Module `auton_survival.models.cph`

Deep Cox Proportional Hazards Model

Sub-modules

auton_survival.models.cph.dcph_torch
auton_survival.models.cph.dcph_utilities

Classes

class DeepCoxPH (layers=None, random_seed=0)

A Deep Cox Proportional Hazards model.

This is the main interface to a Deep Cox Proportional Hazards model. A model is instantiated with approporiate set of hyperparameters and fit on numpy arrays consisting of the features, event/censoring times and the event/censoring indicators.

For full details on Deep Cox Proportional Hazards, refer [1], [2].

References

[1] DeepSurv: personalized treatment recommender system using a Cox proportional hazards deep neural network. BMC medical research methodology (2018)

[2] A neural network model for survival data. Statistics in medicine (1995)

Parameters

k : int: The number of underlying Cox distributions.
layers : list: A list of integers consisting of the number of neurons in each hidden layer.
random_seed : int: Controls the reproducibility of called functions.

Example

>>> from auton_survival import DeepCoxPH
>>> model = DeepCoxPH()
>>> model.fit(x, t, e)

Browse git

Methods

def fit(self, x, t, e, vsize=0.15, val_data=None, iters=1, learning_rate=0.001, batch_size=100, optimizer='Adam')

This method is used to train an instance of the DSM model.

Parameters

x : np.ndarray: A numpy array of the input features, $x$ .
t : np.ndarray: A numpy array of the event/censoring times, $t$ .
e : np.ndarray: A numpy array of the event/censoring indicators, $\delta$ . $\delta = 1$ means the event took place.
vsize : float: Amount of data to set aside as the validation set.
val_data : tuple: A tuple of the validation dataset. If passed vsize is ignored.
iters : int: The maximum number of training iterations on the training dataset.
learning_rate : float: The learning rate for the Adam optimizer.
batch_size : int: learning is performed on mini-batches of input data. this parameter specifies the size of each mini-batch.
optimizer : str: The choice of the gradient based optimization method. One of 'Adam', 'RMSProp' or 'SGD'.

Browse git

def predict_risk(self, x, t=None)

Browse git

def predict_survival(self, x, t=None)

Returns the estimated survival probability at time $t$ , $\widehat{\mathbb{P}}(T > t|X)$ for some input data $x$ .

Parameters

x : np.ndarray: A numpy array of the input features, $x$ .
t : list or float: a list or float of the times at which survival probability is to be computed

Returns

np.array: numpy array of the survival probabilites at each time in t.

Browse git

class DeepRecurrentCoxPH (layers=None, hidden=None, typ='LSTM', random_seed=0)

A deep recurrent Cox PH model.

This model is based on the paper: Leveraging Deep Representations of Radiology Reports in Survival Analysis for Predicting Heart Failure Patient Mortality. NAACL (2021)

Parameters

k : int: The number of underlying Cox distributions.
layers : list: A list of integers consisting of the number of neurons in each hidden layer.
random_seed : int

Controls the reproducibility of called functions. Example

>>> from dsm.contrib import DeepRecurrentCoxPH
>>> model = DeepRecurrentCoxPH()
>>> model.fit(x, t, e)

Browse git

Methods

def fit(self, x, t, e, vsize=0.15, val_data=None, iters=1, learning_rate=0.001, batch_size=100, optimizer='Adam')

Inherited from: DeepCoxPH.fit

This method is used to train an instance of the DSM model.

Parameters

x : np.ndarray: A numpy array of the input features, $x$ .
t : np.ndarray: A numpy array of the event/censoring times, $t$ .
e : np.ndarray: A numpy array of the event/censoring indicators, $\delta$ . $\delta = 1$ means the event took place.
vsize : float: Amount of data to set aside as the validation set.
val_data : tuple: A tuple of the validation dataset. If passed vsize is ignored.
iters : int: The maximum number of training iterations on the training dataset.
learning_rate : float: The learning rate for the Adam optimizer.
batch_size : int: learning is performed on mini-batches of input data. this parameter specifies the size of each mini-batch.
optimizer : str: The choice of the gradient based optimization method. One of 'Adam', 'RMSProp' or 'SGD'.

def predict_survival(self, x, t=None)

Inherited from: DeepCoxPH.predict_survival

Returns the estimated survival probability at time $t$ , $\widehat{\mathbb{P}}(T > t|X)$ for some input data $x$ .

Parameters

x : np.ndarray: A numpy array of the input features, $x$ .
t : list or float: a list or float of the times at which survival probability is to be computed

Returns

np.array: numpy array of the survival probabilites at each time in t.