diff --git a/Dockerfile b/Dockerfile index dd02d11..d489f50 100644 --- a/Dockerfile +++ b/Dockerfile @@ -2,7 +2,7 @@ from ubuntu RUN ["apt-get","update"] RUN ["apt-get","upgrade","-y"] RUN ["apt-get","install","-y","git", "python3-dev","tmux","locales","python3-pip","python3-numpy","python3-pandas","locales"] -RUN ["pip3","install","pandas-gbq","tensorflow"] +RUN ["pip3","install","pandas-gbq","tensorflow","git+https://hiplab.mc.vanderbilt.edu/git/aou/"] RUN ["mkdir","-p","/usr/apps"] WORKDIR /usr/apps -RUN ["git","clone","https://hiplab.mc.vanderbilt.edu/git/gan.git","aou-gan"] +RUN ["git","clone","https://hiplab.mc.vanderbilt.edu/git/aou/bridge.git@release","aou-gan"] diff --git a/README.md b/README.md index 8eb92d1..3dfb291 100644 --- a/README.md +++ b/README.md @@ -1,2 +1,63 @@ -# bridge +## Introduction +This package is designed to generate synthetic data from a dataset from an original dataset using deep learning techniques + + - Generative Adversarial Networks + - With "Earth mover's distance" + +## Installation + + pip install git+https://hiplab.mc.vanderbilt.edu/git/aou/data-maker.git@release + +## Usage + +After installing the easiest way to get started is as follows (using pandas). The process is as follows: + +**Train the GAN on the original/raw dataset** + + + import pandas as pd + import data.maker + + df = pd.read_csv('sample.csv') + column = 'gender' + id = 'id' + context = 'demo' + data.maker.train(context=context,data=df,column=column,id=id,logs='logs') + +The trainer will store the data on disk (for now) in a structured folder that will hold training models that will be used to generate the synthetic data. + + +**Generate a candidate dataset from the learned features** + + + import pandas as pd + import data.maker + + df = pd.read_csv('sample.csv') + id = 'id' + column = 'gender' + context = 'demo' + data.maker.generate(context=context,data=df,id=id,column=column,logs='logs') + +## Limitations + +GANS will generate data assuming the original data has all the value space needed: + +- No new data will be created + + Assuming we have a dataset with an gender attribute with values [M,F]. + + The synthetic data will not be able to generate genders outside [M,F] + +- Not advised on continuous values + + GANS work well on discrete values and thus are not advised to be used. + e.g:measurements (height, blood pressure, ...) +- For now will only perform on a single feature. + +## Credits : + +- [Ziqi Zhang](ziqi.zhang@vanderbilt.edu) +- [Brad Malin](b.malin@vanderbilt.edu) +- [Steve L. Nyemba](steve.l.nyemba@vanderbilt.edu) \ No newline at end of file diff --git a/WGAN.py b/data/WGAN.py similarity index 100% rename from WGAN.py rename to data/WGAN.py diff --git a/data/__init__.py b/data/__init__.py new file mode 100644 index 0000000..98124f1 --- /dev/null +++ b/data/__init__.py @@ -0,0 +1,2 @@ +import data.params as params + diff --git a/bridge.py b/data/bridge.py similarity index 100% rename from bridge.py rename to data/bridge.py diff --git a/gan.py b/data/gan.py similarity index 89% rename from gan.py rename to data/gan.py index 0981411..3391b78 100644 --- a/gan.py +++ b/data/gan.py @@ -11,9 +11,10 @@ import pandas as pd import time import os import sys -from params import SYS_ARGS -from bridge import Binary +from data.params import SYS_ARGS +from data.bridge import Binary import json +import pickle os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID" os.environ['CUDA_VISIBLE_DEVICES'] = "0" @@ -37,10 +38,8 @@ class GNet : self.layers = void() self.layers.normalize = self.normalize - self.get = void() - self.get.variables = self._variable_on_cpu - self.NUM_GPUS = 1 + self.NUM_GPUS = 1 if 'num_gpu' not in args else args['num_gpu'] self.X_SPACE_SIZE = args['real'].shape[1] if 'real' in args else 854 @@ -63,7 +62,11 @@ class GNet : self.ATTRIBUTES = {"id":args['column_id'] if 'column_id' in args else None,"synthetic":args['column'] if 'column' in args else None} self._REAL = args['real'] if 'real' in args else None self._LABEL = args['label'] if 'label' in args else None - + + self.get = void() + self.get.variables = self._variable_on_cpu + self.get.suffix = lambda : "-".join(self.ATTRIBUTES['synthetic']) if isinstance(self.ATTRIBUTES['synthetic'],list) else self.ATTRIBUTES['synthetic'] + self.logger = args['logger'] if 'logger' in args and args['logger'] else None self.init_logs(**args) def init_logs(self,**args): @@ -83,7 +86,9 @@ class GNet : This function is designed to accomodate the uses of the sub-classes outside of a strict dependency model. Because prediction and training can happen independently """ - _name = os.sep.join([self.out_dir,'meta-'+column+'.json']) + # suffix = "-".join(column) if isinstance(column,list)else column + suffix = self.get.suffix() + _name = os.sep.join([self.out_dir,'meta-'+suffix+'.json']) if os.path.exists(_name) : attr = json.loads((open(_name)).read()) for key in attr : @@ -94,7 +99,7 @@ class GNet : def log_meta(self,**args) : - object = { + _object = { 'CONTEXT':self.CONTEXT, 'ATTRIBUTES':self.ATTRIBUTES, 'BATCHSIZE_PER_GPU':self.BATCHSIZE_PER_GPU, @@ -111,9 +116,13 @@ class GNet : key = args['key'] value= args['value'] object[key] = value - _name = os.sep.join([self.out_dir,'meta-'+SYS_ARGS['column']]) + # suffix = "-".join(self.column) if isinstance(self.column,list) else self.column + suffix = self.get.suffix() + _name = os.sep.join([self.out_dir,'meta-'+suffix]) + f = open(_name+'.json','w') - f.write(json.dumps(object)) + f.write(json.dumps(_object)) + return _object def mkdir (self,path): if not os.path.exists(path) : os.mkdir(path) @@ -285,8 +294,10 @@ class Train (GNet): self.discriminator = Discriminator(**args) self._REAL = args['real'] self._LABEL= args['label'] + self.column = args['column'] # print ([" *** ",self.BATCHSIZE_PER_GPU]) - self.log_meta() + + self.meta = self.log_meta() def load_meta(self, column): """ This function will delegate the calls to load meta data to it's dependents @@ -384,7 +395,7 @@ class Train (GNet): # saver = tf.train.Saver() saver = tf.compat.v1.train.Saver() init = tf.global_variables_initializer() - + logs = [] with tf.Session(config=tf.ConfigProto(allow_soft_placement=True, log_device_placement=False)) as sess: sess.run(init) sess.run(iterator_d.initializer, @@ -406,13 +417,22 @@ class Train (GNet): format_str = 'epoch: %d, w_distance = %f (%.1f)' print(format_str % (epoch, -w_sum/(self.STEPS_PER_EPOCH*2), duration)) + # print (dir (w_distance)) + + logs.append({"epoch":epoch,"distance":-w_sum/(self.STEPS_PER_EPOCH*2) }) + if epoch % self.MAX_EPOCHS == 0: - - _name = os.sep.join([self.train_dir,self.ATTRIBUTES['synthetic']]) + # suffix = "-".join(self.ATTRIBUTES['synthetic']) if isinstance(self.ATTRIBUTES['synthetic'],list) else self.ATTRIBUTES['synthetic'] + suffix = self.get.suffix() + _name = os.sep.join([self.train_dir,suffix]) # saver.save(sess, self.train_dir, write_meta_graph=False, global_step=epoch) saver.save(sess, _name, write_meta_graph=False, global_step=epoch) # # + if self.logger : + row = {"logs":logs} #,"model":pickle.dump(sess)} + + self.logger.write(row=row) class Predict(GNet): """ @@ -420,14 +440,16 @@ class Predict(GNet): """ def __init__(self,**args): GNet.__init__(self,**args) - self.generator = Generator(**args) - self.values = values + self.generator = Generator(**args) + self.values = args['values'] def load_meta(self, column): super().load_meta(column) self.generator.load_meta(column) def apply(self,**args): # print (self.train_dir) - model_dir = os.sep.join([self.train_dir,self.ATTRIBUTES['synthetic']+'-'+str(self.MAX_EPOCHS)]) + # suffix = "-".join(self.ATTRIBUTES['synthetic']) if isinstance(self.ATTRIBUTES['synthetic'],list) else self.ATTRIBUTES['synthetic'] + suffix = self.get.suffix() + model_dir = os.sep.join([self.train_dir,suffix+'-'+str(self.MAX_EPOCHS)]) demo = self._LABEL #np.zeros([self.ROW_COUNT,self.NUM_LABELS]) #args['de"shape":{"LABEL":list(self._LABEL.shape)} mo'] tf.compat.v1.reset_default_graph() z = tf.random.normal(shape=[self.BATCHSIZE_PER_GPU, self.Z_DIM]) @@ -450,19 +472,24 @@ class Predict(GNet): # if we are dealing with numeric values only we can perform a simple marginal sum against the indexes # - df = ( pd.DataFrame(np.round(f).astype(np.int32),columns=values)) + df = ( pd.DataFrame(np.round(f).astype(np.int32))) # i = df.T.index.astype(np.int32) #-- These are numeric pseudonyms # df = (i * df).sum(axis=1) # # In case we are dealing with actual values like diagnosis codes we can perform # - r = np.zeros((self.ROW_COUNT,1)) + columns = self.ATTRIBUTES['synthetic'] if isinstance(self.ATTRIBUTES['synthetic'],list)else [self.ATTRIBUTES['synthetic']] + + r = np.zeros((self.ROW_COUNT,len(columns))) for col in df : i = np.where(df[col])[0] r[i] = col - df = pd.DataFrame(r,columns=[self.ATTRIBUTES['synthetic']]) - return df.to_dict(orient='list') + df = pd.DataFrame(r,columns=columns) + + df[df.columns] = (df.apply(lambda value: self.values[ int(value)],axis=1)) + return df.to_dict(orient='lists') + # return df.to_dict(orient='list') # count = str(len(os.listdir(self.out_dir))) # _name = os.sep.join([self.out_dir,self.CONTEXT+'-'+count+'.csv']) # df.to_csv(_name,index=False) @@ -476,7 +503,7 @@ class Predict(GNet): # idx2 = (demo[:, n] == 1) # idx = [idx1[j] and idx2[j] for j in range(len(idx1))] # num = np.sum(idx) - # print ("_____________________") + # print ("___________________list__") # print (idx1) # print (idx2) # print (idx) @@ -531,7 +558,8 @@ if __name__ == '__main__' : elif 'generate' in SYS_ARGS: values = df[column].unique().tolist() values.sort() - p = Predict(context=context,label=LABEL,values=values) + + p = Predict(context=context,label=LABEL,values=values,column=column) p.load_meta(column) r = p.apply() print (df) @@ -539,6 +567,7 @@ if __name__ == '__main__' : df[column] = r[column] print (df) + else: print (SYS_ARGS.keys()) print (__doc__) diff --git a/data/maker/__init__.py b/data/maker/__init__.py new file mode 100644 index 0000000..469a65a --- /dev/null +++ b/data/maker/__init__.py @@ -0,0 +1,75 @@ +""" +(c) 2019 Data Maker, hiplab.mc.vanderbilt.edu +version 1.0.0 + +This package serves as a proxy to the overall usage of the framework. +This package is designed to generate synthetic data from a dataset from an original dataset using deep learning techniques + +@TODO: + - Make configurable GPU, EPOCHS +""" +import pandas as pd +import numpy as np +import data.gan as gan +from transport import factory +def train (**args) : + """ + This function is intended to train the GAN in order to learn about the distribution of the features + :column columns that need to be synthesized (discrete) + :logs where the output of the (location on disk) + :id identifier of the dataset + :data data-frame to be synthesized + :context label of what we are synthesizing + """ + column = args['column'] + + column_id = args['id'] + df = args['data'] + logs = args['logs'] + real = pd.get_dummies(df[column]).astype(np.float32).values + labels = pd.get_dummies(df[column_id]).astype(np.float32).values + + max_epochs = 10 if 'max_epochs' not in args else args['max_epochs'] + context = args['context'] + if 'store' in args : + args['store']['args']['doc'] = context + logger = factory.instance(**args['store']) + + else: + logger = None + + trainer = gan.Train(context=context,max_epochs=max_epochs,real=real,label=labels,column=column,column_id=column_id,logger = logger,logs=logs) + return trainer.apply() + +def generate(**args): + """ + This function will generate a synthetic dataset on the basis of a model that has been learnt for the dataset + @return pandas.DataFrame + + :data data-frame to be synthesized + :column columns that need to be synthesized (discrete) + :id column identifying an entity + :logs location on disk where the learnt knowledge of the dataset is + """ + df = args['data'] + + column = args['column'] + column_id = args['id'] + logs = args['logs'] + context = args['context'] + # + #@TODO: + # If the identifier is not present, we should fine a way to determine or make one + # + #ocolumns= list(set(df.columns.tolist())- set(columns)) + + values = df[column].unique().tolist() + values.sort() + + labels = pd.get_dummies(df[column_id]).astype(np.float32).values + handler = gan.Predict (context=context,label=labels,values=values,column=column) + handler.load_meta(column) + r = handler.apply() + _df = df.copy() + _df[column] = r[column] + return _df diff --git a/data/maker/__main__.py b/data/maker/__main__.py new file mode 100644 index 0000000..e77bf0a --- /dev/null +++ b/data/maker/__main__.py @@ -0,0 +1,10 @@ +import pandas as pd +import data.maker + +df = pd.read_csv('sample.csv') +column = 'gender' +id = 'id' +context = 'demo' +store = {"type":"mongo.MongoWriter","args":{"host":"localhost:27017","dbname":"GAN"}} +max_epochs = 11 +data.maker.train(store=store,max_epochs=max_epochs,context=context,data=df,column=column,id=id,logs='foo') \ No newline at end of file diff --git a/params.py b/data/params.py similarity index 100% rename from params.py rename to data/params.py diff --git a/setup.py b/setup.py new file mode 100644 index 0000000..a7befdf --- /dev/null +++ b/setup.py @@ -0,0 +1,15 @@ +from setuptools import setup, find_packages +import os +import sys + +def read(fname): + return open(os.path.join(os.path.dirname(__file__), fname)).read() +args = {"name":"data-maker","version":"1.0.0","author":"Vanderbilt University Medical Center","author_email":"steve.l.nyemba@vanderbilt.edu","license":"MIT", + "packages":find_packages(),"keywords":["healthcare","data","transport","protocol"]} +args["install_requires"] = ['data-transport@git+https://dev.the-phi.com/git/steve/data-transport.git','numpy','pandas','pandas-gbq','pymongo'] +args['url'] = 'https://hiplab.mc.vanderbilt.edu/aou/gan.git' + +if sys.version_info[0] == 2 : + args['use_2to3'] = False + args['use_2to3_exclude_fixers'] = ['lib2to3.fixes.fix_import'] +setup(**args)