From 9a90f21e5930995cea2a773474f1e8657ceb2355 Mon Sep 17 00:00:00 2001 From: sean-eagles <112004392+sean-eagles@users.noreply.github.com> Date: Fri, 10 Nov 2023 16:33:37 -0500 Subject: [PATCH] Merged Master Script for Python 2.7 scripts One script to rule them all... All of the Python 2.7 scripts have been merged together as their processes followed one after the other. --- MASTER_SCRIPT_Python2x.py | 354 ++++++++++++++++++++++++++++++++++++++ 1 file changed, 354 insertions(+) create mode 100644 MASTER_SCRIPT_Python2x.py diff --git a/MASTER_SCRIPT_Python2x.py b/MASTER_SCRIPT_Python2x.py new file mode 100644 index 0000000..140f638 --- /dev/null +++ b/MASTER_SCRIPT_Python2x.py @@ -0,0 +1,354 @@ +#------------------------------------------------------------------------------- +# Name: MASTER SCRIPT Python 2.7 +# Purpose: +# +# https://stackoverflow.com/questions/73306198/how-to-delete-a-particular-column-in-csv-file-without-pandas-library +# https://www.w3schools.com/python/pandas/pandas_cleaning_empty_cells.asp +# +#------------------------------------------------------------------------------- +# -*- coding: utf-8 -*- +# --------------------------------------------------------------------------- +# Description: + # Author: Sean Eagles + # takes an input CSV and runs through a geodatabase and back out to clean + # the table properly, simplify the output fields and make it work with + # Joins and relates. +# --------------------------------------------------------------------------- +import dbf +import arcpy +import csv +import pandas as pd +import urllib2 +from contextlib import closing +import geojson +import arcpy + +def main(): + pass + +def download_url(url, save_path): + with closing(urllib2.urlopen(url)) as dl_file: + with open(save_path, 'wb') as out_file: + out_file.write(dl_file.read()) + +# Parameters/Variables +# +csv_path = (r"C:\temp\records.csv") +OutFolderPath = "C:/TEMP/" +OutName = "Records.gdb" +InputCSV = OutFolderPath + "records_remove_empty.csv" +GeoDataBase = OutFolderPath + OutName +GDBTableName = "records_cleaned_of_columns" +GDBOutputTableName = "records_column_and_rows_clean" +GDBInputTableName = "records_column_and_rows_clean" +GDBTablePathAndName = GeoDataBase + "/" + GDBTableName +GDBTableOutputPathAndName = GeoDataBase + "/" + GDBOutputTableName +GDBTableInputPathAndName = GeoDataBase + "/" + GDBInputTableName +ExcelOutputTable = OutFolderPath + "records.xls" +DBFName = "records_clean.dbf" +data = "C:/TEMP/records.csv" +data_out = "C:/TEMP/records_remove_empty.csv" + +print "GeoDataBase path = " + GeoDataBase +print "Input CSV = " + InputCSV +print "GDB Name = " + OutName +print "Output Folder Path = " + OutFolderPath +print "Geodatabase Table Name (Cleaned columns) = " + GDBTableName +print "Geodatabase table path and name (Cleaned Columns) = " + GDBTablePathAndName +print "Geodatabase table output path and name (cleaned columns and rows) = " + GDBTableOutputPathAndName +print "Geodatabase table input path and name = " + GDBTableInputPathAndName +print "Excel output table (xls) = " + ExcelOutputTable +print csv_path + " = csv_path" +print "Data table input = " + data +print "Data table output = " + data_out + +if __name__ == '__main__': + + with open(csv_path) as instream: + # Setup the input + reader = csv.DictReader(instream) + rows = list(reader) + + # Setup the output fields + output_fields = reader.fieldnames + output_fields.remove("features_type") + output_fields.remove("features_geometry_type") + output_fields.remove("features_geometry_coordinates") + output_fields.remove("features_properties_title_en") + output_fields.remove("features_properties_title_fr") + output_fields.remove("features_properties_description_en") + output_fields.remove("features_properties_description_fr") + output_fields.remove("features_properties_keywords_en") + output_fields.remove("features_properties_keywords_fr") + output_fields.remove("features_properties_topicCategory") + output_fields.remove("features_properties_parentIdentifier") + output_fields.remove("features_properties_date_published_text") + output_fields.remove("features_properties_date_published_date") + output_fields.remove("features_properties_date_created_text") + output_fields.remove("features_properties_date_created_date") + output_fields.remove("features_properties_date_revision_text") + output_fields.remove("features_properties_date_revision_date") + output_fields.remove("features_properties_date_notavailable_text") + output_fields.remove("features_properties_date_notavailable_date") + output_fields.remove("features_properties_date_inforce_text") + output_fields.remove("features_properties_date_inforce_date") + output_fields.remove("features_properties_date_adopted_text") + output_fields.remove("features_properties_date_adopted_date") + output_fields.remove("features_properties_date_deprecated_text") + output_fields.remove("features_properties_date_deprecated_date") + output_fields.remove("features_properties_date_superceded_text") + output_fields.remove("features_properties_date_superceded_date") + output_fields.remove("features_properties_spatialRepresentation") + output_fields.remove("features_properties_type") + output_fields.remove("features_properties_geometry") + output_fields.remove("features_properties_temporalExtent_begin") + output_fields.remove("features_properties_temporalExtent_end") + output_fields.remove("features_properties_refSys") + output_fields.remove("features_properties_refSys_version") + output_fields.remove("features_properties_status") + output_fields.remove("features_properties_maintenance") + output_fields.remove("features_properties_metadataStandard_en") + output_fields.remove("features_properties_metadataStandard_fr") + output_fields.remove("features_properties_metadataStandardVersion") + output_fields.remove("features_properties_otherConstraints_en") + output_fields.remove("features_properties_otherConstraints_fr") + output_fields.remove("features_properties_graphicOverview") + output_fields.remove("features_properties_distributionFormat_name") + output_fields.remove("features_properties_distributionFormat_format") + output_fields.remove("features_properties_useLimits_en") + output_fields.remove("features_properties_useLimits_fr") + output_fields.remove("features_properties_accessConstraints") + output_fields.remove("features_properties_dateStamp") + output_fields.remove("features_properties_dataSetURI") + output_fields.remove("features_properties_locale_language") + output_fields.remove("features_properties_locale_country") + output_fields.remove("features_properties_locale_encoding") + output_fields.remove("features_properties_language") + output_fields.remove("features_properties_characterSet") + output_fields.remove("features_properties_environmentDescription") + output_fields.remove("features_properties_supplementalInformation_en") + output_fields.remove("features_properties_supplementalInformation_fr") + output_fields.remove("features_properties_contact") + output_fields.remove("features_properties_credits") + output_fields.remove("features_properties_cited") + output_fields.remove("features_properties_distributor") + output_fields.remove("features_properties_options") + output_fields.remove("features_properties_temporalExtent_end_@indeterminatePosition") + output_fields.remove("features_properties_temporalExtent_end_#text") + output_fields.remove("features_properties_plugins") + output_fields.remove("features_properties_sourceSystemName") + output_fields.remove("features_popularity") + output_fields.remove("features_similarity") + + with open(csv_path, "w") as outstream: + # Setup the output + writer = csv.DictWriter( + outstream, + fieldnames=output_fields, + extrasaction="ignore", # Ignore extra dictionary keys/values + ) + + # Write to the output + writer.writeheader() + writer.writerows(rows) + + df = pd.read_csv(data) + + new_df = df.dropna() + + print(new_df.to_string()) + + new_df.to_csv(data_out) + + # Create a file GDB for ease of use. + # + arcpy.CreateFileGDB_management(OutFolderPath, OutName, out_version="CURRENT") + print "Created a file GDB for ease of use" + + # Convert to GDB table for ease of use + # + arcpy.TableToTable_conversion(InputCSV,GeoDataBase, GDBTableName, where_clause="", field_mapping='Field1 "Field1" true true false 4 Long 0 0 ,First,#,C:\TEMP\records_remove_empty.csv,Field1,-1,-1;features_properties_id "features_properties_id" true true false 8000 Text 0 0 ,First,#,C:\TEMP\records_remove_empty.csv,features_properties_id,-1,-1', config_keyword="") + print "Converted to GDB table for ease of use" + + # Create a clean table with all of the elements that you don't need queried out + # + arcpy.TableSelect_analysis(GDBTablePathAndName, GDBTableOutputPathAndName, where_clause="features_properties_id NOT LIKE 'ccmeo%' AND features_properties_id NOT LIKE 'CGDI%' AND features_properties_id NOT LIKE 'eodms%' AND features_properties_id IS NOT NULL") + print " Cleaned table of all rows that are not to be used in this process" + + # add field to table, old field is too long at 8000 characters + # + arcpy.AddField_management(GDBTableInputPathAndName, field_name="Feature_ID", field_type="TEXT", field_precision="", field_scale="", field_length="256", field_alias="", field_is_nullable="NULLABLE", field_is_required="NON_REQUIRED", field_domain="") + print "Add a field to be a better length as opposed to 8000 characters, 256 characters now." + + # Calculate field to hold all values from original ID field from csv table + # + arcpy.CalculateField_management(GDBTableInputPathAndName, field="Feature_ID", expression="!features_properties_id!", expression_type="PYTHON_9.3", code_block="") + print "Move field values over to Feature_ID from original field" + + # Delete extraineous fields + # + arcpy.DeleteField_management(GDBTableInputPathAndName, drop_field="Field1;features_properties_id;Unnamed__0") + print "Delete extraneous fields" + + # Convert cleaned table to Excel file + # + arcpy.TableToExcel_conversion(GDBTableInputPathAndName, ExcelOutputTable, Use_field_alias_as_column_header="NAME", Use_domain_and_subtype_description="CODE") + print "Excel table output created" + + # GDB table to DBF for interoperability + # + arcpy.TableToTable_conversion(GDBTableInputPathAndName, OutFolderPath, DBFName, where_clause="", field_mapping='Feature_ID "Feature_ID" true true false 256 Text 0 0 ,First,#,C:\TEMP\Records.gdb\records_column_and_rows_clean,Feature_ID,-1,-1', config_keyword="") + print "DBF table exported for interoperability" + + # Variables: + # + # to do: change hardcoded to variables + FOLDER = "C:/TEMP" + FILENAME = "catalogue_scrape.dbf" + CATALOGUE_SCRAPE = FOLDER + "//" + FILENAME + #filename_geocore = "34c0dbf2-9595-84f3-679c-7d2d7c90ecfe.geojson" #worked + #filename_geocore = "000183ed-8864-42f0-ae43-c4313a860720.geojson" #worked + #filename_geocore = "cebc283f-bae1-4eae-a91f-a26480cd4e4a.geojson" #worked + #filename_geocore = "2606b1b4-c895-4d23-b466-ad3d64b6381e.geojson" #worked + #filename_geocore = "08166334-889d-4c3a-9b25-b1b85ba48f2c.geojson" #worked + #filename_geocore = "05002515-f6cc-4516-b225-38d510eaaf9c.geojson" #worked + #filename_geocore = "090494a6-8aaf-4c26-b7b3-cf52400bf619.geojson" #worked + #filename_geocore = "054cf636-6637-2508-9aef-0f9139734f4a.geojson" #worked, 42 records long + #filename_geocore = "02d6f853-b0fe-4aa4-bc73-dff0db45d8ae.geojson" + #filename_geocore = "d2af02fe-9e12-413d-8959-06be963bde52.geojson" + FILENAME_GEOCORE = "9e1507cd-f25c-4c64-995b-6563bf9d65bd.geojson" + FILEID = str("9e1507cd-f25c-4c64-995b-6563bf9d65bd") + GEOJSON_URL = 'https://geocore.metadata.geo.ca/' + str(FILENAME_GEOCORE) + SAVE_PATH = "C:/TEMP/{}".format(FILENAME_GEOCORE) + FILE_ID = str(FILENAME_GEOCORE) + #cgp_encoding = "" + + # now download the geojson file and save it to the TEMP directory + # + download_url(GEOJSON_URL, SAVE_PATH) + + with open(SAVE_PATH) as f: + GEOJSON = geojson.load(f) + ITEM_DICTIONARY = GEOJSON + OPTIONS_COUNT = len(ITEM_DICTIONARY['features'][0]['properties']['options']) + print str(OPTIONS_COUNT) + " = options count" + URL_COUNT = OPTIONS_COUNT + print "URL count for all options = " + str(URL_COUNT) + PROTOCOL_COUNT = OPTIONS_COUNT + print "PROTOCOL count for all options = " + str(PROTOCOL_COUNT) + NAME_COUNT = OPTIONS_COUNT + print "NAME count for all options = " + str(NAME_COUNT) + DESCRIPTION_COUNT = len(ITEM_DICTIONARY['features'][0]['properties']['options'][0]['description']) + DESCRIPTION_COUNT = DESCRIPTION_COUNT * OPTIONS_COUNT + print "DESCRIPTION count for all options = " + str(DESCRIPTION_COUNT) + + + print "Downloaded geoJSON GeoCore file" + print "URL = " + str(GEOJSON_URL) + print "save_path = " + str(SAVE_PATH) + + #We now have every attribute counted from the original GeoJSON. + #We are going to create an attribute table to hold all of this data. + #We are going to use the name_count attribute as a number to define how many + #rows we are going to add to the attribute table. + #For this example the number of rows to add is + + print "Number of lines to be added " + unicode(NAME_COUNT) + #the name_count attribute""" + + # Set DBF encoding from ASCII to UTF-8, also created new DBF table + # + dbf.Table(r'C:\TEMP\catalogue_scrape.dbf', 'FILENAME C(100); FILEID C(254); URL C(254); PROTOCOL C(100); NAME C(254); NAME_EN C(254); NAME_FR C(254); DESC_ C(254); DESC_EN C(254); DESC_FR C(254); ROWID_ C(10)', codepage=0xf0) + print "Table created, catalogue_scrape.dbf" + + # Create insert cursor for table + # + ROWS = arcpy.InsertCursor(r"C:\TEMP\catalogue_scrape.dbf") + + # Create new rows based off of how many names + # exist in the options attributes. + # + for x in range(0, NAME_COUNT): + row = ROWS.newRow() + ROWS.insertRow(row) + + # Delete cursor and row objects to remove locks on the data + # + del row + del ROWS + + ROWS = arcpy.UpdateCursor(r"C:\TEMP\catalogue_scrape.dbf") + + + for row in ROWS: + row.setValue("FILENAME", SAVE_PATH) + row.setValue("FILEID", str(FILEID)) + ROWS.updateRow(row) + + del row + del ROWS + + FEATURE = 0 + while FEATURE < OPTIONS_COUNT: + + ROWS = arcpy.UpdateCursor(r"C:\TEMP\catalogue_scrape.dbf") + ROW_NUMBER = 0 + print str(ROW_NUMBER) + " = row number in catalo" + print str(OPTIONS_COUNT) + " = options count" + print str(FEATURE) + " = feature counter" + print "inside loop1" + + for row in ROWS: + print "inside loop2" + #url = gj['features'][0]['properties']['options'][int(feature)]['url'] + #print str(feature) + " = feature" + print str(ROW_NUMBER) + " = row number" + if FEATURE == ROW_NUMBER and FEATURE < OPTIONS_COUNT: + + print "inside if statement" + print FEATURE + url = GEOJSON['features'][0]['properties']['options'][int(FEATURE)]['url'] + protocol = GEOJSON['features'][0]['properties']['options'][int(FEATURE)]['protocol'] + name_en = GEOJSON['features'][0]['properties']['options'][int(FEATURE)]['name']['en'] + name_fr = GEOJSON['features'][0]['properties']['options'][int(FEATURE)]['name']['fr'] + name = unicode(GEOJSON['features'][0]['properties']['options'][int(FEATURE)]['name']['en']) + unicode(GEOJSON['features'][0]['properties']['options'][int(FEATURE)]['name']['fr']) + print name + " = name" + description_en = GEOJSON['features'][0]['properties']['options'][int(FEATURE)]['description']['en'] + description_fr = GEOJSON['features'][0]['properties']['options'][int(FEATURE)]['description']['fr'] + description = unicode(description_en) + unicode(description_fr) + print description + " = description" + row.setValue("NAME", name) + row.setValue("NAME_EN", name_en) + row.setValue("NAME_FR", name_fr) + row.setValue("DESC_", description) + row.setValue("DESC_EN", description_en) + row.setValue("DESC_FR", description_fr) + row.setValue("PROTOCOL", protocol) + row.setValue("URL", url) + row.setValue("ROWID_", FEATURE) + ROWS.updateRow(row) + print "url = " + str(url) + FEATURE = FEATURE + 1 + ROW_NUMBER = ROW_NUMBER + 1 + + print str(ROW_NUMBER) + " = row number" + print str(FEATURE) + " = feature number" + + #print "Option count = " + str(options_count) + #print "Feature = " + str(feature) + #options = GEOJSON['features'][0]['properties']['options'][int(feature)] + + print "url count = " + str(URL_COUNT) + #print options + + del row + del ROWS + + # Clean up all files that are not needed and should be deleted after this + # process is done + arcpy.Delete_management(in_data="C:/TEMP/Records.gdb", data_type="Workspace") + arcpy.Delete_management(in_data="C:/TEMP/Records.gdb", data_type="Folder") + arcpy.Delete_management(in_data="C:/TEMP/records.csv", data_type="TextFile") + arcpy.Delete_management(in_data="C:/TEMP/records.xls", data_type="File") + arcpy.Delete_management(in_data="C:/TEMP/records_remove_empty.csv", data_type="TextFile") +