Date: January 16, 2024

Topic: Using Numpy

Recall

pandas dataframes can be treated as ndarrays

Notes

Relationship to pandas

• The data inside (excluding row and column headers) are ndarrays
• We can also just treat the dataframe as an ndarray directly

Finding properties of an ndarray

• ndarray.size: total number of elements in the array
• ndarray.shape: shape of each dimension starting with the outermost one
• ndarray.dtype: datatype of the array

Slicing is done with colon, starting with the outer array to the inner

Slicing with numpy

To get a specific set of rows and columns:

subset_array = ndarray[start_row:end_row+1, start_col:end_col+1] # Add 1 as upper is exclusive

# Example
ndarray[0:3, 1:3] # Get the first 3 rows, and 2nd and 3rd column

Create arrays with:

• Pre-defined values
• Empty array
• Zeros array

Creating ndarrays

With pre-defined values

• 1D-Array: np.array([2.3.4])
• 2D-Array: np.array([(2,3,4),(5,6,7)])

Empty array (np.empty)

• Add dimensions by specifying more numbers

• e.g., 3D empty array: np.empty((5,4,3)) - Depth of 5, 4 rows, 3 columns

• It does not give an empty array, but shows what ever values were present in corresponding memory location

  

  Possible empty array values

Zeros array (np.zeros)

• Same as above, but replace with np.zeros instead

Specifying datatype

np.ones((5,4), dtype=int) # Use dtype to specify

Create random arrays with:

• random/rand [0.0, 1.0)
• normal distribution
• random integers

Creating random ndarrays

Seed value

To ensure that arrays are generated consistently, define a seed with np.random.seed(<ANY NUMBER>)

Random numbers with sample from [0.0, 1.0)

# Generate an array full of random numbers, uniformly samples from [0.0, 1.0)
np.random.random((5, 4)) # pass in a size tuple

# Generate an array full of random numbers, uniformly samples from [0.0, 1.0)
np.random.rand(5, 4) # function arguments (not a tuple)

Sample from Gaussian distribution

# Sample numbers from a Gaussian (normal) distribution
np.random.normal(size=(2, 3)) # "standard normal" (mean = 0, s.d. = 1)

# Sample numbers from a Gaussian (normal) distribution
np.random.normal(50, 10, size=(2, 3)) # change mean to 50 and s.d. to 10

Random integers

# Random integers
np.random.randint(1) # a single integer in [0, 10)
np.random.randint(0, 10) # same as above, specifying [low, high) explicit
np.random.randint(0, 10, size=5) # 5 random integers as a 1D array
np.random.randint(0, 10, size=(2, 3)) # 2x3 array of random integers

<aside> 📌 SUMMARY: Numpy arrays are similar to dataframes, and can perform a series of operations stated above

</aside>