First solution: RMSE 41960 with sklearn LinearRegression, train on

"year" column

linreg.py

@@ -0,0 +1,56 @@
+import numpy as np
+import pandas as pd
+import sys
+from sklearn .linear_model import LinearRegression
+
+TRAIN_FILE_PATH = 'train/train.tsv'
+
+DEV0_IN = 'dev-0/in.tsv'
+DEV0_OUT = 'dev-0/out.tsv'
+
+TEST_A_IN = 'test-A/in.tsv'
+TEST_A_OUT = 'test-A/out.tsv'
+
+
+def read_data_file(filepath, x_index, y_index):
+    df = pd.read_csv(filepath, sep='\t', header=None, index_col=None)
+    x = df[x_index].tolist() if x_index is not None else None
+    y = df[y_index].tolist() if y_index is not None else None
+
+    return {'x': x, 'y': y}
+
+
+def to_numpy_2d(lst):
+    return np.array(lst).reshape(-1, 1)
+
+
+def get_trained_linreg_model(train_data):
+    x = to_numpy_2d(train_data.get('x'))
+    y = to_numpy_2d(train_data.get('y'))
+
+    model = LinearRegression()
+    model.fit(x, y)
+
+    return model
+
+
+def make_predictions(model, in_file, out_file):
+    input = read_data_file(in_file, 1, None)
+    input_x = to_numpy_2d(input.get('x'))
+    pred_y = model.predict(input_x)
+
+    with open(out_file, 'w') as f:
+        for pred in pred_y:
+            f.write(str(pred[0]) + '\n')
+
+
+def main():
+    train_data = read_data_file(TRAIN_FILE_PATH, 2, 0)
+    model = get_trained_linreg_model(train_data)
+
+    make_predictions(model, DEV0_IN, DEV0_OUT)
+    make_predictions(model, TEST_A_IN, TEST_A_OUT)
+
+
+if __name__ == '__main__':
+    main()