The essential difference between float32 and float64 (the influence of type on deep learning and the use of Python)

First, we need to know what is bits and bytes?

1. bits: Named digit
2. bytes: For bytes
3. A simple number is MB and G The relationship between ！ that 8bits=1bytes, The following is the mutual transformation of various units ！

  that float32 and float64 What's the difference ?

1. The difference between digits
2. One takes up a share of memory 32 and 64 individual bits, That is to say 4bytes or 8bytes
3. The higher the number of digits, the higher the precision of floating-point numbers

It will affect the computational efficiency of deep learning ?

float64 Occupied memory is float32 Twice as many , yes float16 Of 4 times ; For example, for CIFAR10 Data sets , If the float64 To express , need 60000*32*32*3*8/1024**3=1.4G, Just putting data sets into memory requires 1.4G; If the float32, It only needs 0.7G, If the float16, It only needs 0.35G about ; How much memory does it take , It will have a serious impact on the operation efficiency of the system ;（ Therefore, the data set files adopt uint8 To store data , Keep files to a minimum ）

Pandas Table of memory occupied by each data type

Python Use numpy Set the data type to float32

import numpy as np
data = np.array([1,2,3,4,5,6,7,8,9],dtype='float32')

pandas analysis float32 and 64 Memory footprint

pandas The reason why the data processing speed is fast is that it loads the data into the memory for processing , Use pandas Load data to test different data types , Memory occupied by the same data ~

import pandas as pd
data = pd.DataFrame([1.2, 1.3, 1.4, 1.5, 1.6, 1.7, 1.8, 1.9], dtype="float64")
data1 = pd.DataFrame([1.2, 1.3, 1.4, 1.5, 1.6, 1.7, 1.8, 1.9], dtype="float32")
print(data.info(memory_usage=True))
print(data1.info(memory_usage=True))

  The results are as follows , We just need to check the memory size ,float64 Data occupancy 192bytes,float32 Data occupancy 160bytes

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 8 entries, 0 to 7
Data columns (total 1 columns):
 #   Column  Non-Null Count  Dtype  
---  ------  --------------  -----  
 0   0       8 non-null      float64
dtypes: float64(1)
memory usage: 192.0 bytes       -----------------> Memory usage
None
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 8 entries, 0 to 7
Data columns (total 1 columns):
 #   Column  Non-Null Count  Dtype  
---  ------  --------------  -----  
 0   0       8 non-null      float32
dtypes: float32(1)
memory usage: 160.0 bytes        -----------------> Memory usage
None