Logistic Regression with Theano [back]
At this point Theano is already installed in the system.
To work with the MNIST dataset we will use a helper method from Tensorflow package:
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("MNIST_data/", one_hot=False)Model Parameters:
- Number of epochs: 100
- Batch size: 256
- Learning rate: 0.01
- Train samples: 55000
- Validation samples: 5000
- Test samples: 10000
To run code:
$ cd ~/gitlab.altoros/776_DL_Libs_Benchmark.git/
$ git pull
$ cd src/Step02/SubStep-01-Theano/
$ python Theano_LogisticRegression.py
or
$ cd ~/gitlab.altoros/776_DL_Libs_Benchmark.git/
$ cd src/Step02/SubStep-01-Theano/
$ jupyter notebook Theano_LogisticRegression_Notebook.ipynb
and go to the URL http://ec2-54-86-145-119.compute-1.amazonaws.com:9999:
Quick results:
...
-------------
Optimization complete with best validation score of inf %, with test performance 8.253205 %
The code run for 100 epochs, with 3.045976 epochs/sec
The code for file [Theano_LogisticRegression.py] ran for 32.8s
Traing process:
Weights visualisation:
