Reshape & Transpose

import numpy as np

# 1) Reshape — total element count must match
a = np.arange(12)                 # shape (12,)
a.reshape(3, 4)                   # shape (3, 4)
a.reshape(2, 2, 3)                # shape (2, 2, 3)
a.reshape(-1, 4)                  # -1 means 'infer this dim' → (3, 4)

# 2) reshape returns a VIEW when possible (no copy)
b = a.reshape(3, 4)
b[0, 0] = 99
print(a[0])                       # 99 — same memory

# Force copy
c = a.reshape(3, 4).copy()

# 3) ravel / flatten — back to 1-D
m = np.array([[1, 2], [3, 4]])
m.ravel()                         # [1 2 3 4]  view
m.flatten()                       # [1 2 3 4]  always a copy

# 4) Transpose — swap axes (view, no data move)
x = np.arange(6).reshape(2, 3)    # (2, 3)
x.T                               # (3, 2)
x.transpose()                     # same

# 3-D — reorder axes
img = np.zeros((100, 200, 3))     # H, W, C  (RGB)
chw = img.transpose(2, 0, 1)      # C, H, W  (PyTorch format)

# 5) Add / remove unit dimensions
v = np.array([1, 2, 3])           # (3,)
v[:, None]                        # (3, 1)  column vector
v[None, :]                        # (1, 3)  row vector
np.expand_dims(v, 0)              # (1, 3)
v.reshape(1, 3, 1).squeeze()      # (3,)  drops all unit dims

# 6) Broadcasting + reshape together
rows = np.arange(3)[:, None]      # (3, 1)
cols = np.arange(4)[None, :]      # (1, 4)
grid = rows + cols                # (3, 4)  outer-product-like

# 7) Stacking — create a new axis
stack = np.stack([np.eye(3), np.eye(3) * 2])    # (2, 3, 3)

# Concatenate along existing axis
np.concatenate([np.zeros((2, 3)), np.ones((2, 3))], axis=0)  # (4, 3)
np.concatenate([np.zeros((2, 3)), np.ones((2, 3))], axis=1)  # (2, 6)

# 8) Splitting
big = np.arange(24).reshape(4, 6)
np.split(big, 2, axis=1)           # two (4, 3) arrays
np.array_split(big, 3, axis=0)     # uneven split allowed

# 9) Tile + repeat
np.tile(np.array([1, 2]), 3)       # [1 2 1 2 1 2]
np.repeat(np.array([1, 2]), 3)     # [1 1 1 2 2 2]
np.tile([[1, 2]], (2, 3))          # (2, 6)

# 10) Pandas — wide ↔ long
import pandas as pd

long = pd.DataFrame({
    'date':    ['2024-01', '2024-01', '2024-02', '2024-02'],
    'product': ['A', 'B', 'A', 'B'],
    'sales':   [100, 200, 150, 250],
})

# Long → wide
wide = long.pivot(index='date', columns='product', values='sales')
#         product    A    B
# date
# 2024-01           100  200
# 2024-02           150  250

# Wide → long
wide.reset_index().melt(id_vars='date', var_name='product', value_name='sales')

# Pivot with aggregation (handles duplicates)
long.pivot_table(index='date', columns='product', values='sales', aggfunc='sum')

# 11) Stack / unstack — MultiIndex reshape
stacked = wide.stack()              # columns → inner index level
stacked.unstack()                   # back to wide

# 12) Common bugs
#   • reshape((-1,)) returns view; modifying it changes original
#   • transpose doesn't copy — np.ascontiguousarray() if C order needed
#   • PIL/OpenCV use H×W×C; PyTorch uses C×H×W — transpose between them
#   • pivot fails on duplicate (index, columns) pairs → use pivot_table
#   • np.newaxis is the same as None — both add a length-1 dimension
#   • Forgetting axis= in concatenate → uses 0 → wrong shape silently

import numpy as np
a = np.arange(12)
print(a.reshape(3, 4))
print(a.reshape(2, -1))    # -1 infers the size
print(a.reshape(3, 4).T)   # transpose