diff --git a/README.md b/README.md index 4d48dee..ab29cc5 100644 --- a/README.md +++ b/README.md @@ -1,34 +1,16 @@ # deid-risk -This project is intended to compute an estimated value of risk for a given database. +The code below extends a data-frame by adding it the ability to compute de-identification risk (marketer, prosecutor). +Because data-frames can connect to any database/file it will be the responsibility of the user to load the dataset into a data-frame. - 1. Pull meta data of the database and create a dataset via joins - 2. Generate the dataset with random selection of features - 3. Compute risk via SQL using group by -## Python environment +Basic examples that illustrate usage of the the framework are in the notebook folder. The example is derived from +[http://ehelthinformation.ca](http://www.ehealthinformation.ca/wp-content/uploads/2014/08/2009-De-identification-PA-whitepaper1.pdf) -The following are the dependencies needed to run the code: +Dependencies: + numpy + pandas + +Limitations: - pandas - numpy - pandas-gbq - google-cloud-bigquery - - -## Usage - -**Generate The merged dataset** - - python risk.py create --i_dataset --o_dataset --table --path --key [--file ] - - -**Compute risk (marketer, prosecutor)** - - python risk.py compute --i_dataset --table --path --key -## Limitations - - It works against bigquery for now - @TODO: - - Need to write a transport layer (database interface) - - Support for referential integrity, so one table can be selected and a dataset derived given referential integrity - - Add support for journalist risk \ No newline at end of file + - Add support for journalist risk diff --git a/notebooks/risk.ipynb b/notebooks/risk.ipynb index 1109529..299bd35 100644 --- a/notebooks/risk.ipynb +++ b/notebooks/risk.ipynb @@ -2,294 +2,209 @@ "cells": [ { "cell_type": "code", - "execution_count": 1, + "execution_count": 4, "metadata": {}, - "outputs": [ - { - "name": "stdout", - "output_type": "stream", - "text": [ - "dev-deid-600@aou-res-deid-vumc-test.iam.gserviceaccount.com df0ac049-d5b6-416f-ab3c-6321eda919d6 2018-09-25 08:18:34.829000+00:00 DONE\n" - ] - } - ], + "outputs": [], "source": [ + "\"\"\"\n", + " This notebook is intended to show how to use the risk framework:\n", + " There are two basic usages:\n", + " 1. Experiment\n", + " \n", + " Here the framework will select a number of random fields other than the patient id and compute risk for the selection.\n", + " This will repeat over a designated number of runs.\n", + " \n", + " The parameters to pass to enable this mode are id=,nun_runs=\n", + " 2. Assessment\n", + " \n", + " Here the framework assumes you are only interested in a list of quasi identifiers and will run the evaluation once for a given list of quasi identifiers.\n", + " The parameters to enable this mode are id=,quasi_id=\n", + "\"\"\"\n", + "import os\n", "import pandas as pd\n", "import numpy as np\n", - "from google.cloud import bigquery as bq\n", "\n", - "client = bq.Client.from_service_account_json('/home/steve/dev/google-cloud-sdk/accounts/vumc-test.json')\n", - "# pd.read_gbq(query=\"select * from raw.observation limit 10\",private_key='/home/steve/dev/google-cloud-sdk/accounts/vumc-test.json')\n", - "jobs = client.list_jobs()\n", - "for job in jobs :\n", - "# print dir(job)\n", - " print job.user_email,job.job_id,job.started, job.state\n", - " break" - ] - }, - { - "cell_type": "code", - "execution_count": 33, - "metadata": {}, - "outputs": [], - "source": [ - "xo = ['person_id','date_of_birth','race']\n", - "xi = ['person_id','value_as_number','value_source_value']" + "\n", + "#\n", + "#-- Loading a template file\n", + "# The example taken a de-identification white-paper\n", + "# http://www.ehealthinformation.ca/wp-content/uploads/2014/08/2009-De-identification-PA-whitepaper1.pdf\n", + "#\n", + "\n", + "import pandas as pd\n", + "import numpy as np\n", + "from io import StringIO\n", + "csv = \"\"\"\n", + "id,sex,age,profession,drug_test\n", + "1,M,37,doctor,-\n", + "2,F,28,doctor,+\n", + "3,M,37,doctor,-\n", + "4,M,28,doctor,+\n", + "5,M,28,doctor,-\n", + "6,M,37,doctor,-\n", + "\"\"\"\n", + "f = StringIO()\n", + "f.write(unicode(csv))\n", + "f.seek(0)\n", + "MY_DATAFRAME = pd.read_csv(f) " ] }, { "cell_type": "code", - "execution_count": 10, + "execution_count": 2, "metadata": {}, "outputs": [], "source": [ - "def get_tables(client,id,fields=[]):\n", - " \"\"\"\n", - " getting table lists from google\n", - " \"\"\"\n", - " r = []\n", - " ref = client.dataset(id)\n", - " tables = list(client.list_tables(ref))\n", - " for table in tables :\n", - " ref = table.reference\n", - " schema = client.get_table(ref).schema\n", - " names = [f.name for f in schema]\n", - " x = list(set(names) & set(fields))\n", - " if x :\n", - " r.append({\"name\":table.table_id,\"fields\":names})\n", - " return r\n", - " \n", - "def get_fields(**args):\n", - " \"\"\"\n", - " This function will generate a random set of fields from two tables. Tables are structured as follows \n", - " {name,fields:[],\"y\":}, with \n", - " name table name (needed to generate sql query)\n", - " fields list of field names, used in the projection\n", - " y name of the field to be joined.\n", - " @param xo candidate table in the join\n", - " @param xi candidate table in the join\n", - " @param join field by which the tables can be joined.\n", - " \"\"\"\n", - " # The set operation will remove redundancies in the field names (not sure it's a good idea)\n", - "# xo = args['xo']['fields']\n", - "# xi = args['xi']['fields']\n", - "# zi = args['xi']['name']\n", - "# return list(set([ \".\".join([args['xo']['name'],name]) for name in xo]) | set(['.'.join([args['xi']['name'],name]) for name in xi if name != args['join']]) )\n", - " xo = args['xo']\n", - " fields = [\".\".join([args['xo']['name'],name]) for name in args['xo']['fields']]\n", - " if not isinstance(args['xi'],list) :\n", - " x_ = [args['xi']]\n", - " else:\n", - " x_ = args['xi']\n", - " for xi in x_ :\n", - " fields += (['.'.join([xi['name'], name]) for name in xi['fields'] if name != args['join']])\n", - " return fields\n", - "def generate_sql(**args):\n", + "\"\"\"\n", + " Here's the pandas_risk code verbatim. \n", + " NOTE: \n", + "\"\"\"\n", + "@pd.api.extensions.register_dataframe_accessor(\"deid\")\n", + "class deid :\n", " \"\"\"\n", - " This function will generate the SQL query for the resulting join\n", + " This class is a deidentification class that will compute risk (marketer, prosecutor) given a pandas dataframe\n", " \"\"\"\n", + " def __init__(self,df):\n", + " self._df = df\n", " \n", - " xo = args['xo']\n", - " x_ = args['xi']\n", - " xo_name = \".\".join([args['prefix'],xo['name'] ]) if 'prefix' in args else xo['name']\n", - " SQL = \"SELECT :fields FROM :xo.name \".replace(\":xo.name\",xo_name)\n", - " if not isinstance(x_,list):\n", - " x_ = [x_]\n", - " f = []#[\".\".join([args['xo']['name'],args['join']] )] \n", - " INNER_JOINS = []\n", - " for xi in x_ :\n", - " xi_name = \".\".join([args['prefix'],xi['name'] ]) if 'prefix' in args else xi['name']\n", - " JOIN_SQL = \"INNER JOIN :xi.name ON \".replace(':xi.name',xi_name)\n", - " value = \".\".join([xi['name'],args['join']])\n", - " f.append(value) \n", + " def risk(self,**args):\n", + " \"\"\"\n", + " @param id name of patient field \n", + " @params num_runs number of runs (default will be 100)\n", + " @params quasi_id \tlist of quasi identifiers to be used (this will only perform a single run)\n", + " \"\"\"\n", " \n", - " ON_SQL = \"\"\n", - " tmp = []\n", - " for term in f :\n", - " ON_SQL = \":xi.name.:ofield = :xo.name.:ofield\".replace(\":xo.name\",xo['name'])\n", - " ON_SQL = ON_SQL.replace(\":xi.name.:ofield\",term).replace(\":ofield\",args['join'])\n", - " tmp.append(ON_SQL)\n", - " INNER_JOINS += [JOIN_SQL + \" AND \".join(tmp)]\n", - " return SQL + \" \".join(INNER_JOINS)\n", - "def get_final_sql(**args):\n", - " xo = args['xo']\n", - " xi = args['xi']\n", - " join=args['join']\n", - " prefix = args['prefix'] if 'prefix' in args else ''\n", - " fields = get_fields (xo=xo,xi=xi,join=join)\n", - " k = len(fields)\n", - " n = np.random.randint(2,k) #-- number of fields to select\n", - " i = np.random.randint(0,k,size=n)\n", - " fields = [name for name in fields if fields.index(name) in i]\n", - " base_sql = generate_sql(xo=xo,xi=xi,prefix)\n", - " SQL = \"\"\"\n", - " SELECT AVERAGE(count),size,n as selected_features,k as total_features\n", - " FROM(\n", - " SELECT COUNT(*) as count,count(:join) as pop,sum(:n) as N,sum(:k) as k,:fields\n", - " FROM (:sql)\n", - " GROUP BY :fields\n", - " ) \n", - " order by 1\n", - " \n", - " \"\"\".replace(\":sql\",base_sql)\n", - "# sql = \"SELECT :fields FROM :xo.name INNER JOIN :xi.name ON :xi.name.:xi.y = :xo.y \"\n", - "# fields = \",\".join(get_fields(xo=xi,xi=xi,join=xi['y']))\n", - " \n", - " \n", - "# sql = sql.replace(\":fields\",fields).replace(\":xo.name\",xo['name']).replace(\":xi.name\",xi['name'])\n", - "# sql = sql.replace(\":xi.y\",xi['y']).replace(\":xo.y\",xo['y'])\n", - "# return sql\n", - " \n", - " " - ] - }, - { - "cell_type": "code", - "execution_count": 33, - "metadata": {}, - "outputs": [], - "source": [ - "xo = {\"name\":\"person\",\"fields\":['person_id','date_of_birth','race','value_as_number']}\n", - "xi = [{\"name\":\"measurement\",\"fields\":['person_id','value_as_number','value_source_value']}] #,{\"name\":\"observation\",\"fields\":[\"person_id\",\"value_as_string\",\"observation_source_value\"]}]\n", - "# generate_sql(xo=xo,xi=xi,join=\"person_id\",prefix='raw')\n", - "fields = get_fields(xo=xo,xi=xi,join='person_id')\n", - "ofields = list(fields)\n", - "k = len(fields)\n", - "n = np.random.randint(2,k) #-- number of fields to select\n", - "i = np.random.randint(0,k,size=n)\n", - "fields = [name for name in fields if fields.index(name) in i]" - ] - }, - { - "cell_type": "code", - "execution_count": 34, - "metadata": {}, - "outputs": [ - { - "data": { - "text/plain": [ - "['person.race', 'person.value_as_number', 'measurement.value_source_value']" - ] - }, - "execution_count": 34, - "metadata": {}, - "output_type": "execute_result" - } - ], - "source": [ - "fields\n" - ] - }, - { - "cell_type": "code", - "execution_count": 55, - "metadata": {}, - "outputs": [ - { - "data": { - "text/plain": [ - "'SELECT person_id,value_as_number,measurements.value_source_value,measurements.value_as_number,value_source_value FROM person INNER JOIN measurements ON measurements.person_id = person_id '" - ] - }, - "execution_count": 55, - "metadata": {}, - "output_type": "execute_result" - } - ], - "source": [ - "xo = {\"name\":\"person\",\"fields\":['person_id','date_of_birth','race'],\"y\":\"person_id\"}\n", - "xi = {\"name\":\"measurements\",\"fields\":['person_id','value_as_number','value_source_value'],\"y\":\"person_id\"}\n", - "generate_sql(xo=xo,xi=xi)" - ] - }, - { - "cell_type": "code", - "execution_count": 59, - "metadata": {}, - "outputs": [ - { - "data": { - "text/plain": [ - "[('a', 'b'), ('a', 'c'), ('b', 'c')]" - ] - }, - "execution_count": 59, - "metadata": {}, - "output_type": "execute_result" - } - ], - "source": [ - "\"\"\"\n", - " We are designing a process that will take two tables that will generate \n", - "\"\"\"\n", - "import itertools\n", - "list(itertools.combinations(['a','b','c'],2))" + " id = args['id']\n", + " if 'quasi_id' in args :\n", + " num_runs = 1\n", + " columns = list(set(args['quasi_id'])- set(id) )\n", + " else :\n", + " num_runs = args['num_runs'] if 'num_runs' in args else 100\n", + " columns = list(set(self._df.columns) - set([id]))\n", + " r = pd.DataFrame() \n", + " k = len(columns)\n", + " for i in range(0,num_runs) :\n", + " #\n", + " # let's chose a random number of columns and compute marketer and prosecutor risk\n", + " # Once the fields are selected we run a groupby clause\n", + " #\n", + " if 'quasi_id' not in args :\n", + " n = np.random.randint(2,k) #-- number of random fields we are picking\n", + " ii = np.random.choice(k,n,replace=False)\n", + " cols = np.array(columns)[ii].tolist()\n", + " else:\n", + " cols \t= columns\n", + " n \t= len(cols)\n", + " x_ = self._df.groupby(cols).count()[id].values\n", + " r = r.append(\n", + " pd.DataFrame(\n", + " [\n", + " {\n", + " \"selected\":n,\n", + " \"marketer\": x_.size / np.float64(np.sum(x_)),\n", + " \"prosecutor\":1 / np.float64(np.min(x_))\n", + "\n", + " }\n", + " ]\n", + " )\n", + " )\n", + " g_size = x_.size\n", + " n_ids = np.float64(np.sum(x_))\n", + "\n", + " return r" ] }, { "cell_type": "code", - "execution_count": 6, + "execution_count": 7, "metadata": {}, "outputs": [ { "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
marketerprosecutorselected
00.5000001.02
00.5000001.03
00.5000001.03
00.3333331.02
00.3333330.52
\n", + "
" + ], "text/plain": [ - "array([1, 3, 0, 0])" + " marketer prosecutor selected\n", + "0 0.500000 1.0 2\n", + "0 0.500000 1.0 3\n", + "0 0.500000 1.0 3\n", + "0 0.333333 1.0 2\n", + "0 0.333333 0.5 2" ] }, - "execution_count": 6, + "execution_count": 7, "metadata": {}, "output_type": "execute_result" } ], "source": [ "#\n", - "# find every table with person id at the very least or a subset of fields\n", + "# Lets us compute risk here for a random any random selection of quasi identifiers\n", + "# We will run this experiment 5 times\n", "#\n", - "np.random.randint(0,4,size=4)" - ] - }, - { - "cell_type": "code", - "execution_count": 90, - "metadata": {}, - "outputs": [ - { - "data": { - "text/plain": [ - "['a']" - ] - }, - "execution_count": 90, - "metadata": {}, - "output_type": "execute_result" - } - ], - "source": [ - "list(set(['a','b']) & set(['a']))" - ] - }, - { - "cell_type": "code", - "execution_count": 120, - "metadata": {}, - "outputs": [], - "source": [ - "x_ = 1" - ] - }, - { - "cell_type": "code", - "execution_count": 10, - "metadata": {}, - "outputs": [], - "source": [ - "x_ = pd.DataFrame({\"group\":[1,1,1,1,1], \"size\":[2,1,1,1,1]})" + "MY_DATAFRAME.deid.risk(id='id',num_runs=5)" ] }, { "cell_type": "code", - "execution_count": 12, + "execution_count": 8, "metadata": {}, "outputs": [ { @@ -313,35 +228,37 @@ " \n", " \n", " \n", - " size\n", - " \n", - " \n", - " group\n", - " \n", + " marketer\n", + " prosecutor\n", + " selected\n", " \n", " \n", " \n", " \n", - " 1\n", - " 1.2\n", + " 0\n", + " 0.5\n", + " 1.0\n", + " 3\n", " \n", " \n", "\n", "" ], "text/plain": [ - " size\n", - "group \n", - "1 1.2" + " marketer prosecutor selected\n", + "0 0.5 1.0 3" ] }, - "execution_count": 12, + "execution_count": 8, "metadata": {}, "output_type": "execute_result" } ], "source": [ - "x_.groupby(['group']).mean()\n" + "#\n", + "# In this scenario we are just interested in sex,profession,age\n", + "#\n", + "MY_DATAFRAME.deid.risk(id='id',quasi_id=['age','sex','profession'])" ] }, {