text-classify

an exercise

• The dataset is too large to upload. Please download on your own and put it in main.

1.use TF-IDF to process the text.

train

run

python train.py --train_file YOUR_TRAIN_FILE_PATH --val_file YOUR_VAL_FILE_PATH --num_epochs NUM_EPOCHS --batch_size BATCH_sIZE --lr LEARNING_RATE

for example:

python train.py --train_file ./train.tsv --val_file ./test.tsv --num_epochs 50 --batch_size 128 --lr 0.001

predict

run

python predict.py --file_path YOUR_PREDICT_FILE_PATH --model_path YOUR_MODEL_PATH --output_dir OUTPUT_DIRCTORY_PATH

for example:

python predict.py --file_path ./small_test.tsv --model_path D:\CC\
数据集\checkpoint\model_epoch50_ValLoss1.82574892.pth --output_dir ./predict

after predict, confusion_matrix will be saved.

visualize

run

tensorboard --logdir=runs/

After running ,if you see information like this 'TensorBoard 2.16.2 at http://localhost:6006/ (Press CTRL+C to quit)', then you success. Open this link to get results.
Here, you will see the loss and accuracy curve of train dataset and validation dataset, and also the confusion matrix of prediction dataset(after you run 'predict.py').

result

best Train Acc:0.99994, Val Acc:0.96080 model saved to model_epoch35_ValLoss0.18576954506317656.pth

Another two versions.

2. use word2vec model to classify.

run w2vec_model.py

Train acc:0.9945, Test acc:0.9329 confusion matrix:

[[991   0   1   0   1   2   0   2   3   0]
 [  0 975   3   0   4   4   0   7   7   0]
 [  0   3 808  59  12  46  15   4  28  25] 
 [  2   6  10 828   6  12  38   1  12  85] 
 [  1   4   3   3 885   2  35   7  50  10] 
 [  0   8  21   0   5 962   0   2   1   1]
 [  0   2   0  19   6   0 942   1  19  11] 
 [  0   0   3   1   8  11   1 962  12   2] 
 [  0   0   4   0   2   3   1   3 987   0]
 [  0   0   0   4   1   0   6   0   0  989]] 

use network with convolution: Train acc:0.9895, Test acc:0.9337. confusion matrix:

[[990   0   1   0   2   2   1   3   1   0]
 [  0 968   2   1   5   9   4   7   2   2]
 [  0   4 743 133   7  54  20   8   9  22]
 [  0   5  11 890   8  12  30   3   4  37]
 [  1   3   5   2 905   3  17  16  36  12]
 [  1   5   7   1   8 976   0   0   1   1]
 [  0   3   2  38   5   0 935   1   9   7]
 [  0   1   2   0   5   7   0 978   6   1]
 [  0   0   4   1   3   2   0   9 980   1]
 [  0   0   0  21   0   0   7   0   0 972]]