mirror of
https://github.com/wassname/Castor.git
synced 2026-08-20 12:00:37 +08:00
Ashutosh-Adhikari
dc086e895f
* Add ReutersTrainer, ReutersEvaluator options in Factory classes * Add Reuters to Kim-CNN command line arguments * Fix SST dataset path according to changes in Kim-CNN args The dataset path in args.py was made to point at the dataset folder rather than dataset/SST folder. Hence SST folder was added to paths in the SST dataset class * Add Reuters dataset class, and support in __main__ * Add Reuters dataset trainers and evaluators * Remove debug print statement in reuters_evaluator * Fix rounding bug in reuters_trainer and reuters_evaluator * Add LSTM for baseline text classification measurements * Add eval metrics for lstm_baseline * Set batch_first param in lstm_baseline * Remove onnx args from lstm_baseline * Pack padded sequences in LSTM_baseline * Add TensorBoardX support for Reuters trainer * Add Arxiv Academic Paper Dataset (AAPD) * Add Hidden Bottleneck Layer to BiLSTM * Fix packing of padded tensors in Reuters * Add cmdline args for Hidden Bottleneck Layer for BiLSTM * Include pre-padding lengths in AAPD dataset * Remove duplication of preprocessing code in AAPD * Remove batch_size condition in ReutersTrainer * Add ignore_lengths option to ReutersTrainer and ReutersEvaluator * Add AAPDCharQuantized and ReutersCharQuantized * Rename Reuters_hierarchical to ReutersHierarchical * Add CharacterCNN for document classification * Update README.md for CharacterCNN * Fix table in README.md for CharacterCNN * Add AAPDHierarchical for HAN * Update HAN for changes in Reuters dataset endpoints * Fix bug in CharCNN when running on CPU * Add AAPD dataset support for KimCNN * Fix dataset paths for SST-1 * Fix dimensions of FC1 in CharCNN * Add model checkpointing for Reuters based on F1 * Refactor LSTM baseline __main__ * Add precision, recall and F1 to Reuters evaluator * Checkpoint only at the end of an epoch for ReutersTrainer Add detailed log printing for dev evaluations * Fix log_template and dev_log_template in ReutersTrainer * Add IMDB dataset * Fix duplicate printing of header in ReutersTrainer * Add support for single_label datasets in ReutersTrainer * Add support for IMDB dataset in lstm_baseline and lstm_reg * Fix evaluator call in main method of HAN * Add IMDB for HAN * Fix for single_label * Fix evaluate_dataset method for single_label datasets * Reduce default patience to 5 epochs before early stopping * Revert change to save_state rather than the entire model * Add Yelp 2018 dataset * Integrate Yelp2018 with LSTM baseline * Replace Yelp2018 with Yelp2014 dataset * Add Yelp2014 to LSTM Baseline * Integrate Yelp14 into LSTM Regularization * Remove dropout in HBL for LSTM Baseline and Reg * Add Yelp for HAN * Fix the saving issue for HAN * Fix loading for HAN * Fix typo in ReutersEvaluator * Print to STDOUT rather than logger * Print XML-CNN eval to STDOUT rather than logger * Update max_length for IMDB dataset * Add single_label support for char_cnn * Fix evaluation method for char_cnn * Remove unwanted parameters from ReutersTrainer and ReutersEval * Fix code formatting in lstm_reg/args * Add support for IMDB and Yelp in KimCNN * Fix single_label incorporation * Remove unnecessary conditions * Fix num_classes in Yelp2014 * Add single_label support for XML-CNN * Fix call to evaluator in XML-CNN * Address PEP8 issues * Address PEP8 issues * Address PEP8 issues * Address PEP8 issues
66 lines
2.9 KiB
Python
66 lines
2.9 KiB
Python
import torch
|
|
import torch.nn.functional as F
|
|
import numpy as np
|
|
|
|
from sklearn import metrics
|
|
from .evaluator import Evaluator
|
|
|
|
|
|
class ReutersEvaluator(Evaluator):
|
|
|
|
def __init__(self, dataset_cls, model, embedding, data_loader, batch_size, device, keep_results=False):
|
|
super().__init__(dataset_cls, model, embedding, data_loader, batch_size, device, keep_results)
|
|
self.ignore_lengths = False
|
|
self.single_label = False
|
|
|
|
def get_scores(self):
|
|
self.model.eval()
|
|
self.data_loader.init_epoch()
|
|
n_dev_correct = 0
|
|
total_loss = 0
|
|
|
|
# Temp Ave
|
|
if hasattr(self.model, 'beta_ema') and self.model.beta_ema > 0:
|
|
old_params = self.model.get_params()
|
|
self.model.load_ema_params()
|
|
|
|
predicted_labels, target_labels = list(), list()
|
|
for batch_idx, batch in enumerate(self.data_loader):
|
|
if hasattr(self.model, 'TAR') and self.model.TAR: # TAR condition
|
|
if self.ignore_lengths:
|
|
scores, rnn_outs = self.model(batch.text)
|
|
else:
|
|
scores, rnn_outs = self.model(batch.text[0], lengths=batch.text[1])
|
|
else:
|
|
if self.ignore_lengths:
|
|
scores = self.model(batch.text)
|
|
else:
|
|
scores = self.model(batch.text[0], lengths=batch.text[1])
|
|
|
|
if self.single_label:
|
|
predicted_labels.extend(torch.argmax(scores, dim=1).cpu().detach().numpy())
|
|
target_labels.extend(torch.argmax(batch.label, dim=1).cpu().detach().numpy())
|
|
total_loss += F.cross_entropy(scores, torch.argmax(batch.label, dim=1), size_average=False).item()
|
|
else:
|
|
scores_rounded = F.sigmoid(scores).round().long()
|
|
predicted_labels.extend(scores_rounded.cpu().detach().numpy())
|
|
target_labels.extend(batch.label.cpu().detach().numpy())
|
|
total_loss += F.binary_cross_entropy_with_logits(scores, batch.label.float(), size_average=False).item()
|
|
|
|
if hasattr(self.model, 'TAR') and self.model.TAR: # TAR condition
|
|
total_loss += (rnn_outs[1:] - rnn_outs[:-1]).pow(2).mean()
|
|
|
|
predicted_labels = np.array(predicted_labels)
|
|
target_labels = np.array(target_labels)
|
|
accuracy = metrics.accuracy_score(target_labels, predicted_labels)
|
|
precision = metrics.precision_score(target_labels, predicted_labels, average='micro')
|
|
recall = metrics.recall_score(target_labels, predicted_labels, average='micro')
|
|
f1 = metrics.f1_score(target_labels, predicted_labels, average='micro')
|
|
avg_loss = total_loss / len(self.data_loader.dataset.examples)
|
|
|
|
# Temp Ave
|
|
if hasattr(self.model, 'beta_ema') and self.model.beta_ema > 0:
|
|
self.model.load_params(old_params)
|
|
|
|
return [accuracy, precision, recall, f1, avg_loss], ['accuracy', 'precision', 'recall', 'f1', 'cross_entropy_loss']
|