0

I'm running the mnist example of mxnet in Spyder IDE. The training does not progress, just as if the learning rate were 0 (see output below).

If I run the same file on the console using either python or ipython, it works as expected (seeing improvements in the second epoch).

If I run the file using the normal Python console of Spyder, it also works. But if I run the file using the ipython console, I get the output shown below.

I'm new to Python. Any ideas?

Source code

#!/usr/bin/env python2
# -*- coding: utf-8 -*-

#%%
import mxnet as mx

#%% load mnist

mnist = mx.test_utils.get_mnist()

#%% define data iterators

batch_size = 100
train_iter = mx.io.NDArrayIter(mnist['train_data'], mnist['train_label'], batch_size, shuffle=True)
val_iter = mx.io.NDArrayIter(mnist['test_data'], mnist['test_label'], batch_size)

#%% create input variable

data = mx.sym.var('data')
# Flatten the data from 4-D shape into 2-D (batch_size, num_channel*width*height)
data = mx.sym.flatten(data=data)

#%% Multilayer Perceptron with softmax

# The first fully-connected layer and the corresponding activation function
fc1  = mx.sym.FullyConnected(data=data, num_hidden=128)
act1 = mx.sym.Activation(data=fc1, act_type="relu")

# The second fully-connected layer and the corresponding activation function
fc2  = mx.sym.FullyConnected(data=act1, num_hidden = 64)
act2 = mx.sym.Activation(data=fc2, act_type="relu")

# MNIST has 10 classes
fc3  = mx.sym.FullyConnected(data=act2, num_hidden=10)
# Softmax with cross entropy loss
mlp  = mx.sym.SoftmaxOutput(data=fc3, name='softmax')

#%% Training
import logging
logging.getLogger().setLevel(logging.DEBUG)  # logging to stdout
# create a trainable module on CPU
mlp_model = mx.mod.Module(symbol=mlp, context=mx.cpu())
mlp_model.fit(train_iter,  # train data
              eval_data=val_iter,  # validation data
              optimizer='sgd',  # use SGD to train
              optimizer_params={'learning_rate':0.1},  # use fixed learning rate
              eval_metric='acc',  # report accuracy during training
              batch_end_callback = mx.callback.Speedometer(batch_size, 100), # output progress for each 100 data batches
              num_epoch=10)  # train for at most 10 dataset passes

Output of ipython console in Spyder

runfile('/home/thomas/workspace/clover-python/mnist.py', wdir='/home/thomas/workspace/clover-python')
INFO:root:train-labels-idx1-ubyte.gz exists, skipping download
INFO:root:train-images-idx3-ubyte.gz exists, skipping download
INFO:root:t10k-labels-idx1-ubyte.gz exists, skipping download
INFO:root:t10k-images-idx3-ubyte.gz exists, skipping download
INFO:root:Epoch[0] Batch [100]  Speed: 1304.05 samples/sec      accuracy=0.097921
INFO:root:Epoch[0] Batch [200]  Speed: 1059.19 samples/sec      accuracy=0.098200
INFO:root:Epoch[0] Batch [300]  Speed: 1178.64 samples/sec      accuracy=0.099600
INFO:root:Epoch[0] Batch [400]  Speed: 1292.71 samples/sec      accuracy=0.098900
INFO:root:Epoch[0] Batch [500]  Speed: 1394.21 samples/sec      accuracy=0.096500
INFO:root:Epoch[0] Train-accuracy=0.101212
INFO:root:Epoch[0] Time cost=47.798
INFO:root:Epoch[0] Validation-accuracy=0.098000
INFO:root:Epoch[1] Batch [100]  Speed: 1247.47 samples/sec      accuracy=0.097921
INFO:root:Epoch[1] Batch [200]  Speed: 1673.79 samples/sec      accuracy=0.098200
INFO:root:Epoch[1] Batch [300]  Speed: 1283.91 samples/sec      accuracy=0.099600
INFO:root:Epoch[1] Batch [400]  Speed: 1247.79 samples/sec      accuracy=0.098900
INFO:root:Epoch[1] Batch [500]  Speed: 1371.93 samples/sec      accuracy=0.096500
INFO:root:Epoch[1] Train-accuracy=0.101212
INFO:root:Epoch[1] Time cost=44.201
INFO:root:Epoch[1] Validation-accuracy=0.098000
INFO:root:Epoch[2] Batch [100]  Speed: 1387.72 samples/sec      accuracy=0.097921
INFO:root:Epoch[2] Batch [200]  Speed: 1196.37 samples/sec      accuracy=0.098200
INFO:root:Epoch[2] Batch [300]  Speed: 1220.44 samples/sec      accuracy=0.099600
INFO:root:Epoch[2] Batch [400]  Speed: 1387.75 samples/sec      accuracy=0.098900
INFO:root:Epoch[2] Batch [500]  Speed: 1279.58 samples/sec      accuracy=0.096500
INFO:root:Epoch[2] Train-accuracy=0.101212
INFO:root:Epoch[2] Time cost=46.929
INFO:root:Epoch[2] Validation-accuracy=0.098000
INFO:root:Epoch[3] Batch [100]  Speed: 1266.24 samples/sec      accuracy=0.097921

and so on…
Community
  • 1
  • 1
ziggystar
  • 28,410
  • 9
  • 72
  • 124

2 Answers2

0

Do you have multiple versions of MXNet installed? One thing I can think of is to verify if ipython console and python console is using the same version of python and MXNet. In ipython, you can type import mxnet; mxnet? to see what version of MXNet you're using.

eric-haibin-lin
  • 377
  • 2
  • 9
  • I've searched for `mxnet` on my computer. Apparently I only have 0.11.0 for Python 2.7. And `mxnet?` only works in ipython, not in the python console. – ziggystar Nov 17 '17 at 20:15
0

Per Eric's comment, let's verify that you're running the same version of mxnet in each environment. You can type

import mxnet
print(mxnet.__version__)
Simon Corston-Oliver
  • 226
  • 1
  • 6