Datasets:

sammlapp
/

Alberta_SBT_2016_REVI_Localized

# localize loca2024grid1 detections with stricter filters
from opensoundscape.localization import SynchronizedRecorderArray

import numpy as np
import pandas as pd
from glob import glob
from pathlib import Path
from time import time as timer
from tqdm.autonotebook import tqdm
import json

# localization parameters
min_n_receivers = 5
max_receiver_dist = 80
cc_threshold = 0.01  # 0
skip_if_json_exists = True

try:
    from redacted_paths import REDACTED_OUTPUT_DIR, REDACTED_DATA_ROOT
except ImportError:
    print("redacted paths not available")
    REDACTED_OUTPUT_DIR = ""
    REDACTED_DATA_ROOT = ""

# loop through each deployment
data_root = f"{REDACTED_DATA_ROOT}"
deployment_folders = glob(f"{data_root}/SBT*")
detections_path = f"{REDACTED_OUTPUT_DIR}/1_detect_hawkears"
localized_events_path = f"{REDACTED_OUTPUT_DIR}/2_localize_detections"

import datetime
import pytz


def get_dt(f, offset):
    dt = datetime.datetime.strptime(
        "_".join(Path(f).stem.split("_")[-2:]), "%Y%m%d_%H%M%S"
    ) + datetime.timedelta(seconds=offset)
    tz = pytz.timezone("US/Mountain")
    return tz.localize(dt)


# for deployment_folder in deployment_folders:
for deployment_folder in deployment_folders:

    # deployment_folder = deployment_folders[0]  # do one for now
    deployment_name = Path(deployment_folder).name
    print(f"localizing events for {deployment_name}")

    # configure paths
    # score_files = glob(f"{detections_path}/{deployment_name}/*/*.csv")

    audio_files = glob(f"{deployment_folder}/*/*.wav")
    print(f"found {len(audio_files)} files for {deployment_name}")
    if len(audio_files) < 1:
        continue

    # save json of SpatialEvents here:
    out_dir = f"{localized_events_path}/{deployment_name}"
    Path(out_dir).mkdir(exist_ok=True, parents=True)

    # create table providing xyz position for each audio file
    # using the rtk_points table, where point_folder_name matches
    # the name of the folder containing an audio file
    recorder_locations = pd.read_csv(
        f"{deployment_folder}/{deployment_name}_rtk_points.csv"
    )
    point_locs = recorder_locations.set_index("point_folder_name")
    file_position_df = pd.DataFrame({"file": audio_files})
    # point name is name of subfolder with audio files in it:
    file_position_df["point"] = file_position_df["file"].apply(
        lambda x: Path(x).parent.stem
    )
    file_position_df["file"] = file_position_df["file"].apply(lambda x: Path(x).name)
    file_position_df["file"] = [
        f"{deployment_folder}/{p.split('_')[0]}/{p}" for p in file_position_df["file"]
    ]
    # if necessary, remove files from recorders missing/incorrect metadata here
    # add coordinates to the table listing every audio file
    file_position_df[["x", "y"]] = [
        point_locs.loc[c][["EASTING", "NORTHING"]] for c in file_position_df["point"]
    ]  # removed z coordinate, for 2d localization
    # format to match expectation of SynchronizedRecorderArray file_coords argument:
    # index=file path, columns = x,y,(optional z) position

    file_coords = file_position_df.set_index("file")[["x", "y"]]  # , "z"]]

    # load detections saved as sparse df
    pickled_detections = (
        f"{detections_path}/{deployment_name}/dets_{deployment_name}_thresh-1.pkl"
    )
    detections = pd.read_pickle(pickled_detections)
    sp_counts = detections.astype(int).sum()
    sp_counts.sort_values(ascending=False).to_csv(f"{deployment_name}_det_cnts.csv")

    # add timestamp into detections df multi-index, since
    # stamp not available in the audio file metadata
    detections = detections.reset_index(drop=False)
    detections["start_timestamp"] = [
        get_dt(f, st) for f, st in zip(detections["file"], detections["start_time"])
    ]
    detections["file"] = [
        f"{deployment_folder}/{p.split('_')[0]}/{p}" for p in detections["file"]
    ]
    detections = detections.set_index(
        ["file", "start_time", "end_time", "start_timestamp"]
    )

    # skip species if < min_n_receivers detections
    sp_list = ["Red-eyed Vireo"]  # sp_counts[sp_counts > min_n_receivers].index.values
    detections = detections[sp_list]
    import warnings

    species_freq_ranges = pd.read_csv("./species_freq_ranges.csv").set_index(
        "english_common"
    )
    bandpass_ranges = {
        sp: [species_freq_ranges.at[sp, "low_f"], species_freq_ranges.at[sp, "high_f"]]
        for sp in detections.columns
    }

    # warn if species missing bandpass ranges:
    sp_missing_f_range = [
        s  # species_freq_ranges.at[s, "english_common"]
        for s in sp_list
        if np.isnan(bandpass_ranges[s][0]) or bandpass_ranges[s][0] is None
    ]
    if len(sp_missing_f_range) > 0:
        warnings.warn(
            f"{len(sp_missing_f_range)} classes have detections but bandpass range is unspecified: \n{sp_missing_f_range}"
        )

    # create relative positions, to avoid possible numeric convergence issues w/UTM vs elevation
    for c in file_coords.columns:
        file_coords[c] = file_coords[c] - file_coords[c].min()

    # TODO: consider converting back to real UTM/elevation at end

    t0 = timer()
    array = SynchronizedRecorderArray(file_coords)
    for sp in tqdm(sp_list):
        out_file = (
            f'{out_dir}/{deployment_name}_{sp.replace(" ", "_")}_localized_events.json'
        )
        if Path(out_file).exists() and skip_if_json_exists:
            continue
        if bandpass_ranges[sp][0] is None or np.isnan(bandpass_ranges[sp][0]):
            continue  # we'll just skip these classes for now?
        print(f"now localizing {sp} detections")
        localized_events, unlocalized_events = array.localize_detections(
            detections=detections[[sp]],
            max_receiver_dist=max_receiver_dist,
            min_n_receivers=min_n_receivers,
            bandpass_ranges=bandpass_ranges,
            return_unlocalized=True,
            cc_threshold=cc_threshold,
            localization_algorithm="soundfinder",
            num_workers=16,
        )

        if len(localized_events) > 0:
            events_dict = {"localized_events": [e.to_dict() for e in localized_events]}
            with open(out_file, "w+") as f:
                json.dump(events_dict, f)

            print(
                f"saved {len(localized_events)} localized {sp} detections to {out_dir}"
            )
        else:
            print(f"Zero events localized for {sp}")
        print(f"time elapsed: {(timer()-t0)/60} min")

    # print("finished all species")
    # print("breaking after one deployment")
    # break

# example of reloading the events:
# from loca_utils import from_dict
# with open('test_saving.json','r') as f:
#     reloaded_events = json.load(f)
# events = [from_dict(d) for d in reloaded_events['localized_events']]

# post-processing filters to apply:
# residual_rms < 1
#
# underground detections:
# distances = np.linalg.norm(
#     e.receiver_locations - e.location_estimate, axis=1
# )
# nearest_recorder_z_position = e.receiver_locations[np.argmin(distances)][2]
# if e.location_estimate[2] + 3 >= nearest_recorder_z_position:# and e.location_estimate[2]-10 <= nearest_recorder_z_position:
#     new_events.append(e)
#
# manual review using same workflow as 2023

	# localize loca2024grid1 detections with stricter filters
	from opensoundscape.localization import SynchronizedRecorderArray

	import numpy as np
	import pandas as pd
	from glob import glob
	from pathlib import Path
	from time import time as timer
	from tqdm.autonotebook import tqdm
	import json

	# localization parameters
	min_n_receivers = 5
	max_receiver_dist = 80
	cc_threshold = 0.01 # 0
	skip_if_json_exists = True

	try:
	from redacted_paths import REDACTED_OUTPUT_DIR, REDACTED_DATA_ROOT
	except ImportError:
	print("redacted paths not available")
	REDACTED_OUTPUT_DIR = ""
	REDACTED_DATA_ROOT = ""

	# loop through each deployment
	data_root = f"{REDACTED_DATA_ROOT}"
	deployment_folders = glob(f"{data_root}/SBT*")
	detections_path = f"{REDACTED_OUTPUT_DIR}/1_detect_hawkears"
	localized_events_path = f"{REDACTED_OUTPUT_DIR}/2_localize_detections"

	import datetime
	import pytz


	def get_dt(f, offset):
	dt = datetime.datetime.strptime(
	"_".join(Path(f).stem.split("_")[-2:]), "%Y%m%d_%H%M%S"
	) + datetime.timedelta(seconds=offset)
	tz = pytz.timezone("US/Mountain")
	return tz.localize(dt)


	# for deployment_folder in deployment_folders:
	for deployment_folder in deployment_folders:

	# deployment_folder = deployment_folders[0] # do one for now
	deployment_name = Path(deployment_folder).name
	print(f"localizing events for {deployment_name}")

	# configure paths
	# score_files = glob(f"{detections_path}/{deployment_name}//.csv")

	audio_files = glob(f"{deployment_folder}//.wav")
	print(f"found {len(audio_files)} files for {deployment_name}")
	if len(audio_files) < 1:
	continue

	# save json of SpatialEvents here:
	out_dir = f"{localized_events_path}/{deployment_name}"
	Path(out_dir).mkdir(exist_ok=True, parents=True)

	# create table providing xyz position for each audio file
	# using the rtk_points table, where point_folder_name matches
	# the name of the folder containing an audio file
	recorder_locations = pd.read_csv(
	f"{deployment_folder}/{deployment_name}_rtk_points.csv"
	)
	point_locs = recorder_locations.set_index("point_folder_name")
	file_position_df = pd.DataFrame({"file": audio_files})
	# point name is name of subfolder with audio files in it:
	file_position_df["point"] = file_position_df["file"].apply(
	lambda x: Path(x).parent.stem
	)
	file_position_df["file"] = file_position_df["file"].apply(lambda x: Path(x).name)
	file_position_df["file"] = [
	f"{deployment_folder}/{p.split('_')[0]}/{p}" for p in file_position_df["file"]
	]
	# if necessary, remove files from recorders missing/incorrect metadata here
	# add coordinates to the table listing every audio file
	file_position_df[["x", "y"]] = [
	point_locs.loc[c][["EASTING", "NORTHING"]] for c in file_position_df["point"]
	] # removed z coordinate, for 2d localization
	# format to match expectation of SynchronizedRecorderArray file_coords argument:
	# index=file path, columns = x,y,(optional z) position

	file_coords = file_position_df.set_index("file")[["x", "y"]] # , "z"]]

	# load detections saved as sparse df
	pickled_detections = (
	f"{detections_path}/{deployment_name}/dets_{deployment_name}_thresh-1.pkl"
	)
	detections = pd.read_pickle(pickled_detections)
	sp_counts = detections.astype(int).sum()
	sp_counts.sort_values(ascending=False).to_csv(f"{deployment_name}_det_cnts.csv")

	# add timestamp into detections df multi-index, since
	# stamp not available in the audio file metadata
	detections = detections.reset_index(drop=False)
	detections["start_timestamp"] = [
	get_dt(f, st) for f, st in zip(detections["file"], detections["start_time"])
	]
	detections["file"] = [
	f"{deployment_folder}/{p.split('_')[0]}/{p}" for p in detections["file"]
	]
	detections = detections.set_index(
	["file", "start_time", "end_time", "start_timestamp"]
	)

	# skip species if < min_n_receivers detections
	sp_list = ["Red-eyed Vireo"] # sp_counts[sp_counts > min_n_receivers].index.values
	detections = detections[sp_list]
	import warnings

	species_freq_ranges = pd.read_csv("./species_freq_ranges.csv").set_index(
	"english_common"
	)
	bandpass_ranges = {
	sp: [species_freq_ranges.at[sp, "low_f"], species_freq_ranges.at[sp, "high_f"]]
	for sp in detections.columns
	}

	# warn if species missing bandpass ranges:
	sp_missing_f_range = [
	s # species_freq_ranges.at[s, "english_common"]
	for s in sp_list
	if np.isnan(bandpass_ranges[s][0]) or bandpass_ranges[s][0] is None
	]
	if len(sp_missing_f_range) > 0:
	warnings.warn(
	f"{len(sp_missing_f_range)} classes have detections but bandpass range is unspecified: \n{sp_missing_f_range}"
	)

	# create relative positions, to avoid possible numeric convergence issues w/UTM vs elevation
	for c in file_coords.columns:
	file_coords[c] = file_coords[c] - file_coords[c].min()

	# TODO: consider converting back to real UTM/elevation at end

	t0 = timer()
	array = SynchronizedRecorderArray(file_coords)
	for sp in tqdm(sp_list):
	out_file = (
	f'{out_dir}/{deployment_name}_{sp.replace(" ", "_")}_localized_events.json'
	)
	if Path(out_file).exists() and skip_if_json_exists:
	continue
	if bandpass_ranges[sp][0] is None or np.isnan(bandpass_ranges[sp][0]):
	continue # we'll just skip these classes for now?
	print(f"now localizing {sp} detections")
	localized_events, unlocalized_events = array.localize_detections(
	detections=detections[[sp]],
	max_receiver_dist=max_receiver_dist,
	min_n_receivers=min_n_receivers,
	bandpass_ranges=bandpass_ranges,
	return_unlocalized=True,
	cc_threshold=cc_threshold,
	localization_algorithm="soundfinder",
	num_workers=16,
	)

	if len(localized_events) > 0:
	events_dict = {"localized_events": [e.to_dict() for e in localized_events]}
	with open(out_file, "w+") as f:
	json.dump(events_dict, f)

	print(
	f"saved {len(localized_events)} localized {sp} detections to {out_dir}"
	)
	else:
	print(f"Zero events localized for {sp}")
	print(f"time elapsed: {(timer()-t0)/60} min")

	# print("finished all species")
	# print("breaking after one deployment")
	# break

	# example of reloading the events:
	# from loca_utils import from_dict
	# with open('test_saving.json','r') as f:
	# reloaded_events = json.load(f)
	# events = [from_dict(d) for d in reloaded_events['localized_events']]

	# post-processing filters to apply:
	# residual_rms < 1
	#
	# underground detections:
	# distances = np.linalg.norm(
	# e.receiver_locations - e.location_estimate, axis=1
	# )
	# nearest_recorder_z_position = e.receiver_locations[np.argmin(distances)][2]
	# if e.location_estimate[2] + 3 >= nearest_recorder_z_position:# and e.location_estimate[2]-10 <= nearest_recorder_z_position:
	# new_events.append(e)
	#
	# manual review using same workflow as 2023