Examples of evaluation method

Data evaluation

Firstly, activating waveform environment.

$ conda activate waveform
$ cd /workspace/GWAI/docs/source

Then, loading demo data of waveforms: w1.npy, w2.npy, noise.npy, and plotting waveforms and noise.

>>> import matplotlib.pyplot as plt
>>> w1 = np.load('docs/source/w1.npy')
>>> w2 = np.load('docs/source/w2.npy')
>>> noise = np.load('docs/source/noise.npy')
>>> print('w1 shape: ', w1.shape)
w1 shape:  (400,)

import matplotlib.pyplot as plt
plt.plot(w1, label='waveform 1')
plt.plot(w2, label='waveform 2')
plt.plot(noise, label='noise')
plt.legend()
# plt.savefig(f'demo.png', dpi=100)
plt.show()

Source code, demo.png

Tukey Filter

import matplotlib.pyplot as plt
tmp = tukey(w1.shape[0], alpha=0.8)
plt.plot(tmp*w1, label= 'after filtering')
plt.plot(w1, label='original waveform', alpha=0.8)
plt.legend()
# plt.savefig(f'tukey.png', dpi=100)
plt.show()

Source code, tukey.png

SNR

>>> snr = get_amp_snr(w1, noise)
>>> print('snr: ', snr)
snr:  25.198280479785378

Model evaluation

MSE Loss

>>> pred = torch.tensor(w1.reshape(1,-1))
>>> label = torch.tensor(w2.reshape(1,-1))
>>> mse = mse_loss(pred, label)
>>> print('mse loss: ', mse)
mse loss:  tensor(44.5499, dtype=torch.float64)

Downstream task evaluation

Corner plot

corner_plot(w1)

corner1.png