public override distributions.Distribution expand(long[] batch_shape, distributions.Distribution instance = null)
{
if (instance != null && !(instance is Categorical))
{
throw new ArgumentException("expand(): 'instance' must be a Categorical distribution");
}
var param_shape = new List <long>();
param_shape.AddRange(batch_shape);
param_shape.Add(num_events);
var shape = param_shape.ToArray();
var p = _probs?.expand(shape);
var l = _logits?.expand(shape);
var newDistribution = ((instance == null) ? new Categorical(p, l) : instance) as Categorical;
newDistribution.batch_shape = batch_shape;
if (newDistribution == instance)
{
newDistribution.num_events = this.num_events;
newDistribution._probs = p;
newDistribution._logits = l;
}
return(newDistribution);
}
public override distributions.Distribution expand(long[] batch_shape, distributions.Distribution instance = null)
{
if (instance != null && !(instance is Gamma))
{
throw new ArgumentException("expand(): 'instance' must be a Chi2 distribution");
}
if (instance == null)
{
instance = new Chi2(concentration);
}
return(base.expand(batch_shape, instance));
}
/// <summary>
/// Returns a new distribution instance (or populates an existing instance provided by a derived class) with batch dimensions expanded to
/// `batch_shape`. This method calls `torch.Tensor.expand()` on the distribution's parameters. As such, this does not allocate new
/// memory for the expanded distribution instance.
/// </summary>
/// <param name="batch_shape">Tthe desired expanded size.</param>
/// <param name="instance">new instance provided by subclasses that need to override `.expand`.</param>
public override distributions.Distribution expand(long[] batch_shape, distributions.Distribution instance = null)
{
if (instance != null && !(instance is Cauchy))
{
throw new ArgumentException("expand(): 'instance' must be a Cauchy distribution");
}
var newDistribution = ((instance == null) ? new Cauchy(loc.expand(batch_shape), scale.expand(batch_shape)) : instance) as Cauchy;
newDistribution.batch_shape = batch_shape;
if (newDistribution == instance)
{
newDistribution.loc = loc.expand(batch_shape);
newDistribution.scale = scale.expand(batch_shape);
}
return(newDistribution);
}
public override distributions.Distribution expand(long[] batch_shape, distributions.Distribution instance = null)
{
if (instance != null && !(instance is Poisson))
{
throw new ArgumentException("expand(): 'instance' must be a Poisson distribution");
}
var r = rate.expand(batch_shape);
var newDistribution = ((instance == null) ? new Poisson(r) : instance) as Poisson;
newDistribution.batch_shape = batch_shape;
if (newDistribution == instance)
{
newDistribution.rate = r;
}
return(newDistribution);
}
public override distributions.Distribution expand(long[] batch_shape, distributions.Distribution instance = null)
{
if (instance != null && !(instance is Geometric))
{
throw new ArgumentException("expand(): 'instance' must be a Geometric distribution");
}
var newDistribution = ((instance == null) ?
new Geometric(p: _probs?.expand(batch_shape), l: logits?.expand(batch_shape)) :
instance) as Geometric;
newDistribution.batch_shape = batch_shape;
if (newDistribution == instance)
{
newDistribution._probs = _probs?.expand(batch_shape);
newDistribution._logits = _logits?.expand(batch_shape);
}
return(newDistribution);
}
public override distributions.Distribution expand(long[] batch_shape, distributions.Distribution instance = null)
{
if (instance != null && !(instance is Uniform))
{
throw new ArgumentException("expand(): 'instance' must be a Uniform distribution");
}
var newDistribution = ((instance == null) ?
new Uniform(low.expand(batch_shape), high.expand(batch_shape)) :
instance) as Uniform;
newDistribution.batch_shape = batch_shape;
if (newDistribution == instance)
{
newDistribution.low = low.expand(batch_shape);
newDistribution.high = high.expand(batch_shape);
}
return(newDistribution);
}
public override distributions.Distribution expand(long[] batch_shape, distributions.Distribution instance = null)
{
if (instance != null && !(instance is Binomial))
{
throw new ArgumentException("expand(): 'instance' must be a Binomial distribution");
}
var newDistribution = ((instance == null) ?
new Binomial(total_count.expand(batch_shape), p: _probs?.expand(batch_shape), l: logits?.expand(batch_shape)) :
instance) as Binomial;
newDistribution.batch_shape = batch_shape;
if (newDistribution == instance)
{
newDistribution.total_count = total_count.expand(batch_shape);
newDistribution._probs = _probs?.expand(batch_shape);
newDistribution._logits = _logits?.expand(batch_shape);
}
return(newDistribution);
}
/// <summary>
/// Returns a new distribution instance (or populates an existing instance provided by a derived class) with batch dimensions expanded to
/// `batch_shape`. This method calls `torch.Tensor.expand()` on the distribution's parameters. As such, this does not allocate new
/// memory for the expanded distribution instance.
/// </summary>
/// <param name="batch_shape">Tthe desired expanded size.</param>
/// <param name="instance">new instance provided by subclasses that need to override `.expand`.</param>
public override distributions.Distribution expand(long[] batch_shape, distributions.Distribution instance = null)
{
if (instance != null && !(instance is Gamma))
{
throw new ArgumentException("expand(): 'instance' must be a Gamma distribution");
}
var c = concentration.expand(batch_shape);
var r = rate.expand(batch_shape);
var newDistribution = ((instance == null) ? new Gamma(c, r) : instance) as Gamma;
newDistribution.batch_shape = batch_shape;
if (newDistribution == instance)
{
newDistribution.concentration = c;
newDistribution.rate = r;
}
return(newDistribution);
}
/// <summary>
/// Returns a new distribution instance (or populates an existing instance provided by a derived class) with batch dimensions expanded to
/// `batch_shape`. This method calls `torch.Tensor.expand()` on the distribution's parameters. As such, this does not allocate new
/// memory for the expanded distribution instance.
/// </summary>
/// <param name="batch_shape">Tthe desired expanded size.</param>
/// <param name="instance">new instance provided by subclasses that need to override `.expand`.</param>
/// <returns></returns>
public override distributions.Distribution expand(long[] batch_shape, distributions.Distribution instance = null)
{
if (instance != null && !(instance is Bernoulli))
{
throw new ArgumentException("expand(): 'instance' must be a Bernoulli distribution");
}
var p = _probs?.expand(batch_shape);
var l = _logits?.expand(batch_shape);
var newDistribution = ((instance == null) ? new Bernoulli(p, l) : instance) as Bernoulli;
newDistribution.batch_shape = batch_shape;
if (newDistribution == instance)
{
newDistribution._probs = p;
newDistribution._logits = l;
}
return(newDistribution);
}
public override distributions.Distribution expand(long[] batch_shape, distributions.Distribution instance = null)
{
if (instance != null && !(instance is Beta))
{
throw new ArgumentException("expand(): 'instance' must be a Beta distribution");
}
var c0 = concentration0.expand(batch_shape);
var c1 = concentration1.expand(batch_shape);
var newDistribution = ((instance == null) ? new Beta(c1, c0) : instance) as Beta;
newDistribution.batch_shape = batch_shape;
if (newDistribution == instance)
{
newDistribution.dirichlet = new Dirichlet(torch.stack(new Tensor[] { c1, c0 }, -1));
newDistribution.concentration1 = c1;
newDistribution.concentration0 = c0;
}
return(newDistribution);
}
/// <summary>
/// Returns a new distribution instance (or populates an existing instance provided by a derived class) with batch dimensions expanded to
/// `batch_shape`. This method calls `torch.Tensor.expand()` on the distribution's parameters. As such, this does not allocate new
/// memory for the expanded distribution instance.
/// </summary>
/// <param name="batch_shape">Tthe desired expanded size.</param>
/// <param name="instance">new instance provided by subclasses that need to override `.expand`.</param>
public override distributions.Distribution expand(long[] batch_shape, distributions.Distribution instance = null)
{
if (instance != null && !(instance is Multinomial))
{
throw new ArgumentException("expand(): 'instance' must be a Multinomial distribution");
}
var newDistribution = ((instance == null) ?
new Multinomial(total_count, categorical.expand(batch_shape) as Categorical) :
instance) as Multinomial;
if (newDistribution == instance)
{
newDistribution.total_count = total_count;
newDistribution.categorical = categorical.expand(batch_shape) as Categorical;
newDistribution.batch_shape = newDistribution.categorical.batch_shape;
var ps = newDistribution.categorical.param_shape;
newDistribution.event_shape = new long[] { ps[ps.Length - 1] };
}
return(newDistribution);
}
/// <summary>
/// Returns a new distribution instance (or populates an existing instance provided by a derived class) with batch dimensions expanded to
/// `batch_shape`. This method calls `torch.Tensor.expand()` on the distribution's parameters. As such, this does not allocate new
/// memory for the expanded distribution instance.
/// </summary>
/// <param name="batch_shape">Tthe desired expanded size.</param>
/// <param name="instance">new instance provided by subclasses that need to override `.expand`.</param>
public override distributions.Distribution expand(long[] batch_shape, distributions.Distribution instance = null)
{
if (instance != null && !(instance is FisherSnedecor))
{
throw new ArgumentException("expand(): 'instance' must be a FisherSnedecor distribution");
}
var df1 = this.df1.expand(batch_shape);
var df2 = this.df2.expand(batch_shape);
var newDistribution = ((instance == null) ? new FisherSnedecor(df1, df2) : instance) as FisherSnedecor;
newDistribution.batch_shape = batch_shape;
if (newDistribution == instance)
{
newDistribution.df1 = df1;
newDistribution.df2 = df2;
newDistribution.gamma1 = this.gamma1.expand(batch_shape) as Gamma;
newDistribution.gamma2 = this.gamma2.expand(batch_shape) as Gamma;
}
return(newDistribution);
}
/// <summary>
/// Returns a new distribution instance (or populates an existing instance provided by a derived class) with batch dimensions expanded to
/// `batch_shape`. This method calls `torch.Tensor.expand()` on the distribution's parameters. As such, this does not allocate new
/// memory for the expanded distribution instance.
/// </summary>
/// <param name="batch_shape">Tthe desired expanded size.</param>
/// <param name="instance">new instance provided by subclasses that need to override `.expand`.</param>
/// <returns></returns>
public override distributions.Distribution expand(long[] batch_shape, distributions.Distribution instance = null)
{
if (instance != null && !(instance is Dirichlet))
{
throw new ArgumentException("expand(): 'instance' must be a Dirichlet distribution");
}
var shape = new List <long>();
shape.AddRange(batch_shape);
shape.AddRange(event_shape);
var c = concentration.expand(shape.ToArray());
var newDistribution = ((instance == null) ? new Dirichlet(c) : instance) as Dirichlet;
newDistribution.batch_shape = batch_shape;
if (newDistribution == instance)
{
newDistribution.concentration = c;
}
return(newDistribution);
}
