itmat / pade Goto Github PK

Use the geometric mean of (c_i,0 + F) / (c_i,1 + F) for each of the i pairs.

• Rename github repo
• Change package names in code
• Change sphinx documentation

We want to support another bootstrapping method, where we do the sampling separately within each of the groups defined by the reduced model. In other words, permutations will not cross the nuisance group boundaries.

Before doing this bootstrapping, we'll shift the values so that the mean for each group (in the reduced model) equals 0.

I'll implement this alongside the existing technique, so we can choose which method to use via a command-line option.

Currently our model assumes that there are interaction terms. We need to additionally support a model where there are no interaction terms. We'll compute the coefficients using least squares for now.

Add an --equalize-mean option to make pade equalize the means for each group to 0, and make the default behavior to not equalize the mean.

It would be great if the user could choose a subset of columns to analyze, so they don't have to generate a new input file if they want to look at only a subset of the data. I'm thinking we could do this by adding an option to the "run" command that takes a list of column names or numbers.

Currently we pick one group as the baseline and report fold change with respect to that one group. We need to instead report fold change for each value of the nuisance variable. So instead of using one group as the baseline, use one group per value of nuisance variable as the baseline.

There is a case in Elisabetta's input where we have a nan (not a number) symbol for some of the statistic values. This causes all kinds of errors when we try to apply confidence scores based on the statistic values.

If the user specifies a full model that has groups with only one replicate each, we will end up with a zero in the denominator of the f-test for every feature. This won't work, so we should reject any model that results in such a grouping.