def test_single(filename, local=False, **munge_kwargs):
"""
Perform a test on a single file. Prints more diagnostic information than usual.
"""
(new_file, what) = munge(filename, **munge_kwargs)
query_obj = song.util.codegen(new_file, start=-1, duration=-1)
s = fp.best_match_for_query(query_obj[0]["code"], local=local)
if s.TRID is not None:
if local:
metad = _local_bigeval[s.TRID]
else:
metad = fp.metadata_for_track_id(s.TRID)
metad["title"] = metad["track"]
song_metadata = {
"artist": metad.get("artist", ""),
"release": metad.get("release", ""),
"title": metad.get("title", "")
}
print str(song_metadata)
else:
print "No match"
decoded = fp.decode_code_string(query_obj[0]["code"])
print str(len(decoded.split(" ")) / 2) + " codes in original"
response = fp.query_fp(decoded, local=local, rows=15)
if response is not None:
print "From FP flat:"
tracks = {}
scores = {}
for r in response.results:
trid = r["track_id"].split("-")[0]
if local:
metad = _local_bigeval[trid]
else:
metad = fp.metadata_for_track_id(trid)
metad["title"] = metad["track"]
m = {
"artist": metad.get("artist", ""),
"release": metad.get("release", ""),
"title": metad.get("title", "")
}
if m is not None:
actual_match = fp.actual_matches(
decoded, fp.fp_code_for_track_id(r["track_id"],
local=local))
tracks[r["track_id"]] = (m, r["score"], actual_match)
scores[r["track_id"]] = actual_match
else:
print "problem getting metadata for " + r["track_id"]
sorted_scores = sorted(scores.iteritems(),
key=lambda (k, v): (v, k),
reverse=True)
for (trackid, score) in sorted_scores:
(m, score, actual_match) = tracks[trackid]
print trackid + " (" + str(int(score)) + ", " + str(
actual_match) + ") - " + m["artist"] + " - " + m["title"]
else:
print "response from fp flat was None -- decoded code was " + str(
decoded)
os.remove(new_file)
def main():
if not len(sys.argv)==4:
print "usage: python little_eval.py [database_list | disk] query_list [limit]"
sys.exit()
fp_codes = []
limit = int(sys.argv[3])
if sys.argv[1] == "disk":
fp.local_load("disk.pkl")
else:
database_list = open(sys.argv[1]).read().split("\n")[0:limit]
for line in database_list:
(track_id, file) = line.split(" ### ")
print track_id
# TODO - use threaded codegen
j = codegen(file, start=-1, duration=-1)
if len(j):
code_str = fp.decode_code_string(j[0]["code"])
meta = j[0]["metadata"]
l = meta["duration"] * 1000
a = meta["artist"]
r = meta["release"]
t = meta["title"]
v = meta["version"]
fp_codes.append({"track_id": track_id, "fp": code_str, "length": str(l), "codever": str(round(v, 2)), "artist": a, "release": r, "track": t})
fp.ingest(fp_codes, local=True)
fp.local_save("disk.pkl")
counter = 0
actual_win = 0
original_win = 0
bm_win = 0
query_list = open(sys.argv[2]).read().split("\n")[0:limit]
for line in query_list:
(track_id, file) = line.split(" ### ")
print track_id
j = codegen(munge(file))
if len(j):
counter+=1
response = fp.query_fp(fp.decode_code_string(j[0]["code"]), rows=30, local=True, get_data=True)
(winner_actual, winner_original) = get_winners(fp.decode_code_string(j[0]["code"]), response, elbow=8)
winner_actual = winner_actual.split("-")[0]
winner_original = winner_original.split("-")[0]
response = fp.best_match_for_query(j[0]["code"], local=True)
if(response.TRID == track_id):
bm_win+=1
if(winner_actual == track_id):
actual_win+=1
if(winner_original == track_id):
original_win+=1
print "%d / %d actual (%2.2f%%) %d / %d original (%2.2f%%) %d / %d bm (%2.2f%%)" % (actual_win, counter, (float(actual_win)/float(counter))*100.0, \
original_win, counter, (float(original_win)/float(counter))*100.0, \
bm_win, counter, (float(bm_win)/float(counter))*100.0)
def test_single(filename, local=False, **munge_kwargs):
"""
Perform a test on a single file. Prints more diagnostic information than usual.
"""
(new_file, what) = munge(filename, **munge_kwargs)
query_obj = song.util.codegen(new_file, start=-1, duration=-1)
if not local:
s = fp.best_match_for_query(query_obj[0]["code"])
if s.TRID is not None:
metad = _local_bigeval[s.TRID]
song_metadata = {"artist": metad["artist"], "release": metad["release"], "title": metad["title"]}
print str(song_metadata)
else:
print "No match"
else:
s = fp.best_match_for_query(query_obj[0]["code"],local=local)
if s.TRID is not None:
metad = _local_bigeval[s.TRID]
song_metadata = {"artist": metad["artist"], "release": metad["release"], "title": metad["title"]}
print str(song_metadata)
else:
print "No match"
decoded = fp.decode_code_string(query_obj[0]["code"])
print str(len(decoded.split(" "))/2) + " codes in original"
response = fp.query_fp(decoded, local=local, rows=15)
if response is not None:
print "From FP flat:"
tracks = {}
scores = {}
for r in response.results:
trid = r["track_id"].split("-")[0]
metad = _local_bigeval[trid]
m = {"artist": metad["artist"], "release": metad["release"], "title": metad["title"]}
if m is not None:
actual_match = fp.actual_matches(decoded, fp.fp_code_for_track_id(r["track_id"], local=local))
tracks[r["track_id"]] = (m, r["score"], actual_match)
scores[r["track_id"]] = actual_match
else:
print "problem getting metadata for " + r["track_id"]
sorted_scores = sorted(scores.iteritems(), key=lambda (k,v): (v,k), reverse=True)
for (trackid, score) in sorted_scores:
(m, score, actual_match) = tracks[trackid]
print trackid + " (" + str(int(score)) + ", " + str(actual_match) +") - " + m["artist"] + " - " + m["title"]
else:
print "response from fp flat was None -- decoded code was " + str(decoded)
os.remove(new_file)
def magic_matches_list(code_string, elbow=10, local=False):
# DEC strings come in as unicode so we have to force them to ASCII
code_string = code_string.encode("utf8")
# First see if this is a compressed code
if re.match('[A-Za-z\/\+\_\-]', code_string) is not None:
code_string = decode_code_string(code_string)
if code_string is None:
return False
# Query the FP flat directly.
response = query_fp(code_string, rows=30, local=local, get_data=True)
if len(response.results) == 0:
return False
# Get the actual score for all responses
original_scores = {}
actual_scores = {}
trackids = [r["track_id"].encode("utf8") for r in response.results]
tcodes = get_tyrant().multi_get(trackids)
# For each result compute the "actual score" (based on the histogram matching)
for (i, r) in enumerate(response.results):
track_id = r["track_id"]
original_scores[track_id] = int(r["score"])
track_code = tcodes[i]
if track_code is None:
# Solr gave us back a track id but that track
# is not in our keystore
continue
actual_scores[track_id] = actual_matches(code_string, track_code, elbow = elbow)
# Sort the actual scores
sorted_actual_scores = sorted(actual_scores.iteritems(), key=lambda (k,v): (v,k), reverse=True)
# Because we split songs up into multiple parts, sometimes the results will have the same track in the
# first few results. Remove these duplicates so that the falloff is (potentially) higher.
final_results = []
for trid, result in sorted_actual_scores:
trid = trid.split("-")[0]
meta = metadata_for_track_id(trid, local=local)
final_results.append(meta)
return final_results
def main():
if not len(sys.argv) == 4:
print "usage: python little_eval.py [database_list | disk] query_list [limit]"
sys.exit()
fp_codes = []
limit = int(sys.argv[3])
if sys.argv[1] == "disk":
fp.local_load("disk.pkl")
else:
database_list = open(sys.argv[1]).read().split("\n")[0:limit]
for line in database_list:
(track_id, file) = line.split(" ### ")
print track_id
# TODO - use threaded codegen
j = codegen(file, start=-1, duration=-1)
if len(j):
code_str = fp.decode_code_string(j[0]["code"])
meta = j[0]["metadata"]
l = meta["duration"] * 1000
a = meta["artist"]
r = meta["release"]
t = meta["title"]
v = meta["version"]
fp_codes.append({
"track_id": track_id,
"fp": code_str,
"length": str(l),
"codever": str(round(v, 2)),
"artist": a,
"release": r,
"track": t
})
fp.ingest(fp_codes, local=True)
fp.local_save("disk.pkl")
counter = 0
actual_win = 0
original_win = 0
bm_win = 0
query_list = open(sys.argv[2]).read().split("\n")[0:limit]
for line in query_list:
(track_id, file) = line.split(" ### ")
print track_id
j = codegen(munge(file))
if len(j):
counter += 1
response = fp.query_fp(fp.decode_code_string(j[0]["code"]),
rows=30,
local=True,
get_data=True)
(winner_actual, winner_original) = get_winners(
fp.decode_code_string(j[0]["code"]), response, elbow=8)
winner_actual = winner_actual.split("-")[0]
winner_original = winner_original.split("-")[0]
response = fp.best_match_for_query(j[0]["code"], local=True)
if (response.TRID == track_id):
bm_win += 1
if (winner_actual == track_id):
actual_win += 1
if (winner_original == track_id):
original_win += 1
print "%d / %d actual (%2.2f%%) %d / %d original (%2.2f%%) %d / %d bm (%2.2f%%)" % (actual_win, counter, (float(actual_win)/float(counter))*100.0, \
original_win, counter, (float(original_win)/float(counter))*100.0, \
bm_win, counter, (float(bm_win)/float(counter))*100.0)