gokceneraslan / dietnet Goto Github PK

Hey!

Thanks for building this model. I want to give it a try. When running make in the 1000G dir I get this output:

dietnet preprocess genotypes -p affy_samples.20141118.panel -k 5
make: dietnet: Command not found
Makefile:31: recipe for target 'genotypes_x.npy' failed
make: *** [genotypes_x.npy] Error 127

Also, your readme is not up to date with your build/install process. I am not quite sure how to run after I build.

Which version of tensorflow has this software been tested against?

When trying to install this software package I ran into the following : WARNING:tensorflow:From build/bdist.linux-x86_64/egg/dietnet/network.py:135: mean_squared_error (from tensorflow.contrib.losses.python.losses.loss_ops) is deprecated and will be removed after 2016-12-30.
Instructions for updating:
Use tf.losses.mean_squared_error instead.
Traceback (most recent call last):
File "/ri/shared/modules/dietnet/Jun17_2017/bin/dietnet", line 11, in
load_entry_point('dietnet==0.1', 'console_scripts', 'dietnet')()
File "build/bdist.linux-x86_64/egg/dietnet/main.py", line 121, in main
File "build/bdist.linux-x86_64/egg/dietnet/train.py", line 47, in train
File "build/bdist.linux-x86_64/egg/dietnet/network.py", line 135, in diet
File "/ri/shared/modules/dietnet/Jun17_2017/lib/python2.7/site-packages/tensorflow/python/util/deprecation.py", line 136, in new_func
return func(*args, **kwargs)
TypeError: mean_squared_error() got an unexpected keyword argument 'weight'

Looking at current versions of tensorflow :
https://www.tensorflow.org/api_docs/python/tf/losses/mean_squared_error
mean_squared_error(
labels,
predictions,
weights=1.0,
scope=None,
loss_collection=tf.GraphKeys.LOSSES,
reduction=Reduction.SUM_BY_NONZERO_WEIGHTS
)

On a whim I tried modifying
mean_squared_loss = slim.losses.mean_squared_error(xhat,
inputs,
weight=gamma)

to
mean_squared_loss = slim.losses.mean_squared_error(xhat,
inputs,
weights=gamma)

And it seems to work.
.

Hey @gokceneraslan ,

Last time Ill bug you before I build this from scratch. After many different drop out rates, learning rates, regularizations on weights, adding batch_shuffle on queue, multiplying reconstruction loss by different values (like the theano version), adding reduce mean to the softmax_cross_entropy function, and different variances on the weight initialization I have not been able to get the model to get an accuracy better than .06 although the loss seems to have converged.

When I print out the correct labels and the predicted labels, the model usually starts predicting all elements in the batch to the same thing. Meaning, by step 40-70 it begins to predict the same class across the whole batch.

Have you been able to get the accuracy up? Any suggestions?

Thanks again