Ptd Gamma

The Gamma distribution is the workhorse of Bayesian stats - it models wait times, rainfall, insurance claims, and most usefully, it's the conjugate prior for the rate of a Poisson. PtdGamma is how you create one inside ComfyUI without leaving the graph. You give it a shape and a rate, it hands you a PTDISTRIBUTION object, and from there you can sample, score, or invert it with the pack's Ptdm* nodes.

This node comes from ComfyUI-Pt-Wrapper, Hide Inada's no-code PyTorch lab for ComfyUI (a spin-off of his ComfyUI-Data-Analysis). It's not a diffusion node - it's for the crowd doing ML prototyping, statistics, or plotting in the graph instead of in a notebook. If that's not you, this node is dead weight. If it is you, it's genuinely handy.

How it works

Under the hood it builds a torch.distributions.Gamma, then wraps it in a small subclass the pack calls GammaEx. That subclass exists for one reason: PyTorch's Gamma implements cdf but not icdf. The Ex version fills that gap using scipy.stats.gamma.ppf, which is exactly why scipy shows up in the pack's requirements. So PtdGamma -> PtdmIcdf works, where plain PyTorch would throw NotImplementedError.

Inputs that matter

Both inputs are text boxes, not number widgets, and the pack parses them with Python's literal_eval:

• alpha - the shape (concentration) parameter. A scalar like 2.0 or a tuple like (1.0, 2.0, 3.0) for a batch of distributions.
• beta - the rate parameter. Same deal: 1.0, or a tuple to match alpha.

Just type the literal - no quotes, no torch.tensor(...), no function call. 2.0 not 2 if you want float behavior, and keep the two shapes aligned or you'll get a broadcast error.

The single output is PTDISTRIBUTION, which is the pack-wide type that every PtdmCdf, PtdmPdf, PtdmSample, and so on consumes. That's the whole flow: build a distribution with a Ptd* node, then hang query nodes off it.

Installing it

Through ComfyUI Manager: Install Custom Nodes → Search → "Pt Wrapper" (the pack title is ComfyUI-Pt-Wrapper), install, restart. Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/HowToSD/ComfyUI-Pt-Wrapper

Then restart ComfyUI. Note the first install is heavy: the pack pulls matplotlib, pandas, scipy, scikit-learn, transformers, datasets, and pins gensim==4.3.2. Give it a few minutes. Nothing here downloads model files - the distribution nodes are pure math.

Common issues

The big one is mistyping the text inputs. If alpha is empty or has a typo, literal_eval raises a SyntaxError and the node red-flags. Enter a real Python literal or leave the field sensible before you queue. Also remember beta is the rate, not the scale - a common confusion if you're coming from a stats package that parameterizes by scale. And if you run PtdmCdf on a Gamma you'll get a result, but PtdmIcdf relies on the pack's scipy shim, so don't skip the install of the full requirements or that node breaks while its sibling works.