update dockerfile for model training

Dockerfile

@@ -1,4 +1,4 @@
-FROM python:3.11
+FROM condaforge/mambaforge:latest
 
 # seting env variables
 ENV KAGGLE_USERNAME=filippatyk
@@ -9,15 +9,16 @@ ENV RUN_TYPE=""
 WORKDIR /app
 ENV KAGGLE_CONFIG_DIR="./" 
 
-# install python dependencies
-COPY requirements.txt ./
-RUN pip install --no-cache-dir -r requirements.txt
+# create mamba env and activate it
+COPY environment.yml /tmp/environment.yml
+RUN mamba env create -f /tmp/environment.yml && mamba clean -ya
+RUN echo "mamba activate ium" >> ~/.bashrc
+ENV PATH /opt/conda/envs/ium/bin:$PATH
 
-COPY dataset.py ./
+COPY src ./src
 
 #make dir for data
-RUN mkdir -p ./data
+RUN mkdir -p ./data; mkdir -p ./results
 
 
-
-CMD kaggle datasets download -p data --unzip clmentbisaillon/fake-and-real-news-dataset && python ./dataset.py "--$RUN_TYPE"
+CMD kaggle datasets download -p data --unzip clmentbisaillon/fake-and-real-news-dataset && python ./src/main.py "--$RUN_TYPE"

environment.yml

@@ -0,0 +1,18 @@
+name: ium
+channels:
+  - conda-forge
+  - pytorch
+  - nvidia
+dependencies:
+  - python=3.10
+  - kaggle
+  - ca-certificates
+  - openssl
+  - pandas
+  - certifi
+  - scikit-learn
+  - pytorch
+  - pytorch-cuda=11.8
+  - transformers
+
+

src/datasets/dataset.py

@@ -8,27 +8,17 @@ tokenizer = BertTokenizer.from_pretrained("bert-base-cased")
 class Dataset(torch.utils.data.Dataset):
     def __init__(self, data: pd.DataFrame) -> None:
         self.labels = data["label"].to_list()
-        # self.texts = [
-        #     tokenizer(
-        #         text,
-        #         padding="max_length",
-        #         max_length=512,
-        #         truncation=True,
-        #         return_tensors="pt",
-        #     )
-        #     for text in data["text"]
-        # ]
         self.texts = data["text"].to_list()
 
     def __getitem__(self, idx):
         label = self.labels[idx]
         text = tokenizer(
-                self.texts[idx],
-                padding="max_length",
-                max_length=512,
-                truncation=True,
-                return_tensors="pt",
-            )
+            self.texts[idx],
+            padding="max_length",
+            max_length=512,
+            truncation=True,
+            return_tensors="pt",
+        )
 
         return text, label
 

src/datasets/news_dataset.py

@@ -5,7 +5,7 @@ from pathlib import Path
 
 class NewsDataset:
     def __init__(self, data_dir_path: str = "data", data_lenght: int = None) -> None:
-        self.data_dir_path = Path(data_dir_path)
+        self.data_dir_path = Path("./" + data_dir_path)
         self.true_news_path = self.data_dir_path / "True.csv"
         self.fake_news_path = self.data_dir_path / "Fake.csv"
 

src/main.py

@@ -1,7 +1,7 @@
 import random
 from sklearn.model_selection import train_test_split
 import argparse
-
+import torch
 from models import BertClassifier, utils
 from datasets import NewsDataset
 from train import train
@@ -12,7 +12,7 @@ SEED = 2137
 # Hyperparameters
 
 INITIAL_LR = 1e-6
-NUM_EPOCHS = 3
+NUM_EPOCHS = 2
 BATCH_SIZE = 2
 
 
@@ -31,8 +31,9 @@ parser.add_argument("--results_path", type=str, default="results/results.csv")
 if __name__ == "__main__":
     args = parser.parse_args()
 
+    print("CUDA: ", torch.cuda.is_available())
     # loading & spliting data
-    news_dataset = NewsDataset(data_dir_path="data", data_lenght=2000)
+    news_dataset = NewsDataset(data_dir_path="data", data_lenght=1000)
 
     train_val_data, test_data = train_test_split(
         news_dataset.data,