Merge pull request #158 from Srinidhi-Patil/master

GMAE Metric

gtime/metrics/__init__.py

@@ -2,7 +2,8 @@
 The :mod:`gtime.metrics` module contains a collection of different metrics.
 """
 
-from .metrics import smape, max_error, mse, log_mse, r_square, mae, mape, rmse, rmsle
+from .metrics import (smape, max_error, mse, log_mse, 
+                      r_square, mae, mape, rmse, rmsle, gmae)
 
 __all__ = [
     "smape",
@@ -14,4 +15,5 @@
     "r_square",
     "mae",
     "mape",
+    "gmae",
 ]

gtime/metrics/metrics.py

@@ -2,6 +2,7 @@
 
 import numpy as np
 import pandas as pd
+from scipy.stats import gmean
 
 
 def _check_input(y_true: np.ndarray, y_pred: np.ndarray) -> None:
@@ -305,7 +306,8 @@ def mae(
     y_true: Union[pd.DataFrame, List, np.ndarray],
     y_pred: Union[pd.DataFrame, List, np.ndarray],
 ) -> float:
-    """Compute the Mean Absolute Error(also called, Mean Absolute Deviation(MAD)) between two vectors.
+    """Compute the Mean Absolute Error(also called, Mean Absolute Deviation(MAD) or Mean Ratio) 
+    between two vectors.
 
     Parameters
     ----------
@@ -371,3 +373,38 @@ def mape(
     if np.isnan(mape_value):
         raise ValueError("MAPE can not be calculated due to Zero/Zero")
     return mape_value * 100
+
+
+def gmae(
+    y_true: Union[pd.DataFrame, List, np.ndarray],
+    y_pred: Union[pd.DataFrame, List, np.ndarray],
+) -> float:
+    """Compute the Geometric Mean Absolute Error between two vectors.
+
+    Parameters
+    ----------
+    y_true : array-like, shape (length, 1), required
+        The first vector.
+
+    y_pred : array-like, shape (length, 1), required
+        The second vector.
+
+    Returns
+    -------
+    gmae_value : float
+        The geometric mean absolute error between the two vectors.
+
+    Examples
+    --------
+    >>> from gtime.metrics import gmae
+    >>> y_true = [0, 1, 2, 3, 6, 5]
+    >>> y_pred = [-1, 4, 5, 10, 4, 1]
+    >>> gmae(y_true, y_pred)
+    2.82
+
+    """
+    y_true, y_pred = _convert_to_ndarray(y_true, y_pred)
+    _check_input(y_true, y_pred)
+
+    absolute_difference = np.abs(y_pred - y_true)
+    return 0 if 0 in absolute_difference else gmean(absolute_difference)

gtime/metrics/tests/test_loss_functions.py

@@ -1,6 +1,7 @@
 import numpy as np
 import pandas as pd
 import pytest
+from scipy.stats import gmean
 from hypothesis import given
 from hypothesis.extra.numpy import arrays
 from hypothesis.strategies import floats
@@ -15,6 +16,7 @@
     mape,
     rmse,
     rmsle,
+    gmae,
 )
 
 
@@ -812,3 +814,83 @@ def test_mape_random_arrays_finite_values(self, y_true, y_pred):
         print(y_pred)
 
         assert expected_mape == mape_value
+
+
+class TestGMAE:
+    def _correct_gmae(self, y_true, y_pred):
+        y_true = np.array(y_true)
+        y_pred = np.array(y_pred)
+
+        absolute_difference = np.abs(y_true - y_pred)
+        return 0 if 0 in absolute_difference else gmean(absolute_difference)
+
+    def test_wrong_vector_length(self):
+        y_true = np.random.random(5)
+        y_pred = np.random.random(4)
+
+        with pytest.raises(ValueError):
+            gmae(y_true, y_pred)
+
+    def test_nan_values(self):
+        y_true = [np.nan, 1, 2, 3]
+        y_pred = np.random.random(4)
+
+        with pytest.raises(ValueError):
+            gmae(y_true, y_pred)
+
+    def test_infinite_values(self):
+        y_true = np.random.random(4)
+        y_pred = [0, np.inf, 2, 3]
+
+        with pytest.raises(ValueError):
+            gmae(y_true, y_pred)
+
+    def test_gmae_list(self):
+        y_true = [0, 1, 2, 3, 6, 5]
+        y_pred = [-1, 4, 5, 10, 4, 1]
+
+        gmae_value = np.round(gmae(y_true, y_pred), decimals=2)
+        expected_gmae = 2.82
+
+        assert expected_gmae == gmae_value
+
+    def test_gmae_array(self):
+        y_true = np.array([0, 1, 2, 3, 6, 5])
+        y_pred = np.array([-1, 4, 5, 10, 4, 1])
+
+        gmae_value = np.round(gmae(y_true, y_pred), decimals=2)
+        expected_gmae = 2.82
+
+        assert expected_gmae == gmae_value
+
+    def test_gmae_dataframe(self):
+        y_true = pd.DataFrame([0, 1, 2, 3, 6, 5])
+        y_pred = pd.DataFrame([-1, 4, 5, 10, 4, 1])
+
+        gmae_value = np.round(gmae(y_true, y_pred), decimals=2)
+        expected_gmae = 2.82
+
+        assert expected_gmae == gmae_value
+
+    def test_zero_in_difference_gmae(self):
+        #if absolute difference is zero then GMAE is zero
+        y_true = pd.DataFrame([0, 1, 2, 3, 4, 5])
+        y_pred = pd.DataFrame([-1, 4, 5, 10, 4, 1])
+
+        gmae_value = np.round(gmae(y_true, y_pred), decimals=2)
+        expected_gmae = 0
+
+        assert expected_gmae == gmae_value
+
+    @given(
+        arrays(float, shape=30, elements=floats(allow_nan=False, allow_infinity=False)),
+        arrays(float, shape=30, elements=floats(allow_nan=False, allow_infinity=False)),
+    )
+    def test_gmae_random_arrays_finite_values(self, y_true, y_pred):
+        gmae_value = gmae(y_true, y_pred)
+        expected_gmae = self._correct_gmae(y_true, y_pred)
+        print(y_true)
+        print(y_pred)
+
+        assert expected_gmae == gmae_value
+