# Vector

The vector is the major data type in Coprocessor, it's a vector of values in the same type. Usually, it comes from extracting a column from the query result, but you can also construct it in the python script.

The vector is like the array type in the programming language, `Array`

in Apache Arrow or `NDArray`

in NumPy.

## Vector Types

The Coprocessor Engine supports the following vector types:

Type | Description |
---|---|

`vector[str]` | The sting type |

`vector[bool]` | The boolean type |

`vector[u8]` | The 8-bit unsigned integer type |

`vector[u16]` | The 16-bit unsigned integer type |

`vector[u32]` | The 32-bit unsigned integer type |

`vector[u64]` | The 64-bit unsigned integer type |

`vector[i8]` | The 8-bit signed integer type |

`vector[i16]` | The 16-bit signed integer type |

`vector[i32]` | The 32-bit signed integer type |

`vector[i64]` | The 64-bit signed integer type |

`vector[f32]` | The 32-bit floating point type |

`vector[f64]` | The 64-bit floating point type |

`vector[none]` | The any type |

As we see in Hello, world, we can define the return vector types for the coprocessor if we want to use it as SQL UDF. Otherwise, we can ignore the return vector types declaration:

```
@coprocessor(returns=['msg'])
def hello() -> vector[str]:
return "hello, GreptimeDB"
```

The Coprocessor Engine will infer the return vector types by the result. But without the declaration, we can't call it in SQL except by HTTP API.

## Construct a vector

We have already seen the example that extracting vectors from the query result by executing the `sql`

attribute in `@coprocessor`

in Coprocessor.

We can create a vector from literals:

```
@copr(returns=["value"])
def answer() -> vector[i64]:
return 42
```

The result `42`

will be wrapped as a one-element vector of `vector[i64]`

.

```
mysql> select answer();
+----------+
| answer() |
+----------+
| 42 |
+----------+
1 row in set (0.01 sec)
```

We can create a vector from a python list:

```
from greptime import vector
@copr(returns=["value"])
def answer() -> vector[i64]:
return vector([42, 43, 44])
```

The `greptime`

is a built-in module, please refer to Builtin Modules. In fact, the `vector`

function can create a vector from any iterable object in python. But it requires all the element types must be the same, and it chooses the first element's type as its vector type.

## Vector operations

The vector supports a lot of operations:

- Basic arithmetic operators are supported, including
`+`

,`-`

,`*`

,`/`

. - Basic logic operations are supported, including
`&`

,`|`

,`~`

. - Basic comparison operation including
`>`

,`<`

,`>=`

,`<=`

,`==`

,`!=`

are supported too.

Note: Here we override bitwise and

`&`

, bitwise or`|`

, bitwise not`~`

logical operation, because Python doesn't support logical operation override(You can't override`and`

`or`

`not`

). PEP335 made a proposal and was eventually rejected. But bitwise operators have higher precedence than comparison operators, so remember to use a pair of parentheses to make sure the logical operation is what you want. i.e. if you want to filter a vector that's between 0 and 100, you should use`(vector[i32] >= 0) & (vector[i32] <= 100)`

not`vector[i32] >= 0 & vector[i32] <= 100`

. Later one will be evaluated as`vector[i32] >= (0 & vector[i32]) <= 100`

.

For example, you can plus two vectors directly:

```
@copr(args=["n1", "n2"],
returns=["value"],
sql="select number as n1,number as n2 from numbers limit 5")
def add_vectors(n1, n2) -> vector[i32]:
return n1 + n2
```

Or do a comparison with a bool array in a Numpy way:

```
from greptime import vector
@copr(returns=["value"])
def compare() -> vector[bool]:
# This returns a vector([False, False, True])
return vector([1.0, 2.0, 3.0]) > 2.0
```

And using the boolean array indexing:

```
from greptime import vector
@copr(returns=["value"])
def boolean_array() -> vector[f64]:
v = vector([1.0, 2.0, 3.0])
# This returns a vector([2.0])
return v[(v > 1) & (v< 3)]
```

Comparison between two vectors is also supported:

```
from greptime import vector
@copr(returns=["value"])
def compare_vectors() -> vector[bool]:
# This returns a vector([False, False, True])
return vector([1.0, 2.0, 3.0]) > vector([1.0, 2.0, 2.0])
```

Using an indexed bool array to select elements from a vector:

```
from greptime import vector
@copr(returns=["value"])
def select_elements() -> (vector[f64]):
a = vector([1.0, 2.0, 3.0])
# This returns a vector([2.0, 3.0])
return a[a>=2.0]
```

Of course, we can use list comprehension to construct a new vector:

```
from greptime import vector
@copr(returns=["value"])
def list_comprehension() -> (vector[f64]):
a = vector([1.0, 2.0, 3.0])
# This returns a vector([3.0, 4.0])
return [x+1 for x in a if a >= 2.0]
```