Classification Project

This project trains and evaluates classification models on four related datasets, automatically choosing an appropriate pipeline for high‑dimensional and low‑dimensional feature spaces.

Project Structure

• classification.py – main script:
  • loads data for each dataset,
  • cleans and preprocesses features,
  • builds and trains models,
  • writes prediction files.
• classification/ (subdirectory expected by the script):
  • TrainData1.txt … TrainData4.txt – training features.
  • TestData1.txt … TestData4.txt – test features.
  • TrainLabel1.txt … TrainLabel4.txt – training labels.
• Output files (created by the script):
  • <LastName>Classification1.txt … <LastName>Classification4.txt – predicted labels, one per line.

Data & Preprocessing

• Input format
  • All .txt files are whitespace‑separated, with no header.
  • Each TrainData*.txt / TestData*.txt row is a feature vector for one sample.
  • Each TrainLabel*.txt row is the class label for the corresponding training sample.
• Missing values
  • Very large values (greater than 1e98) are treated as missing.
  • Missing values are imputed using the median of each feature.

Model Pipelines

classification.py defines configurations for four datasets:

• Datasets 1 and 2: high_dim = True
• Datasets 3 and 4: high_dim = False

Based on high_dim, different pipelines are used:

• High‑dimensional datasets
  • SimpleImputer(strategy="median")
  • StandardScaler()
  • PCA(n_components=0.95) – reduce dimensionality while preserving ~95% variance.
  • SVC(kernel="linear", random_state=42) – linear Support Vector Classifier.
• Low‑dimensional datasets
  • SimpleImputer(strategy="median")
  • StandardScaler()
  • RandomForestClassifier(n_estimators=100, random_state=42)

For each dataset:

1. Load training and test data plus labels.
2. Clean and preprocess features.
3. Fit the appropriate pipeline.
4. Predict labels for the test set.
5. Save predictions to <LastName>Classification{dataset_id}.txt.

Requirements

• Python 3.8+

Install dependencies with pip (from this directory):

pip install -r requirements.txt