import os
import tqdm
import time
import logging
import numpy as np
from collections import defaultdict
import torch
import torch.nn as nn
[docs]
class GloVeModel(nn.Module):
"""
GloVe Model
:param config_dict: Config Params Dictionary
:type config_dict: dict
"""
def __init__(self, config_dict):
super(GloVeModel, self).__init__()
self.embed_dim = config_dict["model"]["embed_dim"]
self.num_vocab = 1 + config_dict["dataset"]["num_vocab"]
self.cxt_embedding = nn.Embedding(self.num_vocab, self.embed_dim)
self.ctr_embedding = nn.Embedding(self.num_vocab, self.embed_dim)
self.cxt_bias_embedding = nn.Embedding(self.num_vocab, 1)
self.ctr_bias_embedding = nn.Embedding(self.num_vocab, 1)
[docs]
def forward(self, ctr, cxt):
"""
Forward propogation
:param ctr: Center tokens
:type ctr: torch.Tensor (batch_size,)
:param cxt: Context tokens
:type cxt: torch.Tensor (batch_size,)
:return: Center, Context Embeddings and Biases
:rtype: tuple (torch.Tensor [batch_size, embed_dim], torch.Tensor [batch_size, embed_dim],torch.Tensor [batch_size, 1],torch.Tensor [batch_size, 1],)
"""
ctr = ctr.to(dtype=torch.long)
cxt = cxt.to(dtype=torch.long)
ctr_embed = self.ctr_embedding(ctr)
cxt_embed = self.cxt_embedding(cxt)
ctr_bias = self.ctr_bias_embedding(ctr)
cxt_bias = self.cxt_bias_embedding(cxt)
return ctr_embed, cxt_embed, ctr_bias, cxt_bias
[docs]
class GloVeTrainer(nn.Module):
"""
GloVe Trainer
:param model: Seq2Seq model
:type model: torch.nn.Module
:param optimizer: Optimizer
:type optimizer: torch.optim
:param config_dict: Config Params Dictionary
:type config_dict: dict
"""
def __init__(self, model, optimizer, config_dict):
super(GloVeTrainer, self).__init__()
self.logger = logging.getLogger(__name__)
self.model = model
self.optim = optimizer
self.config_dict = config_dict
self.x_max = config_dict["train"]["x_max"]
self.alpha = config_dict["train"]["alpha"]
[docs]
def train_one_epoch(self, data_loader, epoch):
"""
Train step
:param data_loader: Train Data Loader
:type data_loader: torch.utils.data.Dataloader
:param epoch: Epoch number
:type epoch: int
:return: Train Loss
:rtype: torch.float32
"""
self.model.train()
total_loss, num_instances = 0, 0
self.logger.info(
f"-----------Epoch {epoch}/{self.config_dict['train']['epochs']}-----------"
)
pbar = tqdm.tqdm(
enumerate(data_loader), total=len(data_loader), desc="Training"
)
for batch_id, (ctr, cxt, cnt) in pbar:
ctr_embed, cxt_embed, ctr_bias, cxt_bias = self.model(ctr, cxt)
loss = self.loss_fn(ctr_embed, cxt_embed, ctr_bias, cxt_bias, cnt)
loss.backward()
self.optim.step()
self.optim.zero_grad()
total_loss += loss
num_instances += cnt.size(0)
train_loss = total_loss / num_instances
return train_loss
[docs]
@torch.no_grad()
def val_one_epoch(self, data_loader):
"""
Validation step
:param data_loader: Validation Data Loader
:type data_loader: torch.utils.data.Dataloader
:return: Validation Loss
:rtype: torch.float32
"""
self.model.eval()
total_loss, num_instances = 0, 0
pbar = tqdm.tqdm(
enumerate(data_loader), total=len(data_loader), desc="Validation"
)
for batch_id, (ctr, cxt, cnt) in pbar:
ctr_embed, cxt_embed, ctr_bias, cxt_bias = self.model(ctr, cxt)
loss = self.loss_fn(ctr_embed, cxt_embed, ctr_bias, cxt_bias, cnt)
total_loss += loss
num_instances += cnt.size(0)
val_loss = total_loss / num_instances
return val_loss
[docs]
def fit(self, train_loader, val_loader):
"""
Fits the model on dataset. Runs training and Validation steps for given epochs and saves best model based on the evaluation metric
:param train_loader: Train Data loader
:type train_loader: torch.utils.data.DataLoader
:param val_loader: Validaion Data Loader
:type val_loader: torch.utils.data.DataLoader
:return: Training History
:rtype: dict
"""
num_epochs = self.config_dict["train"]["epochs"]
output_folder = self.config_dict["paths"]["output_folder"]
best_val_loss = np.inf
history = defaultdict(list)
start = time.time()
for epoch in range(1, num_epochs + 1):
train_loss = self.train_one_epoch(train_loader, epoch)
val_loss = self.val_one_epoch(val_loader)
self.logger.info(f"Train Loss : {train_loss} - Val Loss : {val_loss}")
history["train_loss"].append(float(train_loss.detach().numpy()))
history["val_loss"].append(float(val_loss.detach().numpy()))
if val_loss <= best_val_loss:
self.logger.info(
f"Validation Loss improved from {best_val_loss} to {val_loss}"
)
best_val_loss = val_loss
torch.save(
self.model.state_dict(),
os.path.join(output_folder, "best_model.pt"),
)
else:
self.logger.info(f"Validation loss didn't improve from {best_val_loss}")
end = time.time()
time_taken = end - start
self.logger.info(
"Training completed in {:.0f}h {:.0f}m {:.0f}s".format(
time_taken // 3600, (time_taken % 3600) // 60, (time_taken % 3600) % 60
)
)
self.logger.info(f"Best Val RMSE: {best_val_loss}")
return history
[docs]
def loss_fn(self, ctr_embed, cxt_embed, ctr_bias, cxt_bias, count):
"""
GloVe loss
:param ctr_embed: Center embedding
:type ctr_embed: torch.Tensor (batch_size, embed_dim)
:param cxt_embed: Context embedding
:type cxt_embed: torch.Tensor (batch_size, embed_dim)
:param ctr_bias: Center Bias
:type ctr_bias: torch.Tensor (batch_size, 1)
:param cxt_bias: Context Bias
:type cxt_bias: torch.Tensor (batch_size, 1)
:param count: Cooccurence matrix element for (center, context)
:type count: float
:return: Loss
:rtype: torch.float32
"""
factor = torch.pow(count / self.x_max, self.alpha)
factor[factor > 1] = 1
log_count = torch.log(1 + count)
embed_product = torch.sum(ctr_embed * cxt_embed, dim=1)
loss = factor * torch.pow(embed_product + ctr_bias + cxt_bias - log_count, 2)
return torch.sum(loss)