Missing Token

id,label
0,0
1,0
…

id,label
0,3
1,0
2,7
…

!pip install gdown

import os, sys, numpy as np, pandas as pd, random
import gdown
from pathlib import Path

np.random.seed(42)
random.seed(42)

train_filename = "train.csv"
test_filename = "test.csv"
sample_filename = "sample_submission.csv"

gdown.download(f"https://drive.google.com/uc?id=1mUXixgltNumMyiATk1unV6yTUJetk2NA", train_filename, quiet=False)
gdown.download(f"https://drive.google.com/uc?id=1KkKisDZKBA2aQXffopzprAqWaCyrqjAo", test_filename, quiet=False)
gdown.download(f"https://drive.google.com/uc?id=1PIBWDOUFE3NtVMyViclj93-8_ChJIrqE", sample_filename, quiet=False)

# alternative links
# gdown.download(f"https://drive.google.com/uc?id=1GMEqfXZL1jSBv7xjFW0ZjsYCrv2rsGEl", train_filename, quiet=False)
# gdown.download(f"https://drive.google.com/uc?id=1NQbEVCEXVk4bDRhU-j9Errhv2ZWA3FBs", test_filename, quiet=False)
# gdown.download(f"https://drive.google.com/uc?id=1ig5p_Dxn7g6sn1elW1WPuYFJOmhM2LMA", sample_filename, quiet=False)

train_df = pd.read_csv(train_filename)
test_df = pd.read_csv(test_filename)
sample = pd.read_csv(sample_filename)

print("\ntrain:", train_df.shape, "| test:", test_df.shape)
train_df.head()

PAD = "<PAD>"
SEQ_COLS = [f"x{i}" for i in range(20)]

def clean_seq(row):
    return [row[c] for c in SEQ_COLS if row[c] != PAD]

# Basic stats
lengths = train_df.apply(lambda r: len(clean_seq(r)), axis=1)
print("Sequence length (incomplete) — min:", lengths.min(),
      "max:", lengths.max(), "mean:", round(lengths.mean(), 1))

def predict_middle(row):
    L = len(clean_seq(row))
    return L // 2

from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_absolute_error # <-- Changed to MAE

# Split train -> train_local / val_local
train_local, val_local = train_test_split(
    train_df, test_size=0.2, random_state=42, shuffle=True
)
train_local = train_local.reset_index(drop=True)
val_local   = val_local.reset_index(drop=True)

print(f"train_local: {len(train_local)} rows  |  val_local: {len(val_local)} rows")

def evaluate(predict_fn, df=None):
    """
    Run predict_fn on val_local (or on the df you pass in) and print the MAE.

    predict_fn : callable(row) -> int
        A function that takes one DataFrame row and returns the predicted position.
    df : DataFrame, optional
        Default: val_local. Must have a `label` column.
    """
    if df is None:
        df = val_local
    preds = df.apply(predict_fn, axis=1).values
    y     = df["label"].values

    mae = mean_absolute_error(y, preds)
    print(f"MAE: {mae:.4f} on {len(df)} samples")
    return mae

# Quick sanity check: middle baseline on val_local
print("\n--- Sanity check: middle baseline ---")
_ = evaluate(predict_middle)

# Generate predictions for test set
test_preds = test_df.apply(predict_middle, axis=1)

# Create submission DataFrame
submission = pd.DataFrame({
    "id": np.arange(len(test_df)),
    "label": test_preds
})

# Save to CSV
submission.to_csv("baseline_submission.csv", index=False)

print("Saved baseline_submission.csv")
print(submission.head())