mirror of
https://github.com/Brandon-Rozek/website-observations.git
synced 2024-12-04 13:57:32 -05:00
258 lines
7.4 KiB
Python
Executable file
258 lines
7.4 KiB
Python
Executable file
#!/usr/bin/env python
|
|
"""
|
|
Script to create Hugo markdown
|
|
files from iNaturalist Observations
|
|
"""
|
|
|
|
from urllib import request
|
|
from glob import glob
|
|
from http.client import HTTPResponse
|
|
from pathlib import Path
|
|
from typing import Any, Dict, Optional
|
|
import json
|
|
import sys
|
|
import time
|
|
|
|
CONTENT_LOCATION = "./"
|
|
USER_ID = "brandonrozek"
|
|
MIN_OBS_ID = -1
|
|
|
|
|
|
def retrieve_data_from_server():
|
|
server_data = []
|
|
server_ids = retrieve_obs_ids_from_server()
|
|
for id_num in server_ids:
|
|
# Grab observation from iNaturalist
|
|
url = f"https://api.inaturalist.org/v1/observations/{id_num}"
|
|
response: Optional[HTTPResponse] = None
|
|
|
|
try:
|
|
response = request.urlopen(url)
|
|
except Exception:
|
|
print(f"Unable to grab observation {id_num} from iNaturalist.")
|
|
|
|
time.sleep(1) # Rate Limit: 1 request/sec
|
|
|
|
if response is None:
|
|
continue
|
|
|
|
# Parse server response
|
|
server_data_part = None
|
|
try:
|
|
server_data_part = json.loads(response.read())['results'][0]
|
|
# Note: there is only one observation as a result
|
|
except Exception:
|
|
print(f"Malformed JSON response for observation {id_num}.")
|
|
continue
|
|
|
|
server_data_part = reformat_obs(id_num, server_data_part)
|
|
server_data.append(server_data_part)
|
|
|
|
|
|
print(f"Successfully obtained {len(server_data)} observations from the server.")
|
|
return server_data
|
|
|
|
|
|
def retrieve_obs_ids_from_server():
|
|
"""
|
|
Grabs observation ids from iNaturalist server
|
|
"""
|
|
global MIN_OBS_ID
|
|
server_data = []
|
|
|
|
finished_retrieving = False
|
|
while not finished_retrieving:
|
|
# Grab observations from iNaturalist
|
|
id_below = "&id_below=" + str(MIN_OBS_ID) \
|
|
if MIN_OBS_ID > 0 else ""
|
|
url = "https://api.inaturalist.org/v1/observations?order=desc&order_by=created_at&only_id=true&user_id=" + USER_ID + id_below
|
|
response: Optional[HTTPResponse] = None
|
|
|
|
try:
|
|
response = request.urlopen(url)
|
|
except Exception:
|
|
print("Unable to grab observations from iNaturalist.")
|
|
|
|
if response is None:
|
|
sys.exit(-1)
|
|
|
|
time.sleep(1) # Rate Limit: 1 request/sec
|
|
|
|
# Parse server response
|
|
server_data_part: Optional[list] = None
|
|
try:
|
|
server_data_part = json.loads(response.read())['results']
|
|
except Exception:
|
|
print("Malformed JSON response from server.")
|
|
|
|
if server_data is None:
|
|
sys.exit(-1)
|
|
|
|
if not isinstance(server_data_part, list):
|
|
print("Unexpected JSON response, should be of form list.")
|
|
sys.exit(-1)
|
|
|
|
# No more to retrieve
|
|
if len(server_data_part) == 0:
|
|
finished_retrieving = True
|
|
break
|
|
|
|
server_data_part = [d['id'] for d in server_data_part]
|
|
|
|
# print(f"Retrieved {len(server_data_part)} observations from server")
|
|
server_data.extend(server_data_part)
|
|
MIN_OBS_ID = server_data_part[-1]
|
|
|
|
print(f"Parsed a total of {len(server_data)} ids from server")
|
|
return server_data
|
|
|
|
|
|
def reformat_obs(obsid, obs_json):
|
|
"""
|
|
Takes a obs_json and
|
|
slightly modifies it to match
|
|
some of the fields Hugo expects.
|
|
"""
|
|
obs_data = dict(
|
|
id=str(obsid),
|
|
metadata={},
|
|
content=""
|
|
)
|
|
|
|
# Turn URL -> Syndication
|
|
obs_data['metadata']['syndication'] = obs_json['uri']
|
|
|
|
# Turn Created At -> Date
|
|
obs_data['metadata']['date'] = obs_json['time_observed_at']
|
|
|
|
# Grab some taxonomy information about the organism
|
|
obs_data['metadata']['taxon'] = dict(
|
|
name=obs_json['taxon']['name'],
|
|
common_name=obs_json['taxon']['preferred_common_name']
|
|
)
|
|
|
|
# Grab only a few fields
|
|
desired_fields = [
|
|
'quality_grade', 'identifications_most_agree',
|
|
'species_guess', 'identifications_most_disagree',
|
|
'captive', 'project_ids',
|
|
'community_taxon_id', 'geojson',
|
|
'owners_identification_from_vision',
|
|
'identifications_count', 'obscured',
|
|
'num_identification_agreements',
|
|
'num_identification_disagreements',
|
|
'place_guess', "photos"
|
|
]
|
|
for key in desired_fields:
|
|
obs_data['metadata'][key] = obs_json[key]
|
|
|
|
return obs_data
|
|
|
|
############################################################################
|
|
|
|
def findall(p, s):
|
|
"""
|
|
Yields all the positions of
|
|
the pattern p in the string s.
|
|
Source: https://stackoverflow.com/a/34445090
|
|
"""
|
|
i = s.find(p)
|
|
while i != -1:
|
|
yield i
|
|
i = s.find(p, i+1)
|
|
|
|
def hugo_markdown_to_json(markdown_contents) -> Optional[Dict[Any, Any]]:
|
|
"""
|
|
Take the contents from a Hugo markdown
|
|
file and read the JSON frontmatter if it
|
|
exists.
|
|
"""
|
|
front_matter_indices = list(findall('---', markdown_contents))
|
|
if len(front_matter_indices) < 2:
|
|
return None
|
|
front_matter = markdown_contents[(front_matter_indices[0] + 3):front_matter_indices[1]]
|
|
json_contents = None
|
|
try:
|
|
json_contents = json.loads(front_matter)
|
|
except Exception:
|
|
pass
|
|
if not isinstance(json_contents, dict):
|
|
json_contents = None
|
|
html_contents = markdown_contents[front_matter_indices[1] + 19:-17]
|
|
return json_contents, html_contents
|
|
|
|
def create_markdown_str(frontmatter, content):
|
|
"""
|
|
Takes a JSON frontmatter
|
|
and creates a string representing
|
|
the contents of a Hugo markdown
|
|
file.
|
|
"""
|
|
return "---\n" + \
|
|
f"{json.dumps(frontmatter)}\n" +\
|
|
"---\n" +\
|
|
"{{< unsafe >}}\n" +\
|
|
f"{content}\n" +\
|
|
"{{< /unsafe >}}\n"
|
|
|
|
def file_from_id(idnum):
|
|
"""Returns filename from id"""
|
|
return f"{CONTENT_LOCATION}/{idnum}.md"
|
|
|
|
def read_hugo_markdown(idnum) -> Optional[Dict[Any, Any]]:
|
|
"""
|
|
Given an id, return the markdown file
|
|
frontmatter and contents stored in Hugo
|
|
if it exists.
|
|
"""
|
|
try:
|
|
with open(file_from_id(idnum), "r", encoding="UTF-8") as hugo_file:
|
|
frontmatter, contents = hugo_markdown_to_json(hugo_file.read())
|
|
return frontmatter, contents
|
|
except Exception:
|
|
return None
|
|
|
|
def write_markdown(id_num, frontmatter, contents):
|
|
"""
|
|
Takes a frontmatter json
|
|
and writes it to a hugo
|
|
markdown content file.
|
|
"""
|
|
try:
|
|
with open(file_from_id(id_num), "w", encoding="UTF-8") as hugo_file:
|
|
hugo_file.write(create_markdown_str(frontmatter, contents))
|
|
except Exception as e:
|
|
print("Failed to write", id_num)
|
|
|
|
|
|
############################################################################
|
|
|
|
# Read in saved data
|
|
saved_filenames = glob(CONTENT_LOCATION + "/*.md")
|
|
saved_ids = { Path(fname).stem for fname in saved_filenames }
|
|
|
|
server_data = retrieve_data_from_server()
|
|
|
|
# Data is structured like [{id: '', metadata: '', contents: ''}]
|
|
# Where metadata is a JSON and contents is HTML
|
|
|
|
for data in server_data:
|
|
id_num = data['id']
|
|
|
|
# If the observation already exists
|
|
if id_num in saved_ids:
|
|
saved_fm, saved_content = read_hugo_markdown(id_num)
|
|
if saved_fm is None:
|
|
print("Unable to read saved data id", id_num)
|
|
|
|
# Only update if observation has changed
|
|
elif saved_fm != data['metadata']:
|
|
print("Updating id", id_num)
|
|
write_markdown(id_num, data['metadata'], data['content'])
|
|
|
|
# New observation found
|
|
else:
|
|
print("Creating id", id_num)
|
|
write_markdown(id_num, data['metadata'], data['content'])
|
|
|
|
print("Completed")
|