# layer2_h input shape req : batch_size x (50*4*4)
layer2_h_input = layer1.output.flatten(2)
# n_in = 50x4x4 pixels; n_out = 500 hidden nodes
layer2_h = HiddenLayer(rng=rng,input=layer2_h_input,n_in=50*4*4,n_out=500)
# In[ ]:
# Layer 3 : Output layer : LogisticRegression
layer3_o = LogisticRegression(input=layer2_h.output,n_in=500,n_out=10)
# In[ ]:
# cost
cost = layer3_o.neg_log_likelihood(y)
# >> setup gradient expression <<
### Need :parameters
params = layer3_o.params + layer2_h.params + layer1.params + layer0.params
gparams = T.grad(cost,params)
# In[ ]:
## Updates ##
updates = [(param, param - gparam*learning_rate)
for param,gparam in zip(params,gparams)]
index = T.lscalar('index')
# compile train
