`genome.models.functions`¶

Model functions

genome.models.functions.errors(self, y)¶

Returns a floatX value of the error

Parameters:	y (theano.tensor.TensorVariable) – A symbolic Tensor variable of the ground-truth data
Returns:	The error rate
Return type:	floatX

genome.models.functions.hessians(self, y)¶

Return a list of hessians w.r.t. to the model parameters

Parameters:	y (theano.tensor.TensorVariable) – A symbolic Tensor variable of the ground-truth data
Returns:	A list of hessian matrices corresponding to each parameter
Return type:	list of theano.shared.TensorSharedVariable

genome.models.functions.mean_squared_error(self, y)¶

Returns a floatX value of the mean squared error (MSE)

Parameters:	y (theano.tensor.TensorVariable) – A symbolic Tensor variable of the ground-truth data
Returns:	The MSE of the linear model
Return type:	floatX

genome.models.functions.negative_log_likelihood(self, y, mean=True)¶

Returns the mean or sum of the negative log likelihood

Parameters:	y (theano.tensor.TensorVariable) – A symbolic Tensor variable of the ground-truth data
Returns:	The negative log-likelihood of the mini-batch
Return type:	floatX

Note

We use mean so the gradient is not dependent on the size of the batch. set mean=False if the optimization requires dependency on the size of the batch.

genome.models.functions¶

`genome.models.functions`¶