home/algorithm.md

 \[\[_TOC_\]\]
 
+
+## Data set selection 
+
+- Run PyChChoo for O3 Scattered light glitches in L1
+- Get the samples with p_belong > 0.9 for the ASL-CSOFT channel, which represent the coincident witness time
+- get the GPS times of these samples
+
 # Architecture
+ 
+Here the architecture is provided. 
+
+## To train 
+
+1. Whitened the strain and witness time series for Scattered light glitches
+2. 
+3. 
+
+
+
+### For HOFT
+- To find the target glitch time series to be subtracted 
+- use a dictionary learning 
+- train the dictionary using the whitened bandpassed (10, 50) Hz 
+- Actually, train multiple dictionaries using different glitch subsets
+- 
+
+
+The plot below is created with the trained dictionaries using the band passed. The reconstructed time series is from the bankpassed series. 
+
+![Screen_Shot_2020-11-30_at_10.37.41_PM](uploads/a5eca19c11956b2b24f462e661d9be30/Screen_Shot_2020-11-30_at_10.37.41_PM.png)
+
+The plot below is created with the trained dictionaries using the band passed. The reconstructed time series is from the full series. 
+
+![Screen_Shot_2020-11-30_at_11.00.28_PM](uploads/04066b95480a661d8f9767c4f31d29f1/Screen_Shot_2020-11-30_at_11.00.28_PM.png)
+
+The reconstructed time series is irrespective whether using the bandpassed or full time series. (Note the dictionary must be trained using the bankpassed time series.)
+
+![Screen_Shot_2020-11-30_at_10.56.12_PM](uploads/6bf89813b60dbdcd66588332e683049e/Screen_Shot_2020-11-30_at_10.56.12_PM.png)
+
+##
+
+