繪圖類庫(kù)：Folium、PlotNine、LeafMap、PyGal、CartoPy、floWeaver)
繼Python生態(tài)繪圖類庫(kù)Plotly、Altair、bqplot、plotlet之后本文繼續(xù)匯總Python生態(tài)下特定領(lǐng)域下的繪圖類庫(kù)。Folium項(xiàng)目主頁(yè)地圖畫圖工具開源GitHub7.4K Star2.3K Fork類庫(kù)基于Leaflet.js可在Jupyter Notebook里畫出可縮放、可拖拽、帶標(biāo)記點(diǎn)的交互式地圖。適用場(chǎng)景場(chǎng)景說明數(shù)據(jù)可視化報(bào)告在Jupyter里展示地理數(shù)據(jù)分布門店/網(wǎng)點(diǎn)分布圖批量打點(diǎn)直觀看到覆蓋范圍路線規(guī)劃展示把GPS軌跡畫出來熱力圖分析如人口密度、事故高發(fā)區(qū)實(shí)戰(zhàn)基于pip安裝pip install folium示例importfolium# 創(chuàng)建地圖定位到北京mfolium.Map(location[39.9042,116.4074],zoom_start12)# 加標(biāo)記點(diǎn)folium.Marker([39.9042,116.4074],popup天安門,iconfolium.Icon(colorred,iconinfo-sign)).add_to(m)# 保存HTML文件m.save(map.html)# 在Jupyter里直接顯示m批量打點(diǎn)importfoliumfromfolium.pluginsimportHeatMapimportpandasaspd# 門店數(shù)據(jù)storespd.DataFrame({name:[店A,店B,店C],lat:[39.90,39.92,39.88],lng:[116.40,116.42,116.38]})mfolium.Map(location[39.90,116.40],zoom_start13)for_,rowinstores.iterrows():folium.Marker([row[lat],row[lng]],popuprow[name]).add_to(m)# 用經(jīng)緯度數(shù)據(jù)生成熱力圖heat_data[[row[lat],row[lng]]for_,rowinstores.iterrows()]HeatMap(heat_data).add_to(m)m效果圖畫區(qū)域、畫路徑# 畫折線folium.PolyLine(locations[(39.90,116.40),(39.92,116.42),(39.88,116.38)],colorblue,weight5).add_to(m)# 畫圓如覆蓋范圍folium.Circle(location[39.90,116.40],radius1000,# 米colorred,fillTrue).add_to(m)展示效果PlotNine官網(wǎng)圖形語(yǔ)法繪圖開源GitHub4.8K Star254 Fork庫(kù)把ggplot2風(fēng)格的Grammar of Graphics帶到Python強(qiáng)調(diào)數(shù)據(jù)、映射、幾何對(duì)象、坐標(biāo)和分面等概念比先畫線再調(diào)樣式的過程更聲明式官方文檔。適合統(tǒng)計(jì)探索、分組對(duì)比、教學(xué)材料、希望用一致語(yǔ)法組織復(fù)雜統(tǒng)計(jì)圖、從R或ggplot2遷移來的團(tuán)隊(duì)以及需要統(tǒng)一圖形語(yǔ)法的分析腳本。不適合極復(fù)雜的交互式Web圖表或需要直接操控Matplotlib底層每個(gè)Artist的精細(xì)場(chǎng)景。特點(diǎn)語(yǔ)法清晰用aes、geom、facet、coord組合圖表表達(dá)意圖很明確適合統(tǒng)計(jì)圖分面、顏色映射、幾何層疊很自然承接PandasDataFrame字段直接進(jìn)入映射分析腳本里銜接順。實(shí)戰(zhàn)基于pip安裝pip install plotnine示例importpandasaspdfromplotnineimportaes,coord_flip,facet_wrap,geom_point,geom_col,ggplot,stat_smooth,theme_xkcd,theme_tufte,theme_minimalfromplotnine.dataimportmtcars(ggplot(mtcars,aes(wt,mpg,colorfactor(gear)))# 調(diào)用mtcars數(shù)據(jù)集以車重wt與每加侖行駛英里數(shù)mpg為坐標(biāo)顏色受gear影響geom_point())繪制效果圖(ggplot(mtcars,aes(wt,mpg,colorfactor(gear)))geom_point()stat_smooth(methodlm)# 繪制線性擬合曲線)效果圖需要等十幾秒鐘(ggplot(mtcars,aes(wt,mpg,colorfactor(gear)))geom_point()stat_smooth(methodlm)facet_wrap(gear)# 按gear進(jìn)行分組以多面展示數(shù)據(jù))效果圖(ggplot(mtcars,aes(wt,mpg,colorfactor(gear)))geom_point()stat_smooth(methodlm)facet_wrap(gear)theme_xkcd()# 繪制手繪風(fēng)格主題# theme_tufte() # Tufte 風(fēng)格主題# theme_minimal() # 系統(tǒng)默認(rèn)# theme_bw() # 系統(tǒng)默認(rèn))可能會(huì)輸出字體不存在渲染沒有什么較大影響findfont: Font family xkcd not found. findfont: Font family Humor Sans not found.不想看到警告信息添加兩行代碼importlogging logging.getLogger(matplotlib.font_manager).setLevel(logging.ERROR)渲染效果也可從xkcd-font官網(wǎng)下載xkcd.ttf字體文件。更多示例dfpd.DataFrame({x:[1,2,3],y:[2,5,3]})pggplot(df,aes(x,y))geom_point()figp.draw(showFalse)print(layers:,len(p.layers))print(axes:,len(fig.axes))dfpd.DataFrame({team:[A,B,C],score:[7,4,9]})pggplot(df,aes(team,score))geom_col(fill#4c78a8)coord_flip()figp.draw(showFalse)print(layers:,len(p.layers))print(coord:,type(p.coordinates).__name__)print(axes:,len(fig.axes))dfpd.DataFrame({x:[1,2,1,2],y:[3,5,2,7],team:[A,A,B,B]})pggplot(df,aes(x,y,colorteam))geom_point(size3)facet_wrap(team)figp.draw(showFalse)print(facets:,type(p.facet).__name__)print(panels:,len(fig.axes))解讀把不同統(tǒng)計(jì)視角拆成geom和facet比在一段imperative代碼里不斷改axes更容易維護(hù)。注意事項(xiàng)固定輸入字段、數(shù)據(jù)量范圍和缺失值策略明確圖表是靜態(tài)導(dǎo)出、Notebook交互、桌面GUI還是Web嵌入在目標(biāo)運(yùn)行環(huán)境里做一次渲染或?qū)С鲵?yàn)證LeafMap官網(wǎng)基于folium開源GitHub3.8K Star469 Fork零代碼地理數(shù)據(jù)分析與交互式繪圖包。能力超多后端隨意切換可切換到folium、ipyleaflet、kepler.gl、pydeck、bokeh等不同后端根據(jù)場(chǎng)景選最順手的工具交互式繪圖工具內(nèi)置圖形界面很多操作如加載數(shù)據(jù)、畫圖不用寫代碼在Jupyter里點(diǎn)點(diǎn)鼠標(biāo)就能完成分析能力強(qiáng)大集成WhiteboxTools分析工具庫(kù)可直接在地圖上進(jìn)行水文分析、地形分析、LiDAR數(shù)據(jù)處理等對(duì)新手和專家都友好初學(xué)者能快速上手出圖專業(yè)用戶也能用它處理復(fù)雜的空間數(shù)據(jù)和分析任務(wù)實(shí)戰(zhàn)基于pip安裝pip install leafmap示例PyGal官網(wǎng)SVG 圖表生成開源GitHub2.8K Star419 Fork。創(chuàng)建圖表對(duì)象、設(shè)置標(biāo)簽、添加series、渲染為SVG。相比大型交互式可視化框架更適合在文檔、郵件圖表、靜態(tài)頁(yè)面和報(bào)告里產(chǎn)出可縮放矢量圖、需要SVG文件的自動(dòng)化腳本以及需要輕量、可嵌入、無需復(fù)雜前端依賴時(shí)。適合在構(gòu)建流程里生成、緩存、嵌入而不是依賴運(yùn)行時(shí)再畫。不適合高度復(fù)雜的交互式分析平臺(tái)或需要海量數(shù)據(jù)點(diǎn)實(shí)時(shí)重繪的科學(xué)計(jì)算界面。特點(diǎn)SVG輸出圖表天然可縮放適合文檔和靜態(tài)頁(yè)面API簡(jiǎn)潔Bar、Line、Radar等圖表對(duì)象直接添加數(shù)據(jù)即可依賴輕不用完整前端框架也能生成可用圖表資產(chǎn)實(shí)戰(zhàn)基于pip安裝pip install pygal示例importpygal chartpygal.Bar()chart.titlesaleschart.x_labels[Jan,Feb,Mar]chart.add(sales,[120,160,90])svgchart.render()print(series:,len(chart.raw_series))print(svg bytes:,len(svg))chartpygal.Line(interpolatecubic)chart.add(A,[1,3,2,5])chart.add(B,[2,2,4,3])names[meta[title]fordata,metainchart.raw_series]svgchart.render()print(series names:,names)print(series count:,len(chart.raw_series))print(svg bytes:,len(svg))chartpygal.Radar()chart.x_labels[speed,quality,cost]chart.add(plan A,[8,7,5])chart.add(plan B,[6,9,7])svgchart.render(is_unicodeTrue)print(radar labels:,list(chart.x_labels))print(has svg:,svg.lstrip().startswith(?xml))CartoPy面向地圖繪制和地理投影的開源GitHub1.6K Star396 Fork庫(kù)常和Matplotlib、NumPy、GeoPandas一起使用。把經(jīng)緯度坐標(biāo)、投影坐標(biāo)、地圖要素和坐標(biāo)軸變換放進(jìn)同一套模型里管理。適合氣象圖、海洋航線、遙感輔助圖、區(qū)域統(tǒng)計(jì)地圖等場(chǎng)景下的可視化官方文檔。特點(diǎn)投影清楚CRS對(duì)象負(fù)責(zé)坐標(biāo)系統(tǒng)和變換適合在圖上混合經(jīng)緯度、墨卡托等坐標(biāo)要素可控Natural Earth要素可按類別和比例尺組織便于控制海岸線、陸地等素材生態(tài)兼容和Matplotlib配合緊密適合把科學(xué)計(jì)算結(jié)果落到地圖視圖里不適合只想快速在網(wǎng)頁(yè)里拖動(dòng)地圖的場(chǎng)景這類需求通常Leaflet、Mapbox或Folium更合適。實(shí)戰(zhàn)基于pip安裝pip install cartopy示例importcartopy.crsasccrsimportcartopy.featureascfeatureimportnumpyasnp srcccrs.PlateCarree()dstccrs.Mercator()ptsdst.transform_points(src,np.array([0,120]),np.array([0,30]))fmtlambdarow:[float(round(v,2))forvinrow[:2]]print(threshold:,round(src.threshold,3))print(points:,pts.shape)print(x0/y0:,fmt(pts[0]))print(x1/y1:,fmt(pts[1]))landcfeature.NaturalEarthFeature(physical,land,110m)print(feature:,land.name)print(category:,land.category)print(scale:,land.scale)解讀把經(jīng)緯度點(diǎn)轉(zhuǎn)換到Web/墨卡托類坐標(biāo)驗(yàn)證數(shù)據(jù)坐標(biāo)能進(jìn)入目標(biāo)投影定義Natural Earth地圖要素后續(xù)可交給Matplotlib軸對(duì)象渲染注意事項(xiàng)固定Python、核心庫(kù)和系統(tǒng)依賴版本避免環(huán)境漂移用小樣本覆蓋輸入邊界、空值、異常值和坐標(biāo)/數(shù)組維度把外部服務(wù)、投影數(shù)據(jù)或編譯依賴寫進(jìn)部署文檔floWeaver開源GitHub464 Star89 Fork流數(shù)據(jù)可視化可將類似于“源-目標(biāo)-數(shù)值”這種表單數(shù)據(jù)變成比例精準(zhǔn)、配色高級(jí)的?；鶊D官方文檔。組件floWeaver負(fù)責(zé)邏輯層ipysankeywidget負(fù)責(zé)在Jupyter里渲染圖表用ProcessGroup定義聚合類別用Waypoint標(biāo)記節(jié)點(diǎn)SankeyDefinition負(fù)責(zé)布局Weave負(fù)責(zé)數(shù)據(jù)映射。實(shí)戰(zhàn)基于pip安裝pip install floweaver示例fromfloweaverimportSankeyDefinition,WeavefromfloweaverimportProcessGroup,Waypointimportpandasaspd group_electProcessGroup(電子產(chǎn)品)group_homeProcessGroup(家居生活)group_clothProcessGroup(服飾美妝)w_electWaypoint(group_elect,title電子)w_homeWaypoint(group_home,title家居)w_clothWaypoint(group_cloth,title服裝)sankey_defSankeyDefinition(flows[(w_elect,w_home),(w_elect,w_cloth),(w_home,w_cloth)],waypoints[w_elect,w_home,w_cloth])datapd.DataFrame({source:[電子,電子,家居],target:[家居,服裝,服裝],value:[100,50,30]})weaveWeave(sankey_def,data,link_columns[source,target,value])weave.to_widget()sankey_def.flows[(w_elect,w_home),(w_home,w_cloth)]weave_newWeave(sankey_def,data,link_columns[source,target,value])weave_new.to_widget()TODO上述代碼在2.0.1版本跑不通。優(yōu)勢(shì)在于它把“流”的語(yǔ)法抽象得非常干凈幾行代碼就能出專業(yè)級(jí)桑基圖且配色省心。不足高度綁定ipysankeywidget只能用于Jupyter Notebook或Lab沒法直接導(dǎo)出PNG或用在Django網(wǎng)站上。 Python生態(tài)繪圖類庫(kù)：Folium、PlotNine、LeafMap、PyGal、CartoPy、floWeaver
繼Python生態(tài)繪圖類庫(kù)Plotly、Altair、bqplot、plotlet之后本文繼續(xù)匯總Python生態(tài)下特定領(lǐng)域下的繪圖類庫(kù)。Folium項(xiàng)目主頁(yè)地圖畫圖工具開源GitHub7.4K Star2.3K Fork類庫(kù)基于Leaflet.js可在Jupyter Notebook里畫出可縮放、可拖拽、帶標(biāo)記點(diǎn)的交互式地圖。適用場(chǎng)景場(chǎng)景說明數(shù)據(jù)可視化報(bào)告在Jupyter里展示地理數(shù)據(jù)分布門店/網(wǎng)點(diǎn)分布圖批量打點(diǎn)直觀看到覆蓋范圍路線規(guī)劃展示把GPS軌跡畫出來熱力圖分析如人口密度、事故高發(fā)區(qū)實(shí)戰(zhàn)基于pip安裝pip install folium示例importfolium# 創(chuàng)建地圖定位到北京mfolium.Map(location[39.9042,116.4074],zoom_start12)# 加標(biāo)記點(diǎn)folium.Marker([39.9042,116.4074],popup天安門,iconfolium.Icon(colorred,iconinfo-sign)).add_to(m)# 保存HTML文件m.save(map.html)# 在Jupyter里直接顯示m批量打點(diǎn)importfoliumfromfolium.pluginsimportHeatMapimportpandasaspd# 門店數(shù)據(jù)storespd.DataFrame({name:[店A,店B,店C],lat:[39.90,39.92,39.88],lng:[116.40,116.42,116.38]})mfolium.Map(location[39.90,116.40],zoom_start13)for_,rowinstores.iterrows():folium.Marker([row[lat],row[lng]],popuprow[name]).add_to(m)# 用經(jīng)緯度數(shù)據(jù)生成熱力圖heat_data[[row[lat],row[lng]]for_,rowinstores.iterrows()]HeatMap(heat_data).add_to(m)m效果圖畫區(qū)域、畫路徑# 畫折線folium.PolyLine(locations[(39.90,116.40),(39.92,116.42),(39.88,116.38)],colorblue,weight5).add_to(m)# 畫圓如覆蓋范圍folium.Circle(location[39.90,116.40],radius1000,# 米colorred,fillTrue).add_to(m)展示效果PlotNine官網(wǎng)圖形語(yǔ)法繪圖開源GitHub4.8K Star254 Fork庫(kù)把ggplot2風(fēng)格的Grammar of Graphics帶到Python強(qiáng)調(diào)數(shù)據(jù)、映射、幾何對(duì)象、坐標(biāo)和分面等概念比先畫線再調(diào)樣式的過程更聲明式官方文檔。適合統(tǒng)計(jì)探索、分組對(duì)比、教學(xué)材料、希望用一致語(yǔ)法組織復(fù)雜統(tǒng)計(jì)圖、從R或ggplot2遷移來的團(tuán)隊(duì)以及需要統(tǒng)一圖形語(yǔ)法的分析腳本。不適合極復(fù)雜的交互式Web圖表或需要直接操控Matplotlib底層每個(gè)Artist的精細(xì)場(chǎng)景。特點(diǎn)語(yǔ)法清晰用aes、geom、facet、coord組合圖表表達(dá)意圖很明確適合統(tǒng)計(jì)圖分面、顏色映射、幾何層疊很自然承接PandasDataFrame字段直接進(jìn)入映射分析腳本里銜接順。實(shí)戰(zhàn)基于pip安裝pip install plotnine示例importpandasaspdfromplotnineimportaes,coord_flip,facet_wrap,geom_point,geom_col,ggplot,stat_smooth,theme_xkcd,theme_tufte,theme_minimalfromplotnine.dataimportmtcars(ggplot(mtcars,aes(wt,mpg,colorfactor(gear)))# 調(diào)用mtcars數(shù)據(jù)集以車重wt與每加侖行駛英里數(shù)mpg為坐標(biāo)顏色受gear影響geom_point())繪制效果圖(ggplot(mtcars,aes(wt,mpg,colorfactor(gear)))geom_point()stat_smooth(methodlm)# 繪制線性擬合曲線)效果圖需要等十幾秒鐘(ggplot(mtcars,aes(wt,mpg,colorfactor(gear)))geom_point()stat_smooth(methodlm)facet_wrap(gear)# 按gear進(jìn)行分組以多面展示數(shù)據(jù))效果圖(ggplot(mtcars,aes(wt,mpg,colorfactor(gear)))geom_point()stat_smooth(methodlm)facet_wrap(gear)theme_xkcd()# 繪制手繪風(fēng)格主題# theme_tufte() # Tufte 風(fēng)格主題# theme_minimal() # 系統(tǒng)默認(rèn)# theme_bw() # 系統(tǒng)默認(rèn))可能會(huì)輸出字體不存在渲染沒有什么較大影響findfont: Font family xkcd not found. findfont: Font family Humor Sans not found.不想看到警告信息添加兩行代碼importlogging logging.getLogger(matplotlib.font_manager).setLevel(logging.ERROR)渲染效果也可從xkcd-font官網(wǎng)下載xkcd.ttf字體文件。更多示例dfpd.DataFrame({x:[1,2,3],y:[2,5,3]})pggplot(df,aes(x,y))geom_point()figp.draw(showFalse)print(layers:,len(p.layers))print(axes:,len(fig.axes))dfpd.DataFrame({team:[A,B,C],score:[7,4,9]})pggplot(df,aes(team,score))geom_col(fill#4c78a8)coord_flip()figp.draw(showFalse)print(layers:,len(p.layers))print(coord:,type(p.coordinates).__name__)print(axes:,len(fig.axes))dfpd.DataFrame({x:[1,2,1,2],y:[3,5,2,7],team:[A,A,B,B]})pggplot(df,aes(x,y,colorteam))geom_point(size3)facet_wrap(team)figp.draw(showFalse)print(facets:,type(p.facet).__name__)print(panels:,len(fig.axes))解讀把不同統(tǒng)計(jì)視角拆成geom和facet比在一段imperative代碼里不斷改axes更容易維護(hù)。注意事項(xiàng)固定輸入字段、數(shù)據(jù)量范圍和缺失值策略明確圖表是靜態(tài)導(dǎo)出、Notebook交互、桌面GUI還是Web嵌入在目標(biāo)運(yùn)行環(huán)境里做一次渲染或?qū)С鲵?yàn)證LeafMap官網(wǎng)基于folium開源GitHub3.8K Star469 Fork零代碼地理數(shù)據(jù)分析與交互式繪圖包。能力超多后端隨意切換可切換到folium、ipyleaflet、kepler.gl、pydeck、bokeh等不同后端根據(jù)場(chǎng)景選最順手的工具交互式繪圖工具內(nèi)置圖形界面很多操作如加載數(shù)據(jù)、畫圖不用寫代碼在Jupyter里點(diǎn)點(diǎn)鼠標(biāo)就能完成分析能力強(qiáng)大集成WhiteboxTools分析工具庫(kù)可直接在地圖上進(jìn)行水文分析、地形分析、LiDAR數(shù)據(jù)處理等對(duì)新手和專家都友好初學(xué)者能快速上手出圖專業(yè)用戶也能用它處理復(fù)雜的空間數(shù)據(jù)和分析任務(wù)實(shí)戰(zhàn)基于pip安裝pip install leafmap示例PyGal官網(wǎng)SVG 圖表生成開源GitHub2.8K Star419 Fork。創(chuàng)建圖表對(duì)象、設(shè)置標(biāo)簽、添加series、渲染為SVG。相比大型交互式可視化框架更適合在文檔、郵件圖表、靜態(tài)頁(yè)面和報(bào)告里產(chǎn)出可縮放矢量圖、需要SVG文件的自動(dòng)化腳本以及需要輕量、可嵌入、無需復(fù)雜前端依賴時(shí)。適合在構(gòu)建流程里生成、緩存、嵌入而不是依賴運(yùn)行時(shí)再畫。不適合高度復(fù)雜的交互式分析平臺(tái)或需要海量數(shù)據(jù)點(diǎn)實(shí)時(shí)重繪的科學(xué)計(jì)算界面。特點(diǎn)SVG輸出圖表天然可縮放適合文檔和靜態(tài)頁(yè)面API簡(jiǎn)潔Bar、Line、Radar等圖表對(duì)象直接添加數(shù)據(jù)即可依賴輕不用完整前端框架也能生成可用圖表資產(chǎn)實(shí)戰(zhàn)基于pip安裝pip install pygal示例importpygal chartpygal.Bar()chart.titlesaleschart.x_labels[Jan,Feb,Mar]chart.add(sales,[120,160,90])svgchart.render()print(series:,len(chart.raw_series))print(svg bytes:,len(svg))chartpygal.Line(interpolatecubic)chart.add(A,[1,3,2,5])chart.add(B,[2,2,4,3])names[meta[title]fordata,metainchart.raw_series]svgchart.render()print(series names:,names)print(series count:,len(chart.raw_series))print(svg bytes:,len(svg))chartpygal.Radar()chart.x_labels[speed,quality,cost]chart.add(plan A,[8,7,5])chart.add(plan B,[6,9,7])svgchart.render(is_unicodeTrue)print(radar labels:,list(chart.x_labels))print(has svg:,svg.lstrip().startswith(?xml))CartoPy面向地圖繪制和地理投影的開源GitHub1.6K Star396 Fork庫(kù)常和Matplotlib、NumPy、GeoPandas一起使用。把經(jīng)緯度坐標(biāo)、投影坐標(biāo)、地圖要素和坐標(biāo)軸變換放進(jìn)同一套模型里管理。適合氣象圖、海洋航線、遙感輔助圖、區(qū)域統(tǒng)計(jì)地圖等場(chǎng)景下的可視化官方文檔。特點(diǎn)投影清楚CRS對(duì)象負(fù)責(zé)坐標(biāo)系統(tǒng)和變換適合在圖上混合經(jīng)緯度、墨卡托等坐標(biāo)要素可控Natural Earth要素可按類別和比例尺組織便于控制海岸線、陸地等素材生態(tài)兼容和Matplotlib配合緊密適合把科學(xué)計(jì)算結(jié)果落到地圖視圖里不適合只想快速在網(wǎng)頁(yè)里拖動(dòng)地圖的場(chǎng)景這類需求通常Leaflet、Mapbox或Folium更合適。實(shí)戰(zhàn)基于pip安裝pip install cartopy示例importcartopy.crsasccrsimportcartopy.featureascfeatureimportnumpyasnp srcccrs.PlateCarree()dstccrs.Mercator()ptsdst.transform_points(src,np.array([0,120]),np.array([0,30]))fmtlambdarow:[float(round(v,2))forvinrow[:2]]print(threshold:,round(src.threshold,3))print(points:,pts.shape)print(x0/y0:,fmt(pts[0]))print(x1/y1:,fmt(pts[1]))landcfeature.NaturalEarthFeature(physical,land,110m)print(feature:,land.name)print(category:,land.category)print(scale:,land.scale)解讀把經(jīng)緯度點(diǎn)轉(zhuǎn)換到Web/墨卡托類坐標(biāo)驗(yàn)證數(shù)據(jù)坐標(biāo)能進(jìn)入目標(biāo)投影定義Natural Earth地圖要素后續(xù)可交給Matplotlib軸對(duì)象渲染注意事項(xiàng)固定Python、核心庫(kù)和系統(tǒng)依賴版本避免環(huán)境漂移用小樣本覆蓋輸入邊界、空值、異常值和坐標(biāo)/數(shù)組維度把外部服務(wù)、投影數(shù)據(jù)或編譯依賴寫進(jìn)部署文檔floWeaver開源GitHub464 Star89 Fork流數(shù)據(jù)可視化可將類似于“源-目標(biāo)-數(shù)值”這種表單數(shù)據(jù)變成比例精準(zhǔn)、配色高級(jí)的?；鶊D官方文檔。組件floWeaver負(fù)責(zé)邏輯層ipysankeywidget負(fù)責(zé)在Jupyter里渲染圖表用ProcessGroup定義聚合類別用Waypoint標(biāo)記節(jié)點(diǎn)SankeyDefinition負(fù)責(zé)布局Weave負(fù)責(zé)數(shù)據(jù)映射。實(shí)戰(zhàn)基于pip安裝pip install floweaver示例fromfloweaverimportSankeyDefinition,WeavefromfloweaverimportProcessGroup,Waypointimportpandasaspd group_electProcessGroup(電子產(chǎn)品)group_homeProcessGroup(家居生活)group_clothProcessGroup(服飾美妝)w_electWaypoint(group_elect,title電子)w_homeWaypoint(group_home,title家居)w_clothWaypoint(group_cloth,title服裝)sankey_defSankeyDefinition(flows[(w_elect,w_home),(w_elect,w_cloth),(w_home,w_cloth)],waypoints[w_elect,w_home,w_cloth])datapd.DataFrame({source:[電子,電子,家居],target:[家居,服裝,服裝],value:[100,50,30]})weaveWeave(sankey_def,data,link_columns[source,target,value])weave.to_widget()sankey_def.flows[(w_elect,w_home),(w_home,w_cloth)]weave_newWeave(sankey_def,data,link_columns[source,target,value])weave_new.to_widget()TODO上述代碼在2.0.1版本跑不通。優(yōu)勢(shì)在于它把“流”的語(yǔ)法抽象得非常干凈幾行代碼就能出專業(yè)級(jí)桑基圖且配色省心。不足高度綁定ipysankeywidget只能用于Jupyter Notebook或Lab沒法直接導(dǎo)出PNG或用在Django網(wǎng)站上。