An epoch can be read as one pass through the pinned samples in a stated order. This lesson counts visits and updates before making any convergence claim.

highlighted = computed this step

Pin the pass order

For this pinned pass, the sample order is s2 then s3. An epoch here means one visit to each sample in that stated order.

s2s3s2\rightarrow s3
One Epoch Is One PassVisit each pinned sample once in this stated order.one epoch is one pinned passpass positionsamplevisitupdatenew (a,b)1s211(1/2,3/2)2s322(-1/4,0)order s2 then s32 visits = 2 updates in one pinned passone pinned pass only; samples are chosen before the pass; NOT convergence; NOT generalization;NOT learning; NOT training success; NOT accuracy; NOT calibration; NOT probability truth; NOTmodel quality

Count the visits

There are 2 pinned samples in the pass, so the visit count is 2. Each sample appears once.

visit count=2\text{visit count}=2
One Epoch Is One PassVisit each pinned sample once in this stated order.one epoch is one pinned passpass positionsamplevisitupdatenew (a,b)1s211(1/2,3/2)2s322(-1/4,0)order s2 then s32 visits = 2 updates in one pinned passone pinned pass only; samples are chosen before the pass; NOT convergence; NOT generalization;NOT learning; NOT training success; NOT accuracy; NOT calibration; NOT probability truth; NOTmodel quality

Each visit makes one update

The first visit, s2, lands at (1/2, 3/2). The second visit, s3, lands at (-1/4, 0). That gives 2 updates in one pass.

update count=2\text{update count}=2
One Epoch Is One PassVisit each pinned sample once in this stated order.one epoch is one pinned passpass positionsamplevisitupdatenew (a,b)1s211(1/2,3/2)2s322(-1/4,0)order s2 then s32 visits = 2 updates in one pinned passone pinned pass only; samples are chosen before the pass; NOT convergence; NOT generalization;NOT learning; NOT training success; NOT accuracy; NOT calibration; NOT probability truth; NOTmodel quality

Boundary

This is pass, count, and order bookkeeping for pinned samples only: NOT convergence, NOT generalization, NOT learning, NOT training success, NOT accuracy, NOT calibration, NOT probability truth, NOT model quality.

one pinned pass only\text{one pinned pass only}
One Epoch Is One PassVisit each pinned sample once in this stated order.one epoch is one pinned passpass positionsamplevisitupdatenew (a,b)1s211(1/2,3/2)2s322(-1/4,0)order s2 then s32 visits = 2 updates in one pinned passone pinned pass only; samples are chosen before the pass; NOT convergence; NOT generalization;NOT learning; NOT training success; NOT accuracy; NOT calibration; NOT probability truth; NOTmodel quality