leconteur

module purge -f
module load compilers/gcc/4.8.5 compilers/java/1.8 apps/buildtools cuda/7.5 libs/cuDNN/5 compilers/swig apps/git apps/bazel/0.4.3 apps/python/3.5.0

source source ~/pythonenvs/python3/bin/activate # My python 3 env
 
OPWD=$(pwd)
TF_COMPILE_PATH=/tmp/${USER}_$(date +'%s')
BAZEL_ROOT_PATH=$TF_COMPILE_PATH/bazel
mkdir -p $TF_COMPILE_PATH; cd $TF_COMPILE_PATH

leconteur

from sklearn.datasets import fetch_mldata
from matplotlib import pyplot as plt
from sklearn.decomposition import PCA
from sklearn.model_selection import train_test_split
from sklearn.neighbors import KNeighborsClassifier
from sklearn.metrics import classification_report

if __name__ == "__main__":
    mnist = fetch_mldata('MNIST original')

leconteur

def cumsum(softmax):
  values = tf.split(1, softmax.get_shape()[1], softmax)
  out = []
  prev = tf.zeros_like(values[0])
  for val in values:
      s = prev + val
      out.append(s)
      prev = s
  cumsum = tf.concat(1, out)
  return cumsum

leconteur

linear = MLP([Identity(), Identity()], [2, 10, 2], weights_init=Constant(1), biases_init=Constant(2))
x = tensor.matrix('x')
y_hat = linear.apply(x)
cost = ....
cg = ComputationGraph(y)
weights = VariableFilter(roles=[WEIGHTS])(cg.variables)
cg = apply_dropout(cg, weights, 0.5)
target_cost = cg.outputs[0]

algorithm = GradientDescent(cost=target_cost, params=cg.parameters,...)