avg function can now perform reblocking

[kkleiner1]

avg function in gather.py now computes averages and standard errors on correlated (via pyqmc's reblock function) or uncorrelated QMC data. See if this version is set up the way you like.

[lkwagner]

I don't think you need the boolean argument. reblock is enough.

Also, actually I just noticed that this is incorrect for DMC; the weights need to be included, similar to this:

def weighted_reblock(array, weight, nblocks):
    """
    Helper function to reblock(); this function actually does the reblocking.
    """
    vals = np.array_split(array, nblocks, axis=0)
    weights = np.array_split(weight, nblocks, axis=0)

    return [ (v*w).mean(axis=0)/w.mean(axis=0) for v,w in zip(vals, weights)]

def read_mc_output(fname, warmup=5, reblock=16):
    ret = {"fname": fname, "warmup": warmup, "reblock": reblock}
    with h5py.File(fname,'r') as f:
        if 'weight' in f.keys():
            wt = f['weight'][warmup:]
        else:
            wt = 1.0
        for k in f.keys():
            if "energy" in k:
                vals = weighted_reblock(f[k][warmup:], wt, reblock)
                ret[k] = np.mean(vals)
                valserr = scipy.stats.sem(vals)
                ret[k + "_err"] = valserr
    return ret