Fix for issue #2

* fix paths to models; fix imports

* update readme

Co-authored-by: inkpad <[email protected]>

Author: ink-pad

README.md

@@ -33,6 +33,11 @@ You would need git-lfs to access the data.
 
 ---
 
+### PRSA Dataset
+For PRSA dataset, one have to download the PRSA dataset from [Kaggle](https://www.kaggle.com/sid321axn/beijing-multisite-airquality-data-set) and place them in [./data/card](/data/card/) directory.
+
+---
+
 ### Tabular BERT
 To train a tabular BERT model on credit card transaction or PRSA dataset run :
 ```

args.py

@@ -28,10 +28,10 @@ def define_main_parser(parser=None):
                         default="card", choices=['card', 'prsa'],
                         help='root directory for files')
     parser.add_argument("--data_root", type=str,
-                        default="./data/",
+                        default="./data/credit_card/",
                         help='root directory for files')
     parser.add_argument("--data_fname", type=str,
-                        default="sd190_trans",
+                        default="card_transaction.v1",
                         help='file name of transaction')
     parser.add_argument("--data_extension", type=str,
                         default="",

main.py

@@ -11,7 +11,7 @@
 from dataset.prsa import PRSADataset
 from dataset.card import TransactionDataset
 from models.modules import TabFormerBertLM, TabFormerGPT2
-from dataset.dataset import random_split_dataset
+from misc.utils import random_split_dataset
 from dataset.datacollator import TransDataCollatorForLanguageModeling
 
 

models/tabformer_bert.py

@@ -3,6 +3,7 @@
 from torch.nn import CrossEntropyLoss
 
 from transformers.modeling_bert import ACT2FN, BertLayerNorm
+from transformers.modeling_bert import BertForMaskedLM
 from transformers.configuration_bert import BertConfig
 from models.custom_criterion import CustomAdaptiveLogSoftmax
 

requirements.txt

@@ -1,5 +1,5 @@
 torch
 torchvision
-transformers
+transformers==3.2.0
 scikit-learn
 pandas