VII. NumPy Input / Output (输入输出)#

NumPy supports two file formats: text files (文本文件) like CSV for human-readable data, and binary files (二进制文件) like .npy / .npz for fast, compact storage. Use text for sharing; use binary for speed.

1. `savetxt()` / `loadtxt()` — Text Files (文本文件)#

Core idea: Save/load arrays as human-readable text (CSV, TSV, etc.).

1
import numpy as np
2

3
a = np.array([[1.0, 2.0, 3.0],
4
              [4.0, 5.0, 6.0]])
5

6
# Save to CSV (保存为CSV)
7
np.savetxt('data.csv', a, delimiter=',', fmt='%.2f',
8
           header='col1,col2,col3', comments='')
9

10
# Load from CSV (从CSV加载)
11
b = np.loadtxt('data.csv', delimiter=',', skiprows=1)
12
print(b)
13
# [[1. 2. 3.]
14
#  [4. 5. 6.]]

Note: loadtxt is for numeric data only. For mixed types (strings + numbers), use np.genfromtxt() or pandas.read_csv().

2. `save()` / `load()` — Binary `.npy` Format (二进制格式)#

Core idea: Save a single array to a .npy binary file — preserves dtype, shape, and is much faster than text.

1
a = np.array([1, 2, 3, 4, 5])
2

3
np.save('my_array.npy', a)       # auto-adds .npy extension
4

5
b = np.load('my_array.npy')
6
print(b)   # [1 2 3 4 5]

Format	Extension	Speed	Human-readable?
Text	`.csv`, `.txt`	Slow	✅ Yes
Binary	`.npy`	Fast	❌ No

3. `savez()` / `load()` — Multiple Arrays (多数组存储)#

Core idea: Save multiple arrays into a single .npz file (a zip of .npy files).

1
a = np.array([1, 2, 3])
2
b = np.array([[4, 5], [6, 7]])
3

4
np.savez('multi.npz', x=a, y=b)            # uncompressed
5
np.savez_compressed('multi.npz', x=a, y=b) # compressed (smaller)
6

7
data = np.load('multi.npz')
8
print(data['x'])   # [1 2 3]
9
print(data['y'])   # [[4 5] [6 7]]

4. `tofile()` / `fromfile()` — Raw Binary (原始二进制)#

Core idea: Write raw bytes to disk — no metadata (shape/dtype) is saved. You must know the dtype and shape to reload correctly.

1
a = np.array([1, 2, 3, 4], dtype=np.int32)
2
a.tofile('raw.bin')
3

4
b = np.fromfile('raw.bin', dtype=np.int32)
5
print(b)   # [1 2 3 4]  — shape info is LOST, always 1-D

Note: tofile/fromfile do NOT save dtype or shape. Always use save/load (.npy) unless you need raw binary for interoperability with C/Fortran code.

5. `genfromtxt()` — Robust Text Loading (鲁棒文本加载)#

Core idea: Like loadtxt but handles missing values (缺失值) and mixed data types.

1
# CSV with missing values
2
# 1,2,NaN
3
# 4,,6
4
data = np.genfromtxt('messy.csv', delimiter=',',
5
                     filling_values=0)   # replace NaN with 0

6. Quick Comparison Table#

Function	Format	Preserves dtype/shape?	Use case
`savetxt / loadtxt`	Text (CSV)	❌	Share data
`save / load`	`.npy` binary	✅	Fast single array
`savez / load`	`.npz` binary	✅	Multiple arrays
`tofile / fromfile`	Raw binary	❌	C interop
`genfromtxt`	Text	❌	Missing values

💡 One-line Takeaway
Use .npy / .npz for fast internal storage, savetxt for sharing CSVs, and genfromtxt when data has missing values.

VII. NumPy Input / Output (输入输出)#

1. savetxt() / loadtxt() — Text Files (文本文件)#

2. save() / load() — Binary .npy Format (二进制格式)#

3. savez() / load() — Multiple Arrays (多数组存储)#

4. tofile() / fromfile() — Raw Binary (原始二进制)#

5. genfromtxt() — Robust Text Loading (鲁棒文本加载)#