sensorutils.metrics

Note

可能ならsklearn.metricsを使用したほうが良い

---

mae(true, pred, axis=None)

Calc Mean Absolute Error.

\[ \frac{1}{N}\sum_{i=0}^{N} |\hat{y}_i - y_i| \]

Parameters:

Name	Type	Description	Default
true	ndarray	true data.	required
pred	ndarray	predicted data.	required

Examples:

>>> a = np.random.randn(2, 3, 4)
>>> b = np.random.randn(2, 3, 4)
>>> mae(a, b)
>>> mae(a, b, axis=2)

Returns:

Type	Description
Union[float, numpy.ndarray]	MAE

---

mape(true, pred, axis=None)

Calc Mean Absolute Persentage Error.

\[ \frac{100}{N}\sum_{i=0}^{N} \left| \frac{\hat{y}_i - y_i}{y_i} \right| \]

Parameters:

Name	Type	Description	Default
true	ndarray	true data.	required
pred	ndarray	predicted data.	required

Examples:

>>> a = np.random.randn(2, 3, 4)
>>> b = np.random.randn(2, 3, 4)
>>> mape(a, b)
>>> mape(a, b, axis=2)

Returns:

Type	Description
Union[float, numpy.ndarray]	MAPE

---

mse(true, pred, axis=None)

Calc Mean Squared Error.

\[ \frac{1}{N}\sum_{i=0}^{N} (dst_i - src_i)^2 \]

Parameters:

Name	Type	Description	Default
true	ndarray	true data.	required
pred	ndarray	predicted data.	required

Examples:

>>> a = np.random.randn(2, 3, 4)
>>> b = np.random.randn(2, 3, 4)
>>> mse(a, b)
>>> mse(a, b, axis=2)

Returns:

Type	Description
Union[float, numpy.ndarray]	MSE

---

rmse(true, pred, axis=None)

Calc Root Mean Squared Error.

\[ \left(\frac{1}{N}\sum_{i=0}^{N} (\hat{y}_i - y_i)^2 \right)^{\frac{1}{2}} \]

Parameters:

Name	Type	Description	Default
true	ndarray	true data.	required
pred	ndarray	predicted data.	required

Examples:

>>> a = np.random.randn(2, 3, 4)
>>> b = np.random.randn(2, 3, 4)
>>> rmse(a, b)
>>> rmse(a, b, axis=2)

Returns:

Type	Description
Union[float, numpy.ndarray]	RMSE

---

rmspe(true, pred, axis=None)

Calc Root Mean Squared Persentage Error.

\[ 100 \left(\frac{1}{N}\sum_{i=0}^{N} (\frac{\hat{y}_i - y_i}{y_i})^2 \right)^{\frac{1}{2}} \]

Parameters:

Name	Type	Description	Default
true	ndarray	true data.	required
pred	ndarray	predicted data.	required

Examples:

>>> a = np.random.randn(2, 3, 4)
>>> b = np.random.randn(2, 3, 4)
>>> rmspe(a, b)
>>> rmspe(a, b, axis=2)

Returns:

Type	Description
Union[float, numpy.ndarray]	RMSPE

---

rmsle(true, pred, axis=None)

Calc root mean squared logarithmic error.

\[ \left(\frac{1}{N}\sum_{i=0}^{N} (\log (\hat{y}_i + 1) - \log (y_i + 1))^2 \right)^{\frac{1}{2}} \]

Parameters:

Name	Type	Description	Default
true	ndarray	clean data	required
pred	ndarray	with noise	required
axis	Optional[int]	mean axis	None

Examples:

>>> a = np.random.randn(2, 3, 4)
>>> b = np.random.randn(2, 3, 4)
>>> rmsle(a, b)
>>> rmsle(a, b, axis=2)

Returns:

Type	Description
Union[float, numpy.ndarray]	RMSLE

---

r2(true, pred)

Calc r2 score(coefficient of determination).

\[ {R^{2}}( \hat{y} ) := 1 - \frac{ \frac{1}{N} \sum_{i=1}^{N} { ( {y}_i - \hat{y}_{i} ) }^{2} }{ \frac{1}{N} \sum_{i=1}^{N} { ( {y}_i - \bar{y}) }^{2} } = 1 - \frac{M S E(\hat{y})}{Var(y)} \]

Parameters:

Name	Type	Description	Default
true	ndarray	clean data	required
pred	ndarray	with noise	required

Examples:

>>> a = np.random.randn(2, 3, 4)
>>> b = np.random.randn(2, 3, 4)
>>> r2(a, b)

Returns:

Type	Description
float	coefficient of determination

---

snr(true, pred, axis=None)

Calc Signal to Noise Ratio.

\[ 10 \log_{10} \left(\frac{\sum_{i=0}^{N}true_i^2}{\sum_{i=0}^{N}(true_i - pred_i)^2} \right) \]

Parameters:

Name	Type	Description	Default
true	ndarray	clean data	required
pred	ndarray	with noise	required
axis	Optional[int]	mean axis (default=None)	None

Examples:

>>> a = np.random.randn(2, 3, 4)
>>> b = np.random.randn(2, 3, 4)
>>> snr(a, b)
>>> snr(a, b, axis=2)

Returns:

Type	Description
Union[float, numpy.ndarray]	SNR

---

lsd(true_spec, pred_spec, axis=None)

Calc Log Spectral Distance.

\[ \mathrm{LSD}(S(\omega),\tilde{S}(\omega)) = \sqrt{\frac{1}{W}\sum_{\omega}^{W} \left(20\log_{10}\left|\frac{S(\omega)}{\tilde{S}(\omega)}\right|\right)^2} \]

\(S(\omega)\)と\(\tilde{S}(\omega)\) は，それぞれ原波形と雑音抑圧波形の対数スペクトル．

複数の短時間スペクトルの距離は各スペクトルで距離を算出した後，平均を取ること．

Parameters:

Name	Type	Description	Default
true_spec	ndarray	spectral 1	required
pred_spec	ndarray	spectral 2	required
axis	Optional[int]	Not use	None

Examples:

>>> a = np.random.randn(2, 3, 4)
>>> b = np.random.randn(2, 3, 4)
>>> lsd(a, b).shape

Returns:

Type	Description
Union[float, numpy.ndarray]	LSD