TODO:
- FInish checking current experiment
- Re run vectorisation with the mean (and tf vectors)
- Re run the replciaiton
- on 8k check verbosity on each items and press release
- plot time series across time
- build earning suprise and see if this relates to number of 8k and number of words in expectations.
- Raw data (csv and json) is stored on an external hard drive that attached to a VM.
- I transformed it into pickles and push that on spratan with clean/news_01_refinitive_merge.py
- Same stuff with third party news in clean/news_03_third_party_merge
- clean/news_05_select_and_merge_tickers_relevant.py select and news with some returns and a single firm.
- news_06_prep_vec_and_news_on_that_day finish the job and kinda get the same data as bryan.
TF RECORDS:
- vec_to_tf_records.py is transforming it into this nice tf format
- train_tf.py is my full training routine.
to push to gadi load_control_coverage load_bryan_data rel_max