los_angeles_midi_dataset_search_and_explore.py

# -*- coding: utf-8 -*-
"""Los_Angeles_MIDI_Dataset_Search_and_Explore.ipynb

Automatically generated by Colaboratory.

Original file is located at
    https://colab.research.google.com/github/asigalov61/Los-Angeles-MIDI-Dataset/blob/main/Los_Angeles_MIDI_Dataset_Search_and_Explore.ipynb

# Los Angeles MIDI Dataset: Search and Explore (ver. 2.1)

***

Powered by tegridy-tools: https://github.com/asigalov61/tegridy-tools

***

#### Project Los Angeles

#### Tegridy Code 2023

***

# (SETUP ENVIRONMENT)
"""

#@title Install all dependencies (run only once per session)
!git clone --depth 1 https://github.com/asigalov61/Los-Angeles-MIDI-Dataset
!pip install huggingface_hub
!pip install matplotlib
!pip install sklearn
!pip install tqdm
!apt install fluidsynth #Pip does not work for some reason. Only apt works
!pip install midi2audio

#@title Import all needed modules

print('Loading core modules...')
import os
import copy
from collections import Counter
import random
import pickle
from tqdm import tqdm

from joblib import Parallel, delayed
import multiprocessing

if not os.path.exists('/content/LAMD'):
    os.makedirs('/content/LAMD')

print('Loading MIDI.py module...')
os.chdir('/content/Los-Angeles-MIDI-Dataset')
import MIDI

print('Loading aux modules...')
from sklearn.metrics import pairwise_distances, pairwise
import matplotlib.pyplot as plt

from midi2audio import FluidSynth
from IPython.display import Audio, display

from huggingface_hub import hf_hub_download

from google.colab import files

os.chdir('/content/')
print('Done!')

"""# (PREP DATA)"""

# Commented out IPython magic to ensure Python compatibility.
#@title Unzip LAMDa data
# %cd /content/Los-Angeles-MIDI-Dataset/META-DATA

print('=' * 70)
print('Unzipping META-DATA...Please wait...')

!cat LAMDa_META_DATA.zip* > LAMDa_META_DATA.zip
print('=' * 70)

!unzip -j LAMDa_META_DATA.zip
print('=' * 70)

print('Done! Enjoy! :)')
print('=' * 70)
# %cd /content/

#================================================

# %cd /content/Los-Angeles-MIDI-Dataset/MIDI-MATRIXES

print('=' * 70)
print('Unzipping MIDI-MATRIXES...Please wait...')

!cat LAMDa_MIDI_MATRIXES.zip* > LAMDa_MIDI_MATRIXES.zip
print('=' * 70)

!unzip -j LAMDa_MIDI_MATRIXES.zip
print('=' * 70)

print('Done! Enjoy! :)')
print('=' * 70)
# %cd /content/

#==================================================

# %cd /content/Los-Angeles-MIDI-Dataset/TOTALS

print('=' * 70)
print('Unzipping TOTALS...Please wait...')

!unzip -j LAMDa_TOTALS.zip
print('=' * 70)

print('Done! Enjoy! :)')
print('=' * 70)
# %cd /content/

#@title Load LAMDa data
print('=' * 70)
print('Loading LAMDa data...Please wait...')
print('=' * 70)
print('Loading LAMDa META-DATA...')
meta_data = pickle.load(open('/content/Los-Angeles-MIDI-Dataset/META-DATA/LAMDa_META_DATA.pickle', 'rb'))
print('Done!')
print('=' * 70)
print('Loading LAMDa MIDI-MATRIXES...')
midi_matrixes = pickle.load(open('/content/Los-Angeles-MIDI-Dataset/MIDI-MATRIXES/LAMDa_MIDI_MATRIXES.pickle', 'rb'))
print('Done!')
print('=' * 70)
print('Loading LAMDa TOTALS...')
totals = pickle.load(open('/content/Los-Angeles-MIDI-Dataset/TOTALS/LAMDa_TOTALS.pickle', 'rb'))
print('Done!')
print('=' * 70)
print('Enjoy!')
print('=' * 70)

"""# (PREP MIDI DATASET)"""

#@title Download the dataset
print('=' * 70)
print('Downloading Los Angeles MIDI Dataset...Please wait...')
print('=' * 70)

hf_hub_download(repo_id='projectlosangeles/Los-Angeles-MIDI-Dataset', 
                filename='Los-Angeles-MIDI-Dataset-Ver-2-0-CC-BY-NC-SA.zip',
                repo_type="dataset",
                local_dir='/content/LAMD', 
                local_dir_use_symlinks=False)
print('=' * 70)
print('Done! Enjoy! :)')
print('=' * 70)

# Commented out IPython magic to ensure Python compatibility.
#@title Unzip the dataset
# %cd /content/LAMD

print('=' * 70)
print('Unzipping Los Angeles MIDI Dataset...Please wait...')
!unzip 'Los-Angeles-MIDI-Dataset-Ver-2-0-CC-BY-NC-SA.zip'
print('=' * 70)

print('Done! Enjoy! :)')
print('=' * 70)
# %cd /content/

#@title Create dataset files list
print('=' * 70)
print('Creating dataset files list...')
dataset_addr = "/content/LAMD/MIDIs"

# os.chdir(dataset_addr)
filez = list()
for (dirpath, dirnames, filenames) in os.walk(dataset_addr):
    filez += [os.path.join(dirpath, file) for file in filenames]

if filez == []:
    print('Could not find any MIDI files. Please check Dataset dir...')
    print('=' * 70)

print('=' * 70)
print('Randomizing file list...')
random.shuffle(filez)
print('=' * 70)

LAMD_files_list = []

for f in tqdm(filez):
  LAMD_files_list.append([f.split('/')[-1].split('.mid')[0], f])
print('Done!')
print('=' * 70)

"""# (PLOT TOTALS)"""

#@title Plot Totals

cos_sim = pairwise.cosine_similarity(
      totals[0][0][4] 
  )
plt.figure(figsize=(8, 8))
plt.imshow(cos_sim, cmap="inferno", interpolation="none")
im_ratio = 1
plt.colorbar(fraction=0.046 * im_ratio, pad=0.04)
plt.title('Times')
plt.xlabel("Position")
plt.ylabel("Position")
plt.tight_layout()
plt.plot()

cos_sim = pairwise.cosine_similarity(
      totals[0][0][5] 
  )
plt.figure(figsize=(8, 8))
plt.imshow(cos_sim, cmap="inferno", interpolation="none")
im_ratio = 1
plt.colorbar(fraction=0.046 * im_ratio, pad=0.04)
plt.title('Durations')
plt.xlabel("Position")
plt.ylabel("Position")
plt.tight_layout()
plt.plot()

cos_sim = pairwise.cosine_similarity(
      totals[0][0][6] 
  )
plt.figure(figsize=(8, 8))
plt.imshow(cos_sim, cmap="inferno", interpolation="none")
im_ratio = 1
plt.colorbar(fraction=0.046 * im_ratio, pad=0.04)
plt.title('Channels')
plt.xlabel("Position")
plt.ylabel("Position")
plt.tight_layout()
plt.plot()

cos_sim = pairwise.cosine_similarity(
      totals[0][0][7] 
  )
plt.figure(figsize=(8, 8))
plt.imshow(cos_sim, cmap="inferno", interpolation="none")
im_ratio = 1
plt.colorbar(fraction=0.046 * im_ratio, pad=0.04)
plt.title('Instruments')
plt.xlabel("Position")
plt.ylabel("Position")
plt.tight_layout()
plt.plot()

cos_sim = pairwise.cosine_similarity(
      totals[0][0][8] 
  )
plt.figure(figsize=(8, 8))
plt.imshow(cos_sim, cmap="inferno", interpolation="none")
im_ratio = 1
plt.colorbar(fraction=0.046 * im_ratio, pad=0.04)
plt.title('Pitches')
plt.xlabel("Position")
plt.ylabel("Position")
plt.tight_layout()
plt.plot()

cos_sim = pairwise.cosine_similarity(
      totals[0][0][9] 
  )
plt.figure(figsize=(8, 8))
plt.imshow(cos_sim, cmap="inferno", interpolation="none")
im_ratio = 1
plt.colorbar(fraction=0.046 * im_ratio, pad=0.04)
plt.title('Velocities')
plt.xlabel("Position")
plt.ylabel("Position")
plt.tight_layout()
plt.plot()

"""# (LOAD SOURCE MIDI)"""

#@title Load source MIDI

full_path_to_source_MIDI = "/content/Los-Angeles-MIDI-Dataset/Come-To-My-Window-Modified-Sample-MIDI.mid" #@param {type:"string"}
render_MIDI_to_audio = False #@param {type:"boolean"}

#=================================================================================

f = full_path_to_source_MIDI

print('=' * 70)
print('Loading MIDI file...')

score = MIDI.midi2ms_score(open(f, 'rb').read())

events_matrix = []

itrack = 1

while itrack < len(score):
    for event in score[itrack]:         
      events_matrix.append(event)
    itrack += 1

# Sorting...
events_matrix.sort(key=lambda x: x[1])

# recalculating timings
for e in events_matrix:
    e[1] = int(e[1] / 10)
    if e[0] == 'note':
      e[2] = int(e[2] / 20)

# final processing...

melody_chords = []

patches = [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]

pe = events_matrix[0]
for e in events_matrix:

  if e[0] == 'note':
    # ['note', start_time, duration, channel, note, velocity]
    time = max(0, min(255, e[1]-pe[1]))
    duration = max(1, min(255, e[2]))
    channel = max(0, min(15, e[3]))

    if e[3] != 9:
      instrument = max(0, min(127, patches[e[3]]))
    else:
      instrument = max(128, min(255, patches[e[3]]+128))

    if e[3] != 9:

      pitch = max(1, min(127, e[4]))
    else:
      pitch = max(129, min(255, e[4]+128))

    if e[3] != 9:
      velocity = max(1, min(127, e[5]))
    else:
      velocity = max(129, min(255, e[5]+128))

    melody_chords.append([time, duration, channel, instrument, pitch, velocity])

  if e[0] == 'patch_change':
    # ['patch_change', dtime, channel, patch]
    time = max(0, min(127, e[1]-pe[1]))
    channel = max(0, min(15, e[2]))
    patch = max(0, min(127, e[3]))

    patches[channel] = patch

  pe = e # Previous event

MATRIX = [[0]*256 for i in range(38)]

for m in melody_chords:

  MATRIX[0][m[0]] += 1
  MATRIX[1][m[1]] += 1
  MATRIX[2][m[2]] += 1 
  MATRIX[3][m[3]] += 1
  MATRIX[4][m[4]] += 1
  MATRIX[5][m[5]] += 1
  MATRIX[m[2]+6][m[3]] += 1
  MATRIX[m[2]+22][m[4]] += 1

#==================================================

score = MIDI.midi2score(open(f, 'rb').read())

events_matrix = []

track_count = 0

for s in score:
    
    if track_count > 0:
        track = s
        track.sort(key=lambda x: x[1])
        events_matrix.extend(track)
    else:
        midi_ticks = s

    track_count += 1
    
events_matrix.sort(key=lambda x: x[1])

mult_pitches_counts = []

for i in range(-6, 6):

  for e in events_matrix:
      if e[0] == 'note':
          if e[3] == 9:
              e[4] = ((e[4] % 128) + 128)
          else:
            e[4] = ((e[4] % 128) + i)

  pitches_counts = [[y[0],y[1]] for y in Counter([y[4] for y in events_matrix if y[0] == 'note']).most_common()]
  pitches_counts.sort(key=lambda x: x[0], reverse=True)
  
  mult_pitches_counts.append(pitches_counts)

patches_list = sorted(list(set([y[3] for y in events_matrix if y[0] == 'patch_change'])))

print('=' * 70)
print('Done!')
print('=' * 70)

#============================================
# MIDI rendering code
#============================================

print('Rendering source MIDI...')
print('=' * 70)

ms_score = MIDI.midi2ms_score(open(f, 'rb').read())

itrack = 1
song_f = []

while itrack < len(ms_score):
    for event in ms_score[itrack]:         
        if event[0] == 'note':
            song_f.append(event)
    itrack += 1

song_f.sort(key=lambda x: x[1])

fname = f.split('.mid')[0]

x = []
y =[]
c = []

colors = ['red', 'yellow', 'green', 'cyan', 'blue', 'pink', 'orange', 'purple', 'gray', 'white', 'gold', 'silver', 'aqua', 'azure', 'bisque', 'coral']

for s in song_f:
  x.append(s[1] / 1000)
  y.append(s[4])
  c.append(colors[s[3]])

if render_MIDI_to_audio:
  FluidSynth("/usr/share/sounds/sf2/FluidR3_GM.sf2", 16000).midi_to_audio(str(fname + '.mid'), str(fname + '.wav'))
  display(Audio(str(fname + '.wav'), rate=16000))

plt.figure(figsize=(14,5))
ax=plt.axes(title=fname)
ax.set_facecolor('black')

plt.scatter(x,y, c=c)
plt.xlabel("Time")
plt.ylabel("Pitch")
plt.show()

"""# (SEARCH AND EXPLORE)"""

#@title Legacy MIDI Matrixes Search (Slow)

#@markdown NOTE: You can stop the search at any time to render partial results

minimum_match_ratio_to_search_for = 0 #@param {type:"slider", min:0, max:500, step:1}
stop_search_on_exact_match = True #@param {type:"boolean"}
skip_exact_matches = False #@param {type:"boolean"}
render_MIDI_to_audio = False #@param {type:"boolean"}

#=================================================================================

matching_type = "minkowski"

def compress_matrix(midi_matrix):

  MX = 38
  MY = 256

  if len(midi_matrix) == MX:

    compressed_matrix = []
    zeros = 0
    zeros_shift = 0
    zeros_count = 0

    for m in midi_matrix:
      for mm in m:
        zeros_shift = max(zeros_shift, mm) + 1

    compressed_matrix.append(zeros_shift)

    for m in midi_matrix:
      if len(m) == MY:
        for mm in m:
          if mm != 0:
            if zeros > 0:
              compressed_matrix.append(zeros+zeros_shift)
              zeros = 0
            compressed_matrix.append(mm)
          
          else:
            zeros += 1
            zeros_count += 1
      
      else:
        print('Wrong matrix format!')
        return [1]

    if zeros > 0:
      compressed_matrix.append(zeros+zeros_shift)

    compressed_matrix.append(zeros_count+zeros_shift)
    compressed_matrix.append(zeros_shift)

    return compressed_matrix

  else:
    print('Wrong matrix format!')
    return [0]

#=================================================================================

def decompress_matrix(compressed_midi_matrix):

  MX = 38
  MY = 256

  zeros_count = 0

  temp_matrix = []
  decompressed_matrix = [[0]*MY for i in range(MX)]

  if compressed_midi_matrix[0] == compressed_midi_matrix[-1]:
    zeros_shift = compressed_midi_matrix[0]
    mcount = 0

    for c in compressed_midi_matrix[1:-2]:
      if c > zeros_shift:
        temp_matrix.extend([0] * (c-zeros_shift))
        zeros_count += (c-zeros_shift)

      else:
        temp_matrix.extend([c])

    if len(temp_matrix) == (MX * MY):

      for i in range(MX):
        for j in range(MY):
          decompressed_matrix[i][j] = copy.deepcopy(temp_matrix[(i*MY) + j])
      
      if len(decompressed_matrix) == MX and zeros_count == (compressed_midi_matrix[-2]-zeros_shift):
        return decompressed_matrix

      else:
        print('Matrix is corrupted!')
        return [len(decompressed_matrix), (MX * MY), zeros_count, (compressed_midi_matrix[-2]-zeros_shift)]
    
    else:
      print('Matrix is corrupted!')
      return [len(temp_matrix), zeros_count]

  else:
    print('Matrix is corrupted!')
    return [0]

#=================================================================================

def batched_scores(matbatch, matrix):

  sco= []
  for D in matbatch:

    dist = pairwise_distances(matrix, decompress_matrix(D[1]), metric=matching_type)[0][0]

    if skip_exact_matches:
      if dist == 0:
        dist = 999999

    if dist <= minimum_match_ratio_to_search_for:
      dist = 999999

    sco.append(dist)

  return sco

#=================================================================================

print('=' * 70)
print('Searching...Please wait...')
print('=' * 70)

scores = []

c_count = multiprocessing.cpu_count()

par = Parallel(n_jobs=c_count)

num_jobs = c_count
scores_per_job = 100

MATRIX_X = [MATRIX] * num_jobs

for i in tqdm(range(0, len(midi_matrixes), (num_jobs*scores_per_job))):
  
  try:

    MAT_BATCHES = []

    for j in range(num_jobs):
      MAT_BATCHES.append(midi_matrixes[i+(j*scores_per_job):i+((j+1)*scores_per_job)])

    output = par(delayed(batched_scores) (MB, MAT) for MB, MAT in zip(MAT_BATCHES, MATRIX_X))

    output1 = []

    for o in output:
      output1.extend(o)

    scores.extend(output1)                            

    if stop_search_on_exact_match:
      if 0 in output1:
        print('=' * 70)
        print('Found exact match!')
        print('Stoping further search...')
        print('=' * 70)
        break

    else:
      if 0 in output1:
        print('=' * 70)
        print('Found exact match!')
        print('=' * 70)
        print('LAMDa Index:', scores.index(min(scores)))
        print('LAMDa File Name:', midi_matrixes[scores.index(min(scores))][0])
        print('=' * 70)
        print('Continuing search...')
        print('=' * 70)

  except KeyboardInterrupt:
    break
  
  except:
    continue

print('Done!')
print('=' * 70)
print('Best match:')
print('=' * 70)   
print(matching_type.title(), 'distance ==', min(scores))
print('LAMDa Index:', scores.index(min(scores)))
print('LAMDa File Name:', midi_matrixes[scores.index(min(scores))][0])
print('=' * 70)

#============================================
# MIDI rendering code
#============================================

print('Rendering source MIDI...')
print('=' * 70)

fn = midi_matrixes[scores.index(min(scores))][0]

try:
  fn_idx = [y[0] for y in LAMD_files_list].index(fn)

  f = LAMD_files_list[fn_idx][1]

  ms_score = MIDI.midi2ms_score(open(f, 'rb').read())

  itrack = 1
  song_f = []

  while itrack < len(ms_score):
      for event in ms_score[itrack]:         
          if event[0] == 'note':
              song_f.append(event)
      itrack += 1

  song_f.sort(key=lambda x: x[1])

  fname = f.split('.mid')[0]

  x = []
  y =[]
  c = []

  colors = ['red', 'yellow', 'green', 'cyan', 'blue', 'pink', 'orange', 'purple', 'gray', 'white', 'gold', 'silver', 'aqua', 'azure', 'bisque', 'coral']

  for s in song_f:
    x.append(s[1] / 1000)
    y.append(s[4])
    c.append(colors[s[3]])

  if render_MIDI_to_audio:
    FluidSynth("/usr/share/sounds/sf2/FluidR3_GM.sf2", 16000).midi_to_audio(str(fname + '.mid'), str(fname + '.wav'))
    display(Audio(str(fname + '.wav'), rate=16000))

  plt.figure(figsize=(14,5))
  ax=plt.axes(title=fname)
  ax.set_facecolor('black')

  plt.scatter(x,y, c=c)
  plt.xlabel("Time")
  plt.ylabel("Pitch")
  plt.show()

except:
  pass

#============================================

print('Top 100 matches')
print('=' * 70)

top_matches = []

for i in range(len(scores)):
  top_matches.append([midi_matrixes[i][0], scores[i]])

top_matches.sort(key=lambda x: x[1])

for t in top_matches[:100]:
  print(t)
  
print('=' * 70)

#@title MIDI Pitches Search (Fast)

#@markdown NOTE: You can stop the search at any time to render partial results

maximum_match_ratio_to_search_for = 1 #@param {type:"slider", min:0, max:1, step:0.01}
pitches_counts_cutoff_threshold_ratio = 0.2 #@param {type:"slider", min:0, max:1, step:0.05}
search_transposed_pitches = False #@param {type:"boolean"}
skip_exact_matches = False #@param {type:"boolean"}
render_MIDI_to_audio = False #@param {type:"boolean"}

print('=' * 70)
print('MIDI Pitches Search')
print('=' * 70)

ratios = []

for d in tqdm(meta_data):

  try:
    p_counts = d[1][3][1]
    p_counts.sort(reverse = True, key = lambda x: x[1])
    max_p_count = p_counts[1][0]
    trimmed_p_counts = [y for y in p_counts if y[1] >= (max_p_count * pitches_counts_cutoff_threshold_ratio)] 
    
    if search_transposed_pitches:
      search_pitches = mult_pitches_counts
    else:
      search_pitches = [mult_pitches_counts[6]]
    
    rat = []

    for m in search_pitches:

      m.sort(reverse = True, key = lambda x: x[1])
      max_pitches_count = m[1][0]
      trimmed_pitches_counts = [y for y in m if y[1] >= (max_pitches_count * pitches_counts_cutoff_threshold_ratio)] 

      num_same_pitches = len(set([T[0] for T in trimmed_p_counts]) & set([m[0] for m in trimmed_pitches_counts]))
      same_pitches_ratio = (num_same_pitches / len(set([m[0] for m in trimmed_p_counts]+[T[0] for T in trimmed_pitches_counts])))

      if skip_exact_matches:
        if same_pitches_ratio == 1:
          same_pitches_ratio = 0

      if same_pitches_ratio > maximum_match_ratio_to_search_for:
        same_pitches_ratio = 0

      rat.append(same_pitches_ratio)

    ratios.append(max(rat))
  
  except KeyboardInterrupt:
    break
  
  except:
    break

max_ratio = max(ratios)
max_ratio_index = ratios.index(max(ratios))

print('FOUND')
print('=' * 70)
print('Match ratio', max_ratio)
print('MIDI file name', meta_data[max_ratio_index][0])
print('=' * 70)
print('First metadata MIDI event', meta_data[max_ratio_index][1][0])
print('=' * 70)

#============================================
# MIDI rendering code
#============================================

print('Rendering source MIDI...')
print('=' * 70)

fn = meta_data[max_ratio_index][0]
fn_idx = [y[0] for y in LAMD_files_list].index(fn)

f = LAMD_files_list[fn_idx][1]

ms_score = MIDI.midi2ms_score(open(f, 'rb').read())

itrack = 1
song_f = []

while itrack < len(ms_score):
    for event in ms_score[itrack]:         
        if event[0] == 'note':
            song_f.append(event)
    itrack += 1

song_f.sort(key=lambda x: x[1])

fname = f.split('.mid')[0]

x = []
y =[]
c = []

colors = ['red', 'yellow', 'green', 'cyan', 'blue', 'pink', 'orange', 'purple', 'gray', 'white', 'gold', 'silver', 'aqua', 'azure', 'bisque', 'coral']

for s in song_f:
  x.append(s[1] / 1000)
  y.append(s[4])
  c.append(colors[s[3]])

if render_MIDI_to_audio:
  FluidSynth("/usr/share/sounds/sf2/FluidR3_GM.sf2", 16000).midi_to_audio(str(fname + '.mid'), str(fname + '.wav'))
  display(Audio(str(fname + '.wav'), rate=16000))

plt.figure(figsize=(14,5))
ax=plt.axes(title=fname)
ax.set_facecolor('black')

plt.scatter(x,y, c=c)
plt.xlabel("Time")
plt.ylabel("Pitch")
plt.show()

#@title MIDI Patches Search (Fast)

#@markdown NOTE: You can stop the search at any time to render partial results

maximum_match_ratio_to_search_for = 1 #@param {type:"slider", min:0, max:1, step:0.01}
skip_exact_matches = False #@param {type:"boolean"}
render_MIDI_to_audio = False #@param {type:"boolean"}

print('=' * 70)
print('MIDI Patches Search')
print('=' * 70)

ratios = []

for d in tqdm(meta_data):

  try:

    p_list= d[1][4][1]

    num_same_patches = len(set(p_list) & set(patches_list))
    
    if len(set(p_list + patches_list)) > 0:
      same_patches_ratio = num_same_patches / len(set(p_list + patches_list))
    else:
      same_patches_ratio = 0

    if skip_exact_matches:
      if same_patches_ratio == 1:
        same_patches_ratio = 0

    if same_patches_ratio > maximum_match_ratio_to_search_for:
      same_patches_ratio = 0
      
    ratios.append(same_patches_ratio)
  
  except KeyboardInterrupt:
    break
  
  except:
    break

max_ratio = max(ratios)
max_ratio_index = ratios.index(max(ratios))

print('FOUND')
print('=' * 70)
print('Match ratio', max_ratio)
print('MIDI file name', meta_data[max_ratio_index][0])
print('=' * 70)
print('Found MIDI patches list', meta_data[max_ratio_index][1][4][1])
print('=' * 70)

#============================================
# MIDI rendering code
#============================================

print('Rendering source MIDI...')
print('=' * 70)

fn = meta_data[max_ratio_index][0]
fn_idx = [y[0] for y in LAMD_files_list].index(fn)

f = LAMD_files_list[fn_idx][1]

ms_score = MIDI.midi2ms_score(open(f, 'rb').read())

itrack = 1
song_f = []

while itrack < len(ms_score):
    for event in ms_score[itrack]:         
        if event[0] == 'note':
            song_f.append(event)
    itrack += 1

song_f.sort(key=lambda x: x[1])

fname = f.split('.mid')[0]

x = []
y =[]
c = []

colors = ['red', 'yellow', 'green', 'cyan', 'blue', 'pink', 'orange', 'purple', 'gray', 'white', 'gold', 'silver', 'aqua', 'azure', 'bisque', 'coral']

for s in song_f:
  x.append(s[1] / 1000)
  y.append(s[4])
  c.append(colors[s[3]])

if render_MIDI_to_audio:
  FluidSynth("/usr/share/sounds/sf2/FluidR3_GM.sf2", 16000).midi_to_audio(str(fname + '.mid'), str(fname + '.wav'))
  display(Audio(str(fname + '.wav'), rate=16000))

plt.figure(figsize=(14,5))
ax=plt.axes(title=fname)
ax.set_facecolor('black')

plt.scatter(x,y, c=c)
plt.xlabel("Time")
plt.ylabel("Pitch")
plt.show()

#@title Metadata Search

#@markdown You can search the metadata by search query or by MIDI md5 hash file name

search_query = "Come To My Window" #@param {type:"string"}
md5_hash_MIDI_file_name = "d9a7e1c6a375b8e560155a5977fc10f8" #@param {type:"string"}
case_sensitive_search = False #@param {type:"boolean"}

fields_to_search = ['track_name', 
                    'text_event', 
                    'lyric', 
                    'copyright_text_event', 
                    'marker',
                    'text_event_08',
                    'text_event_09',
                    'text_event_0a',
                    'text_event_0b',
                    'text_event_0c',
                    'text_event_0d',
                    'text_event_0e',
                    'text_event_0f',
                    ]

print('=' * 70)
print('Los Angeles MIDI Dataset Metadata Search')
print('=' * 70)
print('Searching...')
print('=' * 70)

if md5_hash_MIDI_file_name != '':
  for d in tqdm(meta_data):
    try:
      if d[0] == md5_hash_MIDI_file_name:
        print('Found!')
        print('=' * 70)
        print('Metadata index:', meta_data.index(d))
        print('MIDI file name:', meta_data[meta_data.index(d)][0])
        print('Result:', d[1])
        print('=' * 70)
        break
 
    except KeyboardInterrupt:
      print('Ending search...')
      print('=' * 70)
      break
    
    except:
      print('Ending search...')
      print('=' * 70)
      break

  if d[0] != md5_hash_MIDI_file_name:
    print('Not found!')
    print('=' * 70)
    print('md5 hash was not found!')
    print('Ending search...')
    print('=' * 70)

else:
  for d in tqdm(meta_data):
    try:
      for dd in d[1]:
        if dd[0] in fields_to_search:
          if case_sensitive_search:
            if str(search_query) in str(dd[2]):
              print('Found!')
              print('=' * 70)
              print('Metadata index:', meta_data.index(d))
              print('MIDI file name:', meta_data[meta_data.index(d)][0])
              print('Result:', dd[2])
              print('=' * 70)
          
          else:
            if str(search_query).lower() in str(dd[2]).lower():
              print('Found!')
              print('=' * 70)
              print('Metadata index:', meta_data.index(d))
              print('MIDI file name:', meta_data[meta_data.index(d)][0])
              print('Result:', dd[2])
              print('=' * 70)
    
    except KeyboardInterrupt:
      print('Ending search...')
      print('=' * 70)
      break
    
    except:
      print('Ending search...')
      print('=' * 70)
      break

"""# (MIDI FILE PLAYER)"""

#@title MIDI file player

#@markdown NOTE: You can use md5 hash file name or full MIDI file path to play it

md5_hash_MIDI_file_name = "d9a7e1c6a375b8e560155a5977fc10f8" #@param {type:"string"}
full_path_to_MIDI = "/content/Los-Angeles-MIDI-Dataset/Come-To-My-Window-Modified-Sample-MIDI.mid" #@param {type:"string"}
render_MIDI_to_audio = False #@param {type:"boolean"}

#============================================
# MIDI rendering code
#============================================

print('=' * 70)
print('MIDI file player')
print('=' * 70)

try:

  if os.path.exists(full_path_to_MIDI):
    f = full_path_to_MIDI
    print('Using full path to MIDI')

  else:
    fn = md5_hash_MIDI_file_name
    fn_idx = [y[0] for y in LAMD_files_list].index(fn)
    f = LAMD_files_list[fn_idx][1]
    
    print('Using md5 hash filename')
  
  print('=' * 70)
  print('Rendering MIDI...')
  print('=' * 70)

  ms_score = MIDI.midi2ms_score(open(f, 'rb').read())

  itrack = 1
  song_f = []

  while itrack < len(ms_score):
      for event in ms_score[itrack]:         
          if event[0] == 'note':
              song_f.append(event)
      itrack += 1

  song_f.sort(key=lambda x: x[1])

  fname = f.split('.mid')[0]

  x = []
  y =[]
  c = []

  colors = ['red', 'yellow', 'green', 'cyan', 'blue', 'pink', 'orange', 'purple', 'gray', 'white', 'gold', 'silver', 'aqua', 'azure', 'bisque', 'coral']

  for s in song_f:
    x.append(s[1] / 1000)
    y.append(s[4])
    c.append(colors[s[3]])

  if render_MIDI_to_audio:
    FluidSynth("/usr/share/sounds/sf2/FluidR3_GM.sf2", 16000).midi_to_audio(str(fname + '.mid'), str(fname + '.wav'))
    display(Audio(str(fname + '.wav'), rate=16000))

  plt.figure(figsize=(14,5))
  ax=plt.axes(title=fname)
  ax.set_facecolor('black')

  plt.scatter(x,y, c=c)
  plt.xlabel("Time")
  plt.ylabel("Pitch")
  plt.show()

except:
  print('File not found!!!')
  print('Check the filename!')
  print('=' * 70)

"""# (COLAB MIDI FILES LOCATOR/DOWNLOADER)"""

#@title Loacate and/or download desired MIDI files by MIDI md5 hash file names

MIDI_md5_hash_file_name_1 = "d9a7e1c6a375b8e560155a5977fc10f8" #@param {type:"string"}
MIDI_md5_hash_file_name_2 = "" #@param {type:"string"}
MIDI_md5_hash_file_name_3 = "" #@param {type:"string"}
MIDI_md5_hash_file_name_4 = "" #@param {type:"string"}
MIDI_md5_hash_file_name_5 = "" #@param {type:"string"}
download_located_files = False #@param {type:"boolean"}

print('=' * 70)
print('MIDI files locator and downloader')
print('=' * 70)

md5_list = []

if MIDI_md5_hash_file_name_1 != '':
  md5_list.append(MIDI_md5_hash_file_name_1)

if MIDI_md5_hash_file_name_2 != '':
  md5_list.append(MIDI_md5_hash_file_name_2)

if MIDI_md5_hash_file_name_3 != '':
  md5_list.append(MIDI_md5_hash_file_name_3)

if MIDI_md5_hash_file_name_4 != '':
  md5_list.append(MIDI_md5_hash_file_name_4)

if MIDI_md5_hash_file_name_5 != '':
  md5_list.append(MIDI_md5_hash_file_name_5)

if len(md5_list) > 0:
  for m in md5_list:
    try:

      fn = m
      fn_idx = [y[0] for y in LAMD_files_list].index(fn)
      f = LAMD_files_list[fn_idx][1]

      print('Found md5 hash file name', m)

      location_str = ''

      fl = f.split('/')
      for fa in fl[:-1]:
        if fa != '' and fa != 'content':
          location_str += '/'
          location_str += str(fa)

      print('Colab location/folder', location_str)

      if download_located_files:
        print('Downloading MIDI file', str(m) + '.mid')
        files.download(f)

      print('=' * 70)

    except:
      print('md5 hash file name', m, 'not found!!!')
      print('Check the file name!')
      print('=' * 70)
      continue

else:
  print('No md5 hash file names were specified!')
  print('Check input!')
  print('=' * 70)

"""# (CUSTOM ANALYSIS TEMPLATE)"""

#@title Los Angeles MIDI Dataset Reader

print('=' * 70)
print('Los Angeles MIDI Dataset Reader')
print('=' * 70)
print('Starting up...')
print('=' * 70)

###########

print('Loading MIDI files...')
print('This may take a while on a large dataset in particular.')

dataset_addr = "/content/LAMD/MIDIs"

# os.chdir(dataset_addr)
filez = list()
for (dirpath, dirnames, filenames) in os.walk(dataset_addr):
    filez += [os.path.join(dirpath, file) for file in filenames]

if filez == []:
    print('Could not find any MIDI files. Please check Dataset dir...')
    print('=' * 70)

print('=' * 70)
print('Randomizing file list...')
random.shuffle(filez)
print('=' * 70)

###########

START_FILE_NUMBER = 0
LAST_SAVED_BATCH_COUNT = 0

input_files_count = START_FILE_NUMBER
files_count = LAST_SAVED_BATCH_COUNT

stats = [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]

print('Reading MIDI files. Please wait...')
print('=' * 70)

for f in tqdm(filez[START_FILE_NUMBER:]):
    try:
        input_files_count += 1

        fn = os.path.basename(f)
        fn1 = fn.split('.mid')[0]

        #=======================================================
        # START PROCESSING
        #=======================================================

        # Convering MIDI to score with MIDI.py module
        score = MIDI.midi2score(open(f, 'rb').read())

        events_matrix = []
        
        itrack = 1

        while itrack < len(score):
            for event in score[itrack]:        
              events_matrix.append(event)
            itrack += 1

        # Sorting...
        events_matrix.sort(key=lambda x: x[1])

        if len(events_matrix) > 0:
          
          #=======================================================
          # INSERT YOUR CUSTOM ANAYLSIS CODE RIGHT HERE
          #=======================================================

          # Processed files counter
          files_count += 1

          # Saving every 5000 processed files
          if files_count % 10000 == 0:
            print('=' * 70)
            print('Processed so far:', files_count, 'out of', input_files_count, '===', files_count / input_files_count, 'good files ratio')
            print('=' * 70)

    except KeyboardInterrupt:
        print('Saving current progress and quitting...')
        break  

    except Exception as ex:
        print('WARNING !!!')
        print('=' * 70)
        print('Bad MIDI:', f)
        print('Error detected:', ex)
        print('=' * 70)
        continue

print('=' * 70)
print('Final files counts:', files_count, 'out of', input_files_count, '===', files_count / input_files_count, 'good files ratio')
print('=' * 70)

print('Resulting Stats:')
print('=' * 70)
print('Total good processed MIDI files:', files_count)
print('=' * 70)
print('Done!')   
print('=' * 70)

"""# Congrats! You did it! :)"""