AI Studio Answer Guide: Up / Down

Status: Per-family deep reference (companion to the consolidated Fusion AI Studio & ONNX User Guide) Audience: first-time AI Studio user Answer family: Up / Down Prediction style: Classification

1. The Plain-English Question

Up / Down asks:

This is the simplest directional classification family in AI Studio.

It gives the model only two choices:

• 1 = up
• 0 = not up

That makes it easy to start with, but also easy to misunderstand.

2. What Fusion Is Actually Building

Example:

• Target_UpDown10

Meaning:

Target_UpDown10[t] = 1 if Close[t+10] > Close[t] else 0

Plain-English translation:

• if price closes higher 10 bars later, label the row 1
• otherwise, label the row 0

Important

This is not the same thing as:

1 = long
0 = short

That is the main trap with this family.

3. The One Important Mental Model

Up / Down is a binary direction question, not a full long/flat/short system.

It only asks:

It does not tell you separately whether the bar finished:

• flat
• slightly down
• strongly down

Those all collapse into the same label:

• 0 = not up

4. What 1 and 0 Mean

This family is easiest to understand if you think of it this way:

1 = future close finished above current close
0 = future close did not finish above current close

That means class 0 mixes together:

• flat bars
• down bars

Very important

If you want:

• +1 = long
• 0 = flat
• -1 = short

then Direction (3-Class) is usually the better answer family.

5. What Step 1 Controls This Family

This family uses the normal Step 1:

• Fixed-Horizon Windows (bars)

If Step 1 says:

1,5,10,20

Fusion can create:

• Target_UpDown1
• Target_UpDown5
• Target_UpDown10
• Target_UpDown20

Then Step 2 chooses one of those columns for one model.

Important

For Up / Down:

• Look-ahead Windows matters
• the ATR race section does not matter
• the Step 1 neutral-band / flat-zone field does not define a separate flat class here

6. What The Model Output Means

This is a classification family.

The correct trader meaning is:

1 = up
0 = not up

Important current product note

The saved model contract may still need class decoding to turn the encoded class id back into the human meaning.

So the right mental model is:

not:

7. Why This Family Can Still Be Useful

Even though it is not a clean long/short target, it can still be useful when the user wants:

• the simplest possible first classification workflow
• a pure bullish-vs-not-bullish question
• a first pass before moving to Direction (3-Class)

This can be a good beginner family because it is easy to explain:

8. Why This Family Can Also Be Misleading

This is where users often make a mistake.

They see:

1 = long
0 = short

But that is not what the label means.

It really means:

1 = up
0 = flat or down

So if a user wants a direct short signal family, Up / Down is usually too blunt.

That is why Direction (3-Class) is often the cleaner trader-facing classification family.

9. How This Differs From Direction (3-Class)

Up / Down

Asks:

Output:

• 1 = up
• 0 = not up

Direction (3-Class)

Asks:

Output:

• +1 = up
• 0 = flat
• -1 = down

So the simpler family is:

• Up / Down

But the clearer trading family is often:

• Direction (3-Class)

10. Step-By-Step Up / Down Walkthrough

Step A: Prepare the raw dataset

Open Export and make sure:

• the date range is reasonable
• OHLCV is checked
• the block outputs you want as features are selected

Step B: Create the prepared dataset

In AI Studio Step 1:

• enter Look-ahead Windows, for example 5,10,20
• click Create New Capture

Now Fusion creates answer columns such as:

• Target_UpDown5
• Target_UpDown10
• Target_UpDown20

Step C: Choose the actual training question

In Step 2:

• Prediction Style: Classification
• Answer Column Source: Generated Answer Columns
• Generated Question Family: Up / Down
• choose one horizon

Example:

• 10 bars

Now the model is learning:

• Target_UpDown10

Step D: Choose the features

Pick the inputs that may help answer:

Useful first feature types often include:

• trend context
• momentum
• volatility context
• order flow

Good first rule:

• do not overload the first model with too many random features

Step E: Train

Choose:

• model name
• algorithm
• tuning on/off

Good first algorithms:

• LightGBM Classifier if available
• fallback: Random Forest Classifier

Step F: Read the result

The correct reading is:

1 = price finished above the current close
0 = price did not finish above the current close

That is not the same thing as:

1 = long
0 = short

11. Good First Session

If you want the easiest first classification workflow in AI Studio:

Step 1

• Look-ahead Windows: 5,10,20

Step 2

• Prediction Style: Classification
• Answer Column Source: Generated Answer Columns
• Generated Question Family: Up / Down
• choose one horizon such as 5 or 10 bars

Step 3

• Algorithm: LightGBM Classifier
• fallback: Random Forest Classifier
• tuning off for the first pass

This is a very simple first classification experiment.

12. What Not To Do

Do not treat 0 as a clean short label

Here, 0 means not up, not strictly short.

Do not assume this family has a flat class

It does not. Flat and down are mixed together in the 0 class.

Do not assume the neutral-band field creates a middle class here

That belongs to Direction (3-Class), not Up / Down.

Do not use the raw encoded class plot like a continuous oscillator

Decode the class meaning first.

13. Current Product Notes From The Up / Down Guide Pass

These are the product improvements this guide exposes.

A. The UI should say directly that class 0 means not up

Suggested note:

B. The catalog should visually warn that this is not a clean short label

Suggested compact display:

Up / Down
binary class | 1 = up | 0 = not up

C. The family picker should gently steer short/flat/long users toward Direction (3-Class)

Suggested note:

14. One-Sentence Rule To Remember

Up / Down means: