Better support for large data

[jiangyi15]

Use tf.data.Dataset for large data in the option

data:
    lazy_call: True

[codecov]

Codecov Report

Merging #88 (cc8ba51) into dev (f3e7662) will increase coverage by 0.21%.
The diff coverage is 76.64%.

@@            Coverage Diff             @@
##              dev      #88      +/-   ##
==========================================
+ Coverage   73.83%   74.05%   +0.21%     
==========================================
  Files         104      104              
  Lines       14640    14768     +128     
  Branches     2717     2746      +29     
==========================================
+ Hits        10810    10937     +127     
+ Misses       3177     3163      -14     
- Partials      653      668      +15     

Flag	Coverage Δ
unittests	74.05% <76.64%> (+0.21%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files Changed	Coverage Δ
tf_pwa/data_trans/helicity_angle.py	93.65% <ø> (ø)
tf_pwa/cal_angle.py	76.38% <20.00%> (+1.38%)	⬆️
tf_pwa/config_loader/multi_config.py	62.16% <33.33%> (-1.11%)	⬇️
tf_pwa/fit.py	22.58% <50.00%> (+0.35%)	⬆️
tf_pwa/config_loader/plot.py	69.10% <64.51%> (-0.23%)	⬇️
tf_pwa/amp/core.py	74.43% <66.66%> (-0.27%)	⬇️
tf_pwa/config_loader/data.py	61.66% <70.83%> (+1.12%)	⬆️
tf_pwa/utils.py	66.18% <77.77%> (+0.52%)	⬆️
tf_pwa/model/model.py	50.44% <81.25%> (+0.82%)	⬆️
tf_pwa/data.py	75.88% <83.11%> (+0.04%)	⬆️
... and 4 more

... and 1 file with indirect coverage changes

[jiangyi15]

Currently, the best option (26) for large dataset is

data:   
    lazy_call: True
    use_tf_function: True
    no_id_cached: True
    jit_compile: True
    cached_lazy_call: cached_data/

batch=100000
Step time (nll_grad()) for 1M data + 10M PHSP MC about 10s.