Decrease save file sizes; fix nparams messages

pretrain_g.lua

@@ -89,7 +89,7 @@ function main()
     print("G autoencoder:")
     print(G_AUTOENCODER)
     print(string.format('Number of free parameters in G (total): %d', NN_UTILS.getNumberOfParameters(G_AUTOENCODER)))
-    if OPT.gpu ~= false then
+    if OPT.gpu == false then
         print(string.format('... encoder: %d', NN_UTILS.getNumberOfParameters(G_AUTOENCODER:get(1))))
         print(string.format('... decoder: %d', NN_UTILS.getNumberOfParameters(G_AUTOENCODER:get(2))))
     else
@@ -204,6 +204,7 @@ function epoch()
         print(string.format("<trainer> saving network to %s", filename))
 
         -- Clone the autoencoder and deactivate cuda mode
+        NN_UTILS.prepareNetworkForSave(G_AUTOENCODER)
         local G2 = G_AUTOENCODER:clone()
         G2:float()
         G2 = NN_UTILS.deactivateCuda(G2)

train.lua

@@ -130,7 +130,7 @@ function main()
         MODEL_D = tmp.D
         MODEL_G = tmp.G
         OPTSTATE = tmp.optstate
-        EPOCH = tmp.epoch
+        EPOCH = tmp.epoch + 1
         if NORMALIZE then
             NORMALIZE_MEAN = tmp.normalize_mean
             NORMALIZE_STD = tmp.normalize_std
@@ -255,7 +255,9 @@ function saveAs(filename)
       os.execute(string.format("mv %s %s.old", filename, filename))
     end
     print(string.format("<trainer> saving network to %s", filename))
-    torch.save(filename, {D = MODEL_D, G = MODEL_G, opt = OPT, plot_data = PLOT_DATA, epoch = EPOCH+1, normalize_mean=NORMALIZE_MEAN, normalize_std=NORMALIZE_STD})
+    NN_UTILS.prepareNetworkForSave(MODEL_G)
+    NN_UTILS.prepareNetworkForSave(MODEL_D)
+    torch.save(filename, {D = MODEL_D, G = MODEL_G, opt = OPT, plot_data = PLOT_DATA, epoch = EPOCH, normalize_mean=NORMALIZE_MEAN, normalize_std=NORMALIZE_STD})
 end
 
 main()

train_v.lua

@@ -205,9 +205,7 @@ function epoch()
         os.execute(string.format("mkdir -p %s", sys.dirname(filename)))
         print(string.format("<trainer> saving network to %s", filename))
 
-        -- apparently something in the OPTSTATE is a CudaTensor, so saving it and then loading
-        -- in CPU mode would cause an error
-        --torch.save(filename, {V=NN_UTILS.deactivateCuda(V), opt=OPT, EPOCH=EPOCH+1}) --, optstate=OPTSTATE
+        NN_UTILS.prepareNetworkForSave(V)
         torch.save(filename, {V=V, opt=OPT, EPOCH=EPOCH+1})
     end
 

utils/nn_utils.lua

@@ -411,6 +411,42 @@ function nn_utils.normalize(data, mean_, std_)
     return 0.5, 0.5
 end
 
+-- from https://github.com/torch/DEPRECEATED-torch7-distro/issues/47
+function nn_utils.zeroDataSize(data)
+    if type(data) == 'table' then
+        for i = 1, #data do
+            data[i] = nn_utils.zeroDataSize(data[i])
+        end
+    elseif type(data) == 'userdata' then
+        data = torch.Tensor():typeAs(data)
+    end
+    return data
+end
+
+-- from https://github.com/torch/DEPRECEATED-torch7-distro/issues/47
+-- Resize the output, gradInput, etc temporary tensors to zero (so that the on disk size is smaller)
+function nn_utils.prepareNetworkForSave(node)
+    if node.output ~= nil then
+        node.output = nn_utils.zeroDataSize(node.output)
+    end
+    if node.gradInput ~= nil then
+        node.gradInput = nn_utils.zeroDataSize(node.gradInput)
+    end
+    if node.finput ~= nil then
+        node.finput = nn_utils.zeroDataSize(node.finput)
+    end
+    -- Recurse on nodes with 'modules'
+    if (node.modules ~= nil) then
+        if (type(node.modules) == 'table') then
+            for i = 1, #node.modules do
+                local child = node.modules[i]
+                nn_utils.prepareNetworkForSave(child)
+            end
+        end
+    end
+    collectgarbage()
+end
+
 function nn_utils.getNumberOfParameters(net)
     local nparams = 0
     local dModules = net:listModules()