Training the Model and Observing Results

At this point in the module you are about to run the training step using the modified function. Before you press run, it helps to know roughly how long the wait will be. A learner who expects a few seconds and gets several minutes may assume something has gone wrong. A learner who expects several minutes can plan around the wait. The source gives one concrete estimate for this step, and this section shows how to read that estimate correctly without assuming it applies to every machine.

The key phrase in that definition is specific hardware configuration. Training time is not a fixed property of the model alone. The same model, trained on different hardware, can take a different amount of time. So when someone quotes a training time, the hardware it was measured on is part of the claim. A number with no hardware attached is incomplete.

The source makes one specific claim: training will take about 5 minutes to complete on a MacBook Air or similar laptop. Read that sentence in two parts. The first part, about 5 minutes, is an approximate duration, not an exact one. Treat it as a rough planning figure. The second part names the setup the estimate applies to: a MacBook Air or a laptop similar to it. The estimate describes that class of machine, and nothing beyond it is promised.

The diagram shows the two columns of the comparison. The quoted estimate attaches to the laptop class on the left. Your own machine sits in a separate column, and its time is not covered by the quote. If your hardware differs from the named laptop class, the source's advice is to measure one run yourself and use that measurement in your plans.

An estimate is useful when you apply it to a decision. Suppose you have a half-hour block to experiment with the model and you want to know how many complete training runs you could fit in. If one run takes about 5 minutes on a laptop like the one named in the source, the arithmetic is straightforward. You divide the available time by the time for one run. The result is a planning number that you can later check against what you observe on your own machine.

The timeline makes the accumulation visible. Each run adds about 5 minutes to the running total, and the total reaches 30 minutes exactly when the sixth run ends. The same method works for any session length. A two-hour block at the same 5-minute planning figure would hold about 24 runs (120 divided by 5), though that count is only as reliable as the estimate behind it.

The plan above rests on one number. Its reliability depends on where that number came from. While the number comes from the quoted laptop class, the plan is a reasonable starting point. Once you run a training step on your own machine, the plan can rest on your own measurement instead. The diagram compares these two starting points.