About variable of chainer

Chainer Variable related

This article assumes python 3.x. Also, please note that I am a python amateur. The first thing I wanted to do with machine learning was to do some learning on my own dataset, but the first thing I got stuck in was this variable.

import numpy as np
import chainer as Variable

The type must be Variable when throwing data to a machine learning network defined using chainer. At that time, it is necessary to always use np.array for the data given to the argument. This np.array is an array used by python's numpy, which is different from the standard python list ([]). Looking at the Chainer documentation, what you can specify for variable is [data – Data array of type either numpy.ndarray or cupy.ndarray.] Since it is described as, it is necessary to convert it to a numpy array once instead of using the list data as it is. Also, if you use floating point, you may need to convert to float32. I think the reason for using float32 is that chainer does not support float64, so an error occurs. This time, I considered some conversion patterns up to variable.

Example 1: ** Array **

x_train = [0,0,0,0,0]
x_np_train = np.array(x_train, dtype=np.float32)
x_val_train = Variable(x_np_train, 1)

By doing so, the array is converted to np.array while specifying float32, and it is further converted to variable.

Example 2: ** Multidimensional array ** You can follow the same procedure.

x_train = [[0.1, 0,2],[0.3, 0.4]]
x_np_train = np.array(x_train, dtype=np.float32)
x_val_train = Variable(x_np_train, 1)

Example 3: ** If np.array already exists **

x_train = [[0.1, 0,2],[0.3, 0.4]]
x_np_train = np.array(x_train)
x_np_train.dtype

Then it is dtype ('float64'). When dealing with np.array data from the beginning, I think that there are many cases where you have to cast to float32.

x_np_train.dtype = np.float32
x_val_train = Variable(x_np_train, 1)

Example 4: ** Variable has already been created with another type **

x_train = [[0.1, 0,2],[0.3, 0.4]]
x_np_train = np.array(x_train, dtype=np.float64)
x_val_train = Variable(x_np_train, 1)
x_val_train.data.dtype

If, dtype ('float64') is displayed. If nothing is done, an error will occur.

x_val_train.data.dtype = np.float32
x_val_train.data.dtype

Then, dtype ('float32') is displayed.

Recommended Posts

About variable of chainer
About variable scope. .. ..
About the behavior of enable_backprop of Chainer v2
About all of numpy
About assignment of numpy.ndarray
About MultiIndex of pandas
Implementation of Chainer series learning using variable length mini-batch
About max_iter of LogisticRegression () of scikit-learn
About the ease of Python
About Japanese support of cometchat
About various encodings of Python 3
About all of numpy (2nd)
About cost calculation of MeCab
About the components of Luigi
About 2-variable, 4-branch if statement
Variable parameter expansion of bash
About HOG output of Scikit-Image
About the features of Python
About data management of anvil-app-server
About the return value of pthread_mutex_init ()
About the return value of the histogram.
About the basic type of Go
About the upper limit of threads-max
About the behavior of yield_per of SqlAlchemy
About import error of PyQt5.QtWidgets (Anaconda)
About the size of matplotlib points
About color halftone processing of images
About the basics list of Python basics
About building GUI using TKinter of Python
About the virtual environment of python version 3.7
About sensor_mode and angle of view of picamera
Implementation of dialogue system using Chainer [seq2seq]
Memorandum of python beginners About inclusion notation
A memorandum of understanding about django's QueryDict
About the arguments of the setup function of PyCaret
About Japanese fonts of matplotlib (for Mac)
About the Normal Equation of Linear Regression
Implementation of "blurred" neural network using Chainer
Memo of troubles about coexistence of Python 2/3 system
[Python] Chapter 02-04 Basics of Python Program (About Comments)