Mike Hughes michaelchughes

RECIPE TO TOTALLY REMOVE BAD FILES FROM A GIT REPO WITH MULTIPLE USERS

We assume that there are two users (A and B). A is gonna delete the bad files. B is gonna receive the changes.

At the start, we need A and B to have the SAME version history.

We'll assume there are no branches other than master.

Install working tensorflow or pytorch via standard conda environment workflow.

The recommended conda-based install process works smoothly:

$ # Create a fresh environment

	import argparse
	import pandas as pd
	import os
	import subprocess
	import psutil

	global GPU_UUID_MAP
	GPU_UUID_MAP = dict()

	def get_current_cpu_mem_usage(field='rss', process=None):

	import numpy as np
	import scipy.stats

	from scipy.special import logsumexp, softmax

	np.set_printoptions(precision=3, suppress=1)

	## Settings
	## --------


	## Using notation from HDP-HMM paper in NeurIPS 2015 (Hughes, Stephenson, Sudderth)

	## Start for a single sequence of interest,
	## Compute the following as in the tutorial here:
	https://bnpy.readthedocs.io/en/latest/examples/08_mocap6/plot-03-demo=interpret_hdphmm_params_and_run_viterbi.html#sphx-glr-examples-08-mocap6-plot-03-demo-interpret-hdphmm-params-and-run-viterbi-py

	# start_prob_K : 1D array size K, sums to one
	# trans_proba_KK : 2D array, size K x K, rows sum to one
	# log_lik_seq_TK : 2D array, size T x K

	S = 10 # num samples throughout

	## FIRST, REMEMBER THE COMMON PARAMETERIZATION
	# x ~ Normal(m, sigma^2)
	# m is a real number
	# sigma >= 0 is a standard deviation

	m = 2
	sigma = 0.1

	import pandas as pd
	import numpy as np
	import time

	prng = np.random.RandomState(0)

	n_subj = 20000
	n_rows = 1000000

	# Randomly assign to one of 20000 subjects

	import numpy as np
	import pandas as pd
	import argparse

	def convert_ordered_dx_df_to_indicator_df(
	ordered_dx_df,
	list_of_possible_icd_code_values=None,
	list_of_possible_icd_code_names=None):
	''' Convert rank-ordered diagnosis codes into indicator format