李宏毅_機(jī)器學(xué)習(xí)_作業(yè)4（詳解）_HW4 Classify the speakers

這篇具有很好參考價(jià)值的文章主要介紹了李宏毅_機(jī)器學(xué)習(xí)_作業(yè)4（詳解）_HW4 Classify the speakers。希望對(duì)大家有所幫助。如果存在錯(cuò)誤或未考慮完全的地方，請(qǐng)大家不吝賜教，您也可以點(diǎn)擊"舉報(bào)違法"按鈕提交疑問。

本次作業(yè)需要學(xué)習(xí)完transformer后完成！

Task

做語者辨識(shí)任務(wù)，一共有600個(gè)語者，給了每一個(gè)語者的語音feature進(jìn)行訓(xùn)練，然后通過test_feature進(jìn)行語者辨識(shí)。（本質(zhì)上還是分類任務(wù)Classification）
Simple(0.60824):run sample code and know how to use transformer
Medium(0.70375):know how to adjust parameters of transformer
Strong(0.77750):construct conformer
Boss(0.86500):implement self-attention pooling and additive margin softmax

使用kaggle訓(xùn)練作業(yè)模型

助教樣例code解讀

數(shù)據(jù)集分析

mapping.json文件

將speakers的id映射到編號(hào)0~599，因?yàn)橐还灿?00個(gè)不同的speaker需要對(duì)語音進(jìn)行分類
metadata.json文件

存放的是training data，本次實(shí)驗(yàn)沒有專門設(shè)置validation data，需要從training data中劃分validation data
n_mels：在對(duì)語音數(shù)據(jù)進(jìn)行處理時(shí)，從每一個(gè)時(shí)間維度上選取n_mels個(gè)維度來表示這個(gè)feature
speakers：以key-value形式存放speakers的id和所有feature(每個(gè)speaker都有多個(gè)feature)
feature_path：這個(gè)feature的文件名
mel_len：每一個(gè)feature的長(zhǎng)度（每一個(gè)可能都不一樣，后期需要處理）
testdata.json文件

與metadata形式類似，需要我們進(jìn)行語者辨識(shí)。utterance：話語; 言論

Dataset

本次實(shí)驗(yàn)的數(shù)據(jù)來源于 Voxceleb2語音數(shù)據(jù)集，是真實(shí)世界中語者的語音，作業(yè)中選取了600個(gè)語者，和他們的語音進(jìn)行訓(xùn)練

import os
import json
import torch
import random
from pathlib import Path
from torch.utils.data import Dataset
from torch.nn.utils.rnn import pad_sequence
 
 
class myDataset(Dataset):
	def __init__(self, data_dir, segment_len=128):
		self.data_dir = data_dir
		self.segment_len = segment_len
	
		# Load the mapping from speaker neme to their corresponding id. 
		mapping_path = Path(data_dir) / "mapping.json"  #mapping_path: Dataset\mapping.json
		mapping = json.load(mapping_path.open()) 
		#mapping: {'speaker2id': {'id00464': 0, 'id00559': 1,
		self.speaker2id = mapping["speaker2id"] 
		#self.speaker2id: {'id00464': 0, 'id00559': 1, 'id00578': 2, 'id00905': 3,...
	
		# Load metadata of training data.
		metadata_path = Path(data_dir) / "metadata.json"        
		metadata = json.load(open(metadata_path))["speakers"] #metadata中存放的key是speaker_id,value是每個(gè)speaker的feature和對(duì)應(yīng)長(zhǎng)度
	
		# Get the total number of speaker.
		self.speaker_num = len(metadata.keys())
		self.data = []
		for speaker in metadata.keys():  #遍歷每一個(gè)spearker_id
			for utterances in metadata[speaker]: #通過speaker_id取出speaker的所有feature和len
			"""
                utterances格式：
                {'feature_path': 'uttr-18e375195dc146fd8d14b8a322c29b90.pt', 'mel_len': 435}
               {'feature_path': 'uttr-da9917d5853049178487c065c9e8b718.pt', 'mel_len': 490}...
       """
				self.data.append([utterances["feature_path"], self.speaker2id[speaker]])
        #self.data:[['uttr-18e375195dc146fd8d14b8a322c29b90.pt', 436], 
        #           ['uttr-da9917d5853049178487c065c9e8b718.pt', 436],...
        #一共600個(gè)speaker,436表示第436個(gè)speaker
 
	def __len__(self):
			return len(self.data)
 
	def __getitem__(self, index):
		feat_path, speaker = self.data[index] #feature和speaker編號(hào)[0,599]
		# Load preprocessed mel-spectrogram.
		mel = torch.load(os.path.join(self.data_dir, feat_path)) #加載feature
		#mel.size():torch.Size([490, 40])

		# Segmemt mel-spectrogram into "segment_len" frames.
		if len(mel) > self.segment_len: #將feature切片成固定長(zhǎng)度
			# Randomly get the starting point of the segment.
			start = random.randint(0, len(mel) - self.segment_len)  #隨機(jī)選取切片起始點(diǎn)
			# Get a segment with "segment_len" frames.
			mel = torch.FloatTensor(mel[start:start+self.segment_len])#截取長(zhǎng)度為segment_len的片段 mel.size():torch.Size([128, 40])
		else:
			mel = torch.FloatTensor(mel) #為什么小于segment_len不填充？  填充在dataloader中完成
		# Turn the speaker id into long for computing loss later.
		speaker = torch.FloatTensor([speaker]).long() #將speaker的編號(hào)轉(zhuǎn)為long類型
		return mel, speaker
 
	def get_speaker_number(self):
		return self.speaker_num  #600

Dataloader

主要任務(wù)：1.劃分驗(yàn)證集 2.將長(zhǎng)度小于segment_len的mel進(jìn)行padding 3.生成dataloader

import torch
from torch.utils.data import DataLoader, random_split
from torch.nn.utils.rnn import pad_sequence


def collate_batch(batch):  #用于整理數(shù)據(jù)的函數(shù)，參數(shù)為dataloader中的一個(gè)batch
	# Process features within a batch.
	"""Collate a batch of data."""
	mel, speaker = zip(*batch)  #zip拆包，將一個(gè)batch中的mel和speaker分開，各自單獨(dú)形成一個(gè)數(shù)組
	# Because we train the model batch by batch, we need to pad the features in the same batch to make their lengths the same.
    #mel中元素長(zhǎng)度不相同時(shí)，將所有的mel元素填充到最長(zhǎng)的元素的長(zhǎng)度，填充的值由padding_value決定
	mel = pad_sequence(mel, batch_first=True, padding_value=-20)    # pad log 10^(-20) which is very small value.
	# mel: (batch size, length, 40)
	return mel, torch.FloatTensor(speaker).long()


def get_dataloader(data_dir, batch_size, n_workers):
	"""Generate dataloader"""
	dataset = myDataset(data_dir)
	speaker_num = dataset.get_speaker_number()
	# Split dataset into training dataset and validation dataset
	trainlen = int(0.9 * len(dataset))
	lengths = [trainlen, len(dataset) - trainlen] 
	trainset, validset = random_split(dataset, lengths) #無覆蓋的隨機(jī)劃分訓(xùn)練集和驗(yàn)證集

	train_loader = DataLoader(
		trainset,
		batch_size=batch_size,
		shuffle=True,
		drop_last=True,
		num_workers=n_workers,
		pin_memory=True,
		collate_fn=collate_batch,
	)
	valid_loader = DataLoader(
		validset,
		batch_size=batch_size,
		num_workers=n_workers,
		drop_last=True,
		pin_memory=True,
		collate_fn=collate_batch,
	)

	return train_loader, valid_loader, speaker_num

Model

最關(guān)鍵部分，transformer運(yùn)用
transformer基礎(chǔ)架構(gòu)來自于論文： Attention Is All You Need
論文解讀：李沐大神的論文帶讀，用了都說好

這里是分類任務(wù)，僅需要使用Encoder部分
pytorch官方文檔： torch.nn.TransformerEncoderLayer

import torch
import torch.nn as nn
import torch.nn.functional as F


class Classifier(nn.Module):
	def __init__(self, d_model=80, n_spks=600, dropout=0.1):
		super().__init__()
		# Project the dimension of features from that of input into d_model.
		self.prenet = nn.Linear(40, d_model)
		# TODO:
		#   Change Transformer to Conformer.
		#   https://arxiv.org/abs/2005.08100
        
        #對(duì)于文本分類等下游任務(wù)，只需要用到Encoder部分即可
        #nhead:multi_head_attention中head個(gè)數(shù)
        #d_model:輸入的feature的個(gè)數(shù)
        #dim_feedforward:feedforward network的維度
        #dropout默認(rèn)0.1
		self.encoder_layer = nn.TransformerEncoderLayer(
			d_model=d_model, dim_feedforward=256, nhead=2
		)
		# self.encoder = nn.TransformerEncoder(self.encoder_layer, num_layers=2)

		# Project the the dimension of features from d_model into speaker nums.
		self.pred_layer = nn.Sequential(
			nn.Linear(d_model, d_model),
			nn.ReLU(),
			nn.Linear(d_model, n_spks),
		)

	def forward(self, mels):
		"""
		args:
			mels: (batch size, length, 40)
		return:
			out: (batch size, n_spks)
		"""
		# out: (batch size, length, d_model)   length=segment_len
		out = self.prenet(mels)
		# out: (length, batch size, d_model)
		out = out.permute(1, 0, 2) #交換dim=0和dim=1
		# The encoder layer expect features in the shape of (length, batch size, d_model).
		out = self.encoder_layer(out)
		# out: (batch size, length, d_model)
		out = out.transpose(0, 1)  #轉(zhuǎn)置dim=0和dim=1
		# mean pooling
		stats = out.mean(dim=1) #可以理解為求平均并去除維度1  stats.size():（batch_size,d_model）

		# out: (batch, n_spks)
		out = self.pred_layer(stats)
		return out

Learning rate schedule

當(dāng)batch設(shè)置的比較大的時(shí)候通常需要比較大的學(xué)習(xí)率（通常batch_size和學(xué)習(xí)率成正比），但在剛開始訓(xùn)練時(shí)，參數(shù)是隨機(jī)初始化的，梯度也比較大，這時(shí)學(xué)習(xí)率也比較大，會(huì)使得訓(xùn)練不穩(wěn)定。
warm up 方法就是在最初幾輪迭代采用比較小的學(xué)習(xí)率，等梯度下降到一定程度再恢復(fù)初始學(xué)習(xí)率
------《神經(jīng)網(wǎng)絡(luò)與深度學(xué)習(xí)》

import math

import torch
from torch.optim import Optimizer
from torch.optim.lr_scheduler import LambdaLR


def get_cosine_schedule_with_warmup(
	optimizer: Optimizer,
	num_warmup_steps: int,
	num_training_steps: int,
	num_cycles: float = 0.5,
	last_epoch: int = -1,
):
	"""
	Create a schedule with a learning rate that decreases following the values of the cosine function between the
	initial lr set in the optimizer to 0, after a warmup period during which it increases linearly between 0 and the
	initial lr set in the optimizer.

	Args:
		optimizer (:class:`~torch.optim.Optimizer`):
		The optimizer for which to schedule the learning rate.
		num_warmup_steps (:obj:`int`):
		The number of steps for the warmup phase.
		num_training_steps (:obj:`int`):
		The total number of training steps.
		num_cycles (:obj:`float`, `optional`, defaults to 0.5):
		The number of waves in the cosine schedule (the defaults is to just decrease from the max value to 0
		following a half-cosine).
		last_epoch (:obj:`int`, `optional`, defaults to -1):
		The index of the last epoch when resuming training.

	Return:
		:obj:`torch.optim.lr_scheduler.LambdaLR` with the appropriate schedule.
	"""
	def lr_lambda(current_step):
		# Warmup
		if current_step < num_warmup_steps:
			return float(current_step) / float(max(1, num_warmup_steps))
		# decadence
		progress = float(current_step - num_warmup_steps) / float(
			max(1, num_training_steps - num_warmup_steps)
		)
		return max(
			0.0, 0.5 * (1.0 + math.cos(math.pi * float(num_cycles) * 2.0 * progress))
		)

	return LambdaLR(optimizer, lr_lambda, last_epoch)

Model Function

調(diào)用自定義model的forward部分，每遍歷一個(gè)batch都要調(diào)用一次model_fn

import torch


def model_fn(batch, model, criterion, device):
	"""Forward a batch through the model."""

	mels, labels = batch
  
	#print("model_fn_mels.size():",mels.size())  
    # out:torch.Size([16, 128, 40]) [batch_size,segment_len,40]
	mels = mels.to(device)
	labels = labels.to(device)

	outs = model(mels)

	loss = criterion(outs, labels)

	# Get the speaker id with highest probability.
	preds = outs.argmax(1)
	# Compute accuracy.
	accuracy = torch.mean((preds == labels).float())

	return loss, accuracy

Validate

計(jì)算驗(yàn)證集上的準(zhǔn)確率

from tqdm import tqdm
import torch


def valid(dataloader, model, criterion, device): 
	"""Validate on validation set."""

	model.eval()
	running_loss = 0.0
	running_accuracy = 0.0
	#驗(yàn)證集5667個(gè)
	pbar = tqdm(total=len(dataloader.dataset), ncols=0, desc="Valid", unit=" uttr")

	for i, batch in enumerate(dataloader):
		with torch.no_grad():
			loss, accuracy = model_fn(batch, model, criterion, device)
			running_loss += loss.item()
			running_accuracy += accuracy.item()

		pbar.update(dataloader.batch_size)
		pbar.set_postfix(
			loss=f"{running_loss / (i+1):.2f}",
			accuracy=f"{running_accuracy / (i+1):.2f}",
		)

	pbar.close()
	model.train()

	return running_accuracy / len(dataloader)

Main function

開始跑模型，這里與之前的作業(yè)有不同的地方。前幾個(gè)作業(yè)是跑完一個(gè)epoch也就是完整訓(xùn)練集，再開始跑驗(yàn)證集。這里是跑valid_steps個(gè)batch，跑一遍驗(yàn)證集。

from tqdm import tqdm

import torch
import torch.nn as nn
from torch.optim import AdamW
from torch.utils.data import DataLoader, random_split


def parse_args():
	"""arguments"""
	config = {
		"data_dir": "./Dataset",
		"save_path": "model.ckpt",
		"batch_size": 16,
		"n_workers": 0,
		"valid_steps": 2000,
		"warmup_steps": 1000,
		"save_steps": 10000,
		"total_steps": 70000,
	}

	return config


def main(
	data_dir,
	save_path,
	batch_size,
	n_workers,
	valid_steps,
	warmup_steps,
	total_steps,
	save_steps,
):
	"""Main function."""
	device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
	print(f"[Info]: Use {device} now!")

	train_loader, valid_loader, speaker_num = get_dataloader(data_dir, batch_size, n_workers)
	train_iterator = iter(train_loader) #iter()生成迭代器，以batch為單位
	#print("train_iterator:",train_iterator) #<torch.utils.data.dataloader._SingleProcessDataLoaderIter object at 0x000001FD07C558D0>
	print(f"[Info]: Finish loading data!",flush = True)

	model = Classifier(n_spks=speaker_num).to(device)
	criterion = nn.CrossEntropyLoss()
	optimizer = AdamW(model.parameters(), lr=1e-3)
	scheduler = get_cosine_schedule_with_warmup(optimizer, warmup_steps, total_steps) #上面定義的warm up函數(shù)
	print(f"[Info]: Finish creating model!",flush = True)

	best_accuracy = -1.0
	best_state_dict = None

	pbar = tqdm(total=valid_steps, ncols=0, desc="Train", unit=" step") 
	#train valid_steps個(gè)batch再跑驗(yàn)證集

	for step in range(total_steps): #一共運(yùn)行total_Steps輪，這里沒有epoch的概念
		# Get data
		try:
			batch = next(train_iterator) #next()返回迭代器的下一個(gè)項(xiàng)目，即下一個(gè)batch
			#print("batch[0].size():",batch[0].size())    
			#out:torch.Size([16, 128, 40]) [batch_size,segment_len,40]       
		except StopIteration:  # 不指定 default 且迭代器元素耗盡, 將引發(fā) StopIteration 異常
			train_iterator = iter(train_loader)
			batch = next(train_iterator)

		loss, accuracy = model_fn(batch, model, criterion, device) #計(jì)算當(dāng)前batch的loss和acc
		#print("loss:",loss) #tensor(6.3915, device='cuda:0', grad_fn=<NllLossBackward0>)            
		batch_loss = loss.item() # loss是張量，item()可以取出張量中的值
		#print("batch_loss:",batch_loss) #batch_loss: 6.391468048095703
		batch_accuracy = accuracy.item()

		# Updata model 反向傳播更新參數(shù)，每跑一個(gè)batch都會(huì)更新
		loss.backward()
		optimizer.step()
		scheduler.step()
		optimizer.zero_grad()

		# Log
		pbar.update() #打印當(dāng)前l(fā)oss和acc
		pbar.set_postfix(
			loss=f"{batch_loss:.2f}",
			accuracy=f"{batch_accuracy:.2f}",
			step=step + 1,
		)

		# Do validation
		if (step + 1) % valid_steps == 0: #經(jīng)過valid_steps開始跑驗(yàn)證集
			pbar.close()

			valid_accuracy = valid(valid_loader, model, criterion, device) #計(jì)算valid_acc

			# keep the best model
			if valid_accuracy > best_accuracy:
				best_accuracy = valid_accuracy
				best_state_dict = model.state_dict() #保存模型參數(shù)

			pbar = tqdm(total=valid_steps, ncols=0, desc="Train", unit=" step")

		# Save the best model so far.
		if (step + 1) % save_steps == 0 and best_state_dict is not None: #每save_steps輪會(huì)保存一次當(dāng)前最好模型
			torch.save(best_state_dict, save_path)
			pbar.write(f"Step {step + 1}, best model saved. (accuracy={best_accuracy:.4f})")

	pbar.close()


if __name__ == "__main__":
	main(**parse_args())

李宏毅_機(jī)器學(xué)習(xí)_作業(yè)4（詳解）_HW4 Classify the speakers

Inference

inference:推理，就是跑testing data
類比training即可

Main function of inference

類似Main function

樣例code得分

李宏毅_機(jī)器學(xué)習(xí)_作業(yè)4（詳解）_HW4 Classify the speakers

Medium

調(diào)整參數(shù)過medium
d_model=160
n_head=8
num_layers=2
linear layer:1層
total_steps=100000
李宏毅_機(jī)器學(xué)習(xí)_作業(yè)4（詳解）_HW4 Classify the speakers

李宏毅_機(jī)器學(xué)習(xí)_作業(yè)4（詳解）_HW4 Classify the speakers 這一輪train上準(zhǔn)確率100%，只雖然只進(jìn)行了13步，但從loss上可以看出是有過擬合的

Strong

Transformer->Conformer

先上結(jié)果，未過strong
李宏毅_機(jī)器學(xué)習(xí)_作業(yè)4（詳解）_HW4 Classify the speakers
嚴(yán)重過擬合，在訓(xùn)練集和驗(yàn)證集上均有過擬合現(xiàn)象，驗(yàn)證集上的準(zhǔn)確率遠(yuǎn)高于測(cè)試集上結(jié)果

論文地址: Conformer
conformer的思路很簡(jiǎn)單，就是將Transformer和CNN進(jìn)行結(jié)合。原因：
1.Transformer中由于attention機(jī)制，擁有很好的全局性。
2.CNN擁有較好的局部性，可以對(duì)細(xì)粒度的信息進(jìn)行提取。
兩者結(jié)合在語音上有較好的效果。論文中闡述了具體的model架構(gòu)。

首先 pip conformer包

!pip install conformer

導(dǎo)入conformer包

from conformer import ConformerBlock

修改module

import torch
import torch.nn as nn
import torch.nn.functional as F


class Classifier(nn.Module):
	def __init__(self, d_model=512, n_spks=600, dropout=0.1):
		super().__init__()
		# Project the dimension of features from that of input into d_model.
		self.prenet = nn.Linear(40, d_model)
		# TODO:
		#   Change Transformer to Conformer.
		#   https://arxiv.org/abs/2005.08100
        
        #對(duì)于文本分類等下游任務(wù)，只需要用到Encoder部分即可
        #nhead:multi_head_attention中head個(gè)數(shù)
        #d_model:輸入的feature的個(gè)數(shù)
        #dim_feedforward:feedforward network的維度
        #dropout默認(rèn)0.1
		#self.encoder_layer = nn.TransformerEncoderLayer(
			#d_model=d_model, dim_feedforward=256, nhead=8
		#)
		#self.encoder = nn.TransformerEncoder(self.encoder_layer, num_layers=2)
		self.conformer_block=ConformerBlock(
		dim=d_model,
		dim_head=64,
		heads=8,
		ff_mult=4,
		conv_expansion_factor=2,
		conv_kernel_size=31,
		attn_dropout=dropout,
		ff_dropout=dropout,
		conv_dropout=dropout
		)        
		# Project the the dimension of features from d_model into speaker nums.
		self.pred_layer = nn.Sequential(
			#nn.Linear(d_model, d_model),
			#nn.ReLU(),
			nn.Linear(d_model, n_spks),
		)

	def forward(self, mels):
		"""
		args:
			mels: (batch size, length, 40)
		return:
			out: (batch size, n_spks)
		"""
		# out: (batch size, length, d_model)   length=segment_len
		out = self.prenet(mels)
		# out: (length, batch size, d_model)
		out = out.permute(1, 0, 2) #交換dim=0和dim=1
		# The encoder layer expect features in the shape of (length, batch size, d_model).
		out = self.conformer_block(out)
		# out: (batch size, length, d_model)
		out = out.transpose(0, 1)  #轉(zhuǎn)置dim=0和dim=1
		# mean pooling
		stats = out.mean(dim=1) #可以理解為求平均并去除維度1  stats.size():（batch_size,d_model）

		# out: (batch, n_spks)
		out = self.pred_layer(stats)
		return out

Self-attention pooling

self attention pooling論文
主要看論文中的self-attention pooling架構(gòu)，和mean pooling相比之下，self-attention pooling是通過可學(xué)習(xí)參數(shù)來進(jìn)行pooling，相比mean pooling可以提取到一些信息。
參考大佬視頻講解
代碼：

#self attention pooling類實(shí)現(xiàn)
import torch.nn.functional as F
import torch.nn as nn
class Self_Attentive_Pooling(nn.Module):
   def __init__(self,dim):
       super(Self_Attentive_Pooling,self).__init__()
       self.sap_linear=nn.Linear(dim,dim)
       self.attention=nn.Parameter(torch.FloatTensor(dim,1))
       
   def forward(self,x):
       x=x.permute(0,2,1)
       h=torch.tanh(self.sap_linear(x))
       w=torch.matmul(h,self.attention).squeeze(dim=2)
       w=F.softmax(w,dim=1).view(x.size(0),x.size(1),1)
       x=torch.sum(x*w,dim=1)
       return x

修改model:

import torch
import torch.nn as nn
import torch.nn.functional as F


class Classifier(nn.Module):
   def __init__(self, d_model=512, n_spks=600, dropout=0.1):
   	super().__init__()
   	# Project the dimension of features from that of input into d_model.
   	self.prenet = nn.Linear(40, d_model)
   	# TODO:
   	#   Change Transformer to Conformer.
   	#   https://arxiv.org/abs/2005.08100
       
       #對(duì)于文本分類等下游任務(wù)，只需要用到Encoder部分即可
       #nhead:multi_head_attention中head個(gè)數(shù)
       #d_model:輸入的feature的個(gè)數(shù)
       #dim_feedforward:feedforward network的維度
       #dropout默認(rèn)0.1
   	#self.encoder_layer = nn.TransformerEncoderLayer(
   		#d_model=d_model, dim_feedforward=256, nhead=8
   	#)
   	#self.encoder = nn.TransformerEncoder(self.encoder_layer, num_layers=2)
   	self.conformer_block=ConformerBlock(
   	dim=d_model,
   	dim_head=64,
   	heads=8,
   	ff_mult=4,
   	conv_expansion_factor=2,
   	conv_kernel_size=31,
   	attn_dropout=dropout,
   	ff_dropout=dropout,
   	conv_dropout=dropout
   	)        
   	# Project the the dimension of features from d_model into speaker nums.
   	self.pooling=Self_Attentive_Pooling(d_model)
   	self.pred_layer = nn.Sequential(
   		#nn.Linear(d_model, d_model),
   		#nn.ReLU(),
   		nn.Linear(d_model, n_spks),
   	)

   def forward(self, mels):
   	"""
   	args:
   		mels: (batch size, length, 40)
   	return:
   		out: (batch size, n_spks)
   	"""
   	# out: (batch size, length, d_model)   length=segment_len
   	out = self.prenet(mels)
   	# out: (length, batch size, d_model)
   	out = out.permute(1, 0, 2) #交換dim=0和dim=1
   	# The encoder layer expect features in the shape of (length, batch size, d_model).
   	out = self.conformer_block(out)
   	# out: (batch size, length, d_model)
   	#out = out.transpose(0, 1)  #轉(zhuǎn)置dim=0和dim=1
   	# mean pooling
   	#stats = out.mean(dim=1) #可以理解為求平均并去除維度1  stats.size():（batch_size,d_model）
       
   	out=out.permute(1,2,0)
   	stats=self.pooling(out)
   	

   	# out: (batch, n_spks)
   	out = self.pred_layer(stats)
   	return out

total_steps=70000
李宏毅_機(jī)器學(xué)習(xí)_作業(yè)4（詳解）_HW4 Classify the speakers
total_steps=100000
文章來源地址http://www.zghlxwxcb.cn/news/detail-427294.html

到了這里，關(guān)于李宏毅_機(jī)器學(xué)習(xí)_作業(yè)4（詳解）_HW4 Classify the speakers的文章就介紹完了。如果您還想了解更多內(nèi)容，請(qǐng)?jiān)谟疑辖撬阉鱐OY模板網(wǎng)以前的文章或繼續(xù)瀏覽下面的相關(guān)文章，希望大家以后多多支持TOY模板網(wǎng)！