diff --git a/.gitignore b/.gitignore
new file mode 100644
index 0000000..496ee2c
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1 @@
+.DS_Store
\ No newline at end of file
diff --git a/RNN/.gitignore b/RNN/.gitignore
index 8a768dd..3dc20a9 100644
--- a/RNN/.gitignore
+++ b/RNN/.gitignore
@@ -2,4 +2,4 @@ gen/
 runjob-fullVideo.sh
 log/
 trash/
-
+.DS_Store
diff --git a/RNN/main.lua b/RNN/main.lua
index b7dbb98..6e9f5e4 100644
--- a/RNN/main.lua
+++ b/RNN/main.lua
@@ -45,7 +45,7 @@ cmd:option('-silent', 				false, 			'don`t print anything to stdout')
 cmd:option('-uniform', 				0.1, 			'initialize parameters using uniform distribution between -uniform and uniform. -1 means default initialization')
 cmd:option('-seqCriterion', 		false, 			'use SequencerCriterion or not. Note to use smaller LR')
 -- recurrent layer 
-cmd:option('-lstm', 				true, 			'use Long Short Term Memory (nn.LSTM instead of nn.Recurrent)')
+cmd:option('-lstm', 				false, 			'use Long Short Term Memory (nn.LSTM instead of nn.Recurrent)')
 cmd:option('-bn', 					false, 			'use batch normalization. Only supported with --lstm')
 cmd:option('-gru', 					false, 			'use Gated Recurrent Units (nn.GRU instead of nn.Recurrent)')
 cmd:option('-numSegment',			3, 			    'number of segments for each video')
@@ -128,6 +128,9 @@ local data  = require 'data-ucf101'
 local train = require 'train'
 local test  = require 'test'
 
+-- initialize bestAcc
+bestAcc = 0
+
 if opt.testOnly then
 	-- Begin testing with trained model
 	test(data.testData, data.testTarget)
@@ -137,8 +140,7 @@ end
 ------------------------------------------------------------
 -- Run
 ------------------------------------------------------------
--- initialize bestAcc
-bestAcc = 0
+
 
 print(sys.COLORS.red .. '==> training!')
 for iteration = 1, opt.maxEpoch do
diff --git a/RNN/test.lua b/RNN/test.lua
index 4f74a1c..a51c141 100644
--- a/RNN/test.lua
+++ b/RNN/test.lua
@@ -42,11 +42,16 @@ local prob = {}
 predsFrames = torch.Tensor(opt.batchSize, nClass, opt.rho - opt.numSegment + 1)
 -- end
 
+local logsoftmax = nn.LogSoftMax()
+
 if opt.cuda == true then
 	inputs = inputs:cuda()
 	targets = targets:cuda()
+	logsoftmax = logsoftmax:cuda()
 end
 
+
+
 -- test function
 function test(testData, testTarget)
 
@@ -162,7 +167,7 @@ function test(testData, testTarget)
 			end
 
 			preds = model:forward(inputsSegments)
-			preds:exp()
+			preds = logsoftmax:forward(preds):exp()
 		end
 
 		-- discard the redundant predictions and targets
@@ -233,7 +238,7 @@ function test(testData, testTarget)
 		torch.save(opt.save .. '/prob.txt', prob,'ascii')
 
 		if opt.saveModel == true then
-			if confusion.totalValid * 100 >= 92 then
+			if confusion.totalValid * 100 >= 94.2 then
 				checkpoints.save(epoch-1, model, optimState, bestModel, confusion.totalValid*100)
 			end
 		end