#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
from dataset_utils import load_training, load_validation
from model_utils import model_builder
from training_utils import train_model
import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
import os
### Load Data
#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# training
training_foldername = '../../nesmdb24_seprsco/train/'
train_save_filename = 'transformed_dataset.json'
dataset , labels2int_map , int2labels_map = \
load_training(training_foldername, train_save_filename)
# validation
validation_foldername = '../../nesmdb24_seprsco/valid/'
val_save_filename = 'transformed_val_dataset.json'
val_dataset = load_validation(validation_foldername,\
labels2int_map, val_save_filename)
### Build Model
#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
### Model Parameters
rnn_units = 512
input_dim = len(int2labels_map)
emb_output_dim = 256
batch_size = 100
lstm_maxnorm, dense_maxnorm = 4, 4
import numpy as np
import matplotlib.pyplot as plt
import os
### Load Data
#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# Parameters for shape of dataset (note these are also used for model def.)
measures = 8
measure_len = 96
# training
training_foldername = '../../nesmdb24_seprsco/train/'
train_save_filename = 'transformed_dataset.json'
dataset , labels2int_map , int2labels_map = \
load_training(training_foldername, train_save_filename,
measures = measures, measure_len = measure_len)
# validation
validation_foldername = '../../nesmdb24_seprsco/valid/'
val_save_filename = 'transformed_val_dataset.json'
val_dataset = load_validation(validation_foldername,\
labels2int_map, val_save_filename,
measures = measures, measure_len = measure_len)
### Build Model
#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
### Model Parameters
latent_dim = 124
input_dim = len(int2labels_map) - 1
dropout = .1
# NOTE - nesmdb folder manually added to environment libraries
from dataset_utils import load_training, load_track
from model_utils import model_builder
from generation_utils import generate_track, generate_seprsco
import nesmdb
from nesmdb.vgm.vgm_to_wav import save_vgmwav
import tensorflow as tf
import numpy as np
import os
### Load Mappings
#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
data_foldername = '../../nesmdb24_seprsco/train/'
save_filename = 'transformed_dataset.json'
dataset , labels2int_map , int2labels_map = load_training(data_foldername, save_filename)
# Delete dataset to free up memory
del dataset
### Reinitiate Model
#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# Define Model Structure
model = model_builder(rnn_units = 512,
input_dim = len(int2labels_map),
emb_output_dim = 256,
batch_size = 1,
lstm_maxnorm = 4,
dense_maxnorm = 4,
lstm_dropout = .5,
dense_dropout = .5)