diff --git a/Thursday/practical-part1/keras-mnist-bettercnn-vis-maxact3.py b/Thursday/practical-part1/keras-mnist-bettercnn-vis-maxact3.py
index ed65b075..9633d1cb 100644
--- a/Thursday/practical-part1/keras-mnist-bettercnn-vis-maxact3.py
+++ b/Thursday/practical-part1/keras-mnist-bettercnn-vis-maxact3.py
@@ -13,9 +13,8 @@
 step=1
 
 # we're interested in maximising outputs of the 3rd layer:
-layer_output = model.layers[3].output
-
-for i in xrange(0,15):
+layer_output = model.layers[2].output
+for i in range(0,15):
 	# build a loss function that maximizes the activation
 	# of the nth filter of the layer considered
 	loss = K.mean(layer_output[:, :, :, i])
@@ -33,7 +32,7 @@
 	input_img_data = np.random.random((1, 28, 28, 1)) * 0.07 + 0.5
 	
 	# run gradient ascent for 50 steps
-	for j in range(50):
+	for j in range(30):
 		loss_value, grads_value = iterate([input_img_data])
 		input_img_data += grads_value * step
 
diff --git a/Thursday/practical-part1/keras-tutorial.md b/Thursday/practical-part1/keras-tutorial.md
index 11af1208..cdf733b5 100644
--- a/Thursday/practical-part1/keras-tutorial.md
+++ b/Thursday/practical-part1/keras-tutorial.md
@@ -631,7 +631,7 @@ layer_output = model.layers[3].output
 for i in xrange(0,15):
 	# build a loss function that maximizes the activation
 	# of the nth filter of the layer considered
-	loss = K.mean(layer_output[:, i, :, :])
+	loss = K.mean(layer_output[:, :, :, i])
 
 	# compute the gradient of the input picture wrt this loss
 	grads = K.gradients(loss, input_img)[0]
diff --git a/Thursday/practical-part1/keras-tutorial.pdf b/Thursday/practical-part1/keras-tutorial.pdf
index 3a9e5acd..5f6d907b 100644
Binary files a/Thursday/practical-part1/keras-tutorial.pdf and b/Thursday/practical-part1/keras-tutorial.pdf differ
diff --git a/Thursday/practical-part1/maxact.png b/Thursday/practical-part1/maxact.png
index 7e454e59..66a64804 100644
Binary files a/Thursday/practical-part1/maxact.png and b/Thursday/practical-part1/maxact.png differ