Max-n arrays are 4D with shape (ny, nx, ncat, n) where ny and nx are the number of pixels, ncat the number of categories (will be 1 if not using a categorical reduction) and the last axis containing n values in descending order. If there are fewer than n values it is padded with nans. Return the first array. NrjrbrcrXrr|s rnanmax_n_in_place_4dr-cyyBD" YYr]rAT{s1a9~uQ3Y/?@#rc|j\}}}tj|D](}t|D]}t |||f|||f*y)zL3d version of nanmax_n_in_place_4d, taking arrays of shape (ny, nx, n). Nrrs rnanmax_n_in_place_3dr=MJBB YYr]rA 3q!t9eAqDk 2rct|}d}|D]E}t|ryt||D]'}t||s |||kst|||}EGy)adSingle pixel implementation of nanmin_n_in_place. ret_pixel and other_pixel are both 1D arrays of the same length. Walk along other_pixel a value at a time, find insertion index in ret_pixel and shift values along to insert. Next other_pixel value is inserted at a higher index, so this walks the two pixel arrays just once each. rNrrss r_nanmin_n_implrGrrc |j\}}}}tj|D]:}t|D]*}t|D]}t ||||f||||f,<y)aCombine two min-n arrays, taking nans into account. Min-n arrays are 4D with shape (ny, nx, ncat, n) where ny and nx are the number of pixels, ncat the number of categories (will be 1 if not using a categorical reduction) and the last axis containing n values in ascending order. If there are fewer than n values it is padded with nans. Return the first array. NrjrbrcrXrr|s rnanmin_n_in_place_4dr]rrc|j\}}}tj|D](}t|D]}t |||f|||f*y)zL3d version of nanmin_n_in_place_4d, taking arrays of shape (ny, nx, n). Nrrs rnanmin_n_in_place_3drmrrc|j}|j}tjt|D]G}t ||rt ||r ||||<)t ||r8||xx||z cc<Iy)zSum of 2 arrays but taking nans into account. Could use np.nansum but would need to replace zeros with nans where both arrays are nans. Return the first array. Nr`rds rnansum_in_placerwss ))+C KKME YYs3x  #a&>%(#qAa! FeAh F !rc|j}|j}tt|D]'}||dkDs ||dk(s ||||kDs ||||<)y)zMaximum of 2 arrays of row indexes. Row indexes are integers from 0 upwards, missing data is -1. Return the first array. riNrarXr~rds rrow_max_in_placer] ))+C KKME 3s8_ 8b=c!fleAhQ.?1XCFrc|j}|j}tt|D]'}||dkDs ||dk(s ||||ks ||||<)y)zMinimum of 2 arrays of row indexes. Row indexes are integers from 0 upwards, missing data is -1. Return the first array. riNrrds rrow_min_in_placerrrct|}d}|D]9}|dk(ryt||D]!}||dk(s |||kDst|||}9;y)aeSingle pixel implementation of row_max_n_in_place. ret_pixel and other_pixel are both 1D arrays of the same length. Walk along other_pixel a value at a time, find insertion index in ret_pixel and shift values along to insert. Next other_pixel value is inserted at a higher index, so this walks the two pixel arrays just once each. rriNr~rXrqrss r_row_max_n_implr` IA F" "  61%Q<2%y|)C-iaHF& #rc |j\}}}}t|D]:}t|D]*}t|D]}t||||f||||f,<y)zCombine two row_max_n signed integer arrays. Equivalent to nanmax_n_in_place with -1 replacing NaN for missing data. Return the first array. NrjrXrr|s rrow_max_n_in_place_4dr_ yyBD" 2YrAT{Aq#IaCi0@A#rc|j\}}}t|D](}t|D]}t|||f|||f*yrrrs rrow_max_n_in_place_3drGJBB 2YrA C1IuQT{ 3rct|}d}|D]9}|dk(ryt||D]!}||dk(s |||kst|||}9;y)aeSingle pixel implementation of row_min_n_in_place. ret_pixel and other_pixel are both 1D arrays of the same length. Walk along other_pixel a value at a time, find insertion index in ret_pixel and shift values along to insert. Next other_pixel value is inserted at a higher index, so this walks the two pixel arrays just once each. rriNrrss r_row_min_n_implrrrc |j\}}}}t|D]:}t|D]*}t|D]}t||||f||||f,<y)zCombine two row_min_n signed integer arrays. Equivalent to nanmin_n_in_place with -1 replacing NaN for missing data. Return the first array. NrjrXrr|s rrow_min_n_in_place_4drrrc|j\}}}t|D](}t|D]}t|||f|||f*yrrrs rrow_min_n_in_place_3drrr)NN)r)r)z.pngT.r)^ __future__rrreinspectrnumbarbnumpyr,pandasrtoolzrxarrayrdatashader.datashaperIdask.dataframe dataframer ImportErrordatashader.datatypesrrr Exceptiongeopandas.arrayrr spatialpandas.geometry UserWarningr jitngjitngjit_parallelr+match __version__groupsr numba_versionrr2rQrTrfrvrrrrrrrrrrrr%r)r2rGrNrVrXr\r^rfrhrjrlrqrxrrrrrrrrrrrrrrrrrrrr)rs0rrs#" (0#B4   D)TDAxrxx;NN(,,2FH(56(5!s1v(567 D@@<P$L$7. )6X4B%6-,   $:(V=  @A  B  3><$)6%%     8.CC55*BB44* A A33* A A33      * B B44* B B44{ B t*K  D #T # JM"6s^HH$H8II7I.H! H!$ H54H58II II I+*I+