adnn: A Python repository from jbinas

 # Some info on how to use these tools.


 ## run_exp.py

 Run measurements and/or training/testing
 
 Command line args:
  * demo - run demo on small network
  * measure - run measurements with specified parameters
  * train - run training using measured data (use together with measure or specify model parameters using load arg)
  * test - run subset of test set through circuit (use together with load)
  * rndseed=42 - specify numpy rng seed
  * dataset=data/datasets/<my-dataset>.pkz - specify which dataset to use
  * single=n - only run the nth measurement
  * load=data/measured/<my-measurement>.npy - load set of measurements to be used in training
  * params=data/devparams/<my-params>.pkl - device params to be used in simulation
  * numthreads=8 - number of threads to be used
  * N=196x49x10 - network dimensions
  * numtest=50 - number of test set samples to run

 Example use cases:

 Start automatic parameter extraction
 $ python run_exp.py measure

 Start training
 $ python run_exp.py train

 Use specific model parameters for training
 $ python run_exp.py train load=data/<your-measurement-data>.npy

 Test trained network
 $ python run_exp.py train test load=data/<your-measurement-data>.npy
 


 ## run_single.py

 Run transient simulation

 Command line args:
  * rndseed=42 - numpy random seed
  * N=196x49x10 - network dimensions
  * inp=100 - maximum input current to be applied to input nodes (in nA)
  * input1=1 - input pattern 1 from dataset
  * input2=2 - input pattern 2 from dataset
  * weights=data/weights/<my-weights>.npy - set of weights to be used in simulation
  * params=data/devparams/<my-params>.pkl - device params to be used in simulation
  * dataset=data/datasets/<my-dataset>.pkz - specify which dataset to use

 $ python run_single.py rndseed=42 N=196x49x10 inp=200 datasets=data/datasets/mnist_14x14.pkl.gz weights=data/weights/weights_clipped_regtest.npy params=data/devparams/params_mnist14_-5185923438954998828.pkl inpat1=21 inpat2=121



 ## analysis.py

 Run analysis (plot/print output)

 Command line args:
  * noplot - do not plot output (used for batch processing)
  * noprint - do not print verbose output message (used for batch processing)

 $ python analysis.py data/analysis/<my-transient-data>.out.npy



 ## cat_exp.py

 Concatenate measurement files to obtain one data file
 (single input measurements have to be available in data/cache dir)

 $ python cat_exp.py
jbinas/adnn
