from Data.DataGeneratorsLeadMeta import CombinedGenerator as CGLeadMeta
from Nets.CombinedNetwork import BarEmbedding, RhythmEncoder, RhythmNetwork,\
MelodyEncoder, MelodyNetwork,\
CombinedNetwork
from Nets.MetaEmbedding import MetaEmbedding
from Nets.MetaPredictor import MetaPredictor
import os
#%% META
meta_embedder = MetaEmbedding.from_saved_custom("meta_saved")
meta_embed_size = meta_embedder.embed_size
meta_predictor = MetaPredictor.from_saved_custom("meta_saved")
meta_predictor.freeze()
#%%
rc = 2
mc = 2
cg = CGLeadMeta("Data/lessfiles",
save_conversion_params=False,
to_list=False,
meta_prep_f=meta_embedder.predict)
cg.get_num_pieces()
def without_lead(cg_inst):
top_dir = "Trainings"
# save_dir = asctime().split()
# save_dir = "_".join([*save_dir[0:3], *save_dir[3].split(":")[:2]])
save_dir = "meta_no_embed_wrong_loss2"
if not os.path.isdir("/".join([top_dir, save_dir, "chord"])):
os.makedirs("/".join([top_dir, save_dir, "chord"]))
# META
# meta_embedder = MetaEmbedding.from_saved_custom("/".join([top_dir, save_dir, "meta"]))
# meta_embed_size = meta_embedder.embed_size
meta_predictor = MetaPredictor.from_saved_custom("/".join([top_dir, save_dir, "meta"]))
meta_predictor.freeze()
#%%
music_dir = "../../Data/music21"
ch_gen = ChordGenerator(music_dir, save_conversion_params="/".join([top_dir, save_dir]),
to_list=False, meta_prep_f=None) # None
# data_iter = ch_gen.generate_forever(batch_size=24)
x, y = ch_gen.list_data()
#%%
comb_net = CombinedNetwork.from_saved_custom("/".join([top_dir, save_dir, "weights"]),
# meta_emb, eval_results = get_meta_embedder(meta_examples,
# embed_size=10,
# epochs=10,
# evaluate=True, verbose=1)
#
# print("MetaEmbedding trained!\n\tevaluation results:\n\t",
# eval_results)
#%%
pred_meta_gen = gen_preds_and_meta(cg, forever=True)
r_params = (None, cg.rhythm_V)
m_params = (48, cg.melody_V)
mp = MetaPredictor(r_params, m_params, 10, #meta_embed_size
12, 24)
#%%
mp.fit_generator(pred_meta_gen,
steps_per_epoch=cg.num_pieces,
epochs=6)
#%%
# meta_emb.save_model_custom("/".join([top_dir, save_dir, "meta"]))
#%%
mp.save_model_custom("/".join([top_dir, save_dir, "meta"]))
# -*- coding: utf-8 -*- from Nets.MetaPredictor import MetaPredictor from Nets.CombinedNetwork import CombinedNetwork #%% save_dir = "Trainings/meta_no_embed_wrong_loss/" meta_pred = MetaPredictor.from_saved_custom(save_dir + "meta") #%% comb_net = CombinedNetwork.from_saved_custom(save_dir + "weights", meta_pred)
meta_emb, eval_results = get_meta_embedder(meta_examples,
embed_size=10,
epochs=10,
evaluate=True,
verbose=1)
print("MetaEmbedding trained!\n\tevaluation results:\n\t", eval_results)
#%%
pred_meta_gen = gen_preds_and_meta(cg, meta_emb, forever=True)
r_params = (None, cg.rhythm_V)
m_params = (48, cg.melody_V)
mp = MetaPredictor(r_params, m_params, meta_emb.embed_size, 12, 24)
#%%
mp.fit_generator(pred_meta_gen, steps_per_epoch=cg.num_pieces, epochs=2)
#%%
meta_emb.save_model_custom("/".join([top_dir, save_dir, "meta"]))
#%%
mp.save_model_custom("/".join([top_dir, save_dir, "meta"]))
#%%