To round off my current Machine Learning adventures I gave a talk at a local Machine Learning group.



# Talking about learning

For a while now I have been attending the [Perth Machine Learning Meetup Group (PMLG)](https://www.meetup.com/Perth-Machine-Learning-Group/) and really enjoying the talks by the various incredibly talented data-science and machine learning experts that attend that group. So when they asked for more talks from community members it was with much hesitation that I put my hand up to talk about my minor fumblings in the world of machine learning in game development.

The talk basically went over my posts:

1. [A Game Developer Learns Machine Learning - Getting Started](/machine-learning/a-game-developer-learns-machine-learning-getting-started/)

2. [A Game Developer Learns Machine Learning - A Little Deeper](/machine-learning/a-game-developer-learns-machine-learning-a-little-deeper/)

3. [A Game Developer Learns Machine Learning - Mr Nibbles Basics](/machine-learning/a-game-developer-learns-machine-learning-mr-nibbles-basics/)

The talk went well despite a few internet issues. Most people hadnt done any Unity development before so even thought the ML aspect was probably a little basic for them, at least they will have learnt a little bit about how Unity works :)

I got some great feedback at the end on how I could potentially improve the model's learning through different application of rewards, I took notes and will definately be applying those suggestions next time I do some work on the project.

Its been almost exactly a week since the last developmental release of Mr Nibbles Forever. According to [the plan](https://www.mikecann.blog/myprojects/mr-nibbles-forever-getting-it-done/) this week should have been about implementing the landing page and gathering feedback from people, as I managed to get the landing page sorted in the first week I decided to spend the time developing the game instead.



So this week I have concentrated on decreasing the loading times for the game. On my Nexus 7 the loading time was about one minute which is way too long. The main reason for the load times is the way I construct the levels.

As mentioned [way back in my second post](https://www.mikecann.blog/myprojects/mr-nibbles-forever-prototype-2/) for the game the levels are made up of pre-defined chunks:

[![2015-07-12_15-24-52](https://www.mikecann.blog/wp-content/uploads/2015/07/2015-07-12_15-24-52-1024x589.png)](https://www.mikecann.blog/wp-content/uploads/2015/07/2015-07-12_15-24-52.png)

These chunks are individual scenes. I did it this way because Unity doesn't support nested prefabs and I wanted the ability to change properties of some of the elements of the levels as I was developing. The problem with this however is that to construct the levels I must first load the chunk scene, then extract its contents and build a level.

Unity's scene loading system is very slow and causes the game to pause and stutter if you try to do it at run time. As a result I put a "build" button into the editor which runs through a small script which loads all the chunks into a specially built "Release" scene:

[![2015-07-12_15-31-15](https://www.mikecann.blog/wp-content/uploads/2015/07/2015-07-12_15-31-15.png)](https://www.mikecann.blog/wp-content/uploads/2015/07/2015-07-12_15-31-15.png)

So now all the chunks are available in one single scene at startup removing the scene loading issue while the game is running. This is great but it does now mean that the initial load of the game is gigantic as it has to load 80 chunks with thousands of game objects at startup. The result is the game takes a minute to load.

To solve this problem I modified my "build" process so now instead of simply importing the chunk scenes it also processes them a little bit. One common technique in optomising game performance is to use object pooling to reduce the amount of new objects that are created at runtime.

I know that the levels are going to have large numbers of certain types of objects such as spiders, nibbles and level tiles, so what I decided to do was record the unique properties for each instance of the objects such as their position, rotation, sprite type etc. I could then delete those objects from each chunk.

Now when a chunk is used in a level, it simply requests an existing instance of that object type from a global pool, then sets the unique properties on the object such as position and rotation.

The result of all that work is that the load times have been cut in half. So now just 30 seconds on my 2012 Nexus 7\. There is probably more room for improvements there but considering that the deadline for the game is coming up, im happy with it for now.

[![2015-07-12_15-47-42](https://www.mikecann.blog/wp-content/uploads/2015/07/2015-07-12_15-47-42-1024x419.png)](https://www.mikecann.blog/wp-content/uploads/2015/07/2015-07-12_15-47-42.png)

In addition to that performance work, I also had a little more time to work through a few other small bugs and one more feature.

A few people had suggested to me that it would be good to add a "remove ads" option to the game as many people would be happy to remove the ads for a small fee. Well it wasnt a huge amount of work to do it so I put it in.

[![2015-07-12_15-50-59](https://www.mikecann.blog/wp-content/uploads/2015/07/2015-07-12_15-50-59.png)](https://www.mikecann.blog/wp-content/uploads/2015/07/2015-07-12_15-50-59.png)

[![2015-07-12_15-51-30](https://www.mikecann.blog/wp-content/uploads/2015/07/2015-07-12_15-51-30.png)](https://www.mikecann.blog/wp-content/uploads/2015/07/2015-07-12_15-51-30.png)

So now, you no longer need to watch an ad when you want to start or restart a challenge:

[![2015-07-12_15-53-33](https://www.mikecann.blog/wp-content/uploads/2015/07/2015-07-12_15-53-33.png)](https://www.mikecann.blog/wp-content/uploads/2015/07/2015-07-12_15-53-33.png)

Well that's it for this week. The plan for next week is to plan the game trailer and to experiment with full-screen post process effects again which will hopefully add a whole lot more to the individual visual aspect of each outfit.

As per usual if you are interested in testing the game on Android then join the community page: https://plus.google.com/u/0/communities/100690884724496136044 or for iOS email me: mike.cann@gmail.com

I haven't posted an update on my new game Mr Nibbles Forever for a couple of weeks not because I haven't been doing anything (quite the opposite) rather the game has been in a little flux and thus was a little too broken to blog about.



Well if you haven't seen it in a couple of weeks there has been quite a lot of changes to the meta-game:

<iframe width="1280" height="750" src="https://www.youtube.com/embed/3hzO1Io995o" frameborder="0" allowfullscreen></iframe>

There are now several "meta-game" features which I hope will encourage the user to keep playing the game, find new and interesting things, continue to challenge themselves and (if I have done things right) generate some revenue.

I introduce the player to the meta-game by initially giving them a "free gift":

[![2015-02-23_10-45-10](https://www.mikecann.blog/wp-content/uploads/2015/02/2015-02-23_10-45-10-1024x756.png)](https://www.mikecann.blog/wp-content/uploads/2015/02/2015-02-23_10-45-10.png)

This is timed mechanic which I borrowed from social games and commonly used to encourage the player to return to the game at some point in the future. In Mr Nibbles Forever it also serves the duel purpose of providing the player with some early "nibbles" which they can then immediately use on the wheel of fortune:

[![2015-02-23_10-47-06](https://www.mikecann.blog/wp-content/uploads/2015/02/2015-02-23_10-47-06-1024x764.png)](https://www.mikecann.blog/wp-content/uploads/2015/02/2015-02-23_10-47-06.png)

[![2015-02-23_10-47-41](https://www.mikecann.blog/wp-content/uploads/2015/02/2015-02-23_10-47-41-1024x768.png)](https://www.mikecann.blog/wp-content/uploads/2015/02/2015-02-23_10-47-41.png)

[![2015-02-23_10-48-07](https://www.mikecann.blog/wp-content/uploads/2015/02/2015-02-23_10-48-07-1024x765.png)](https://www.mikecann.blog/wp-content/uploads/2015/02/2015-02-23_10-48-07.png)

The initial spin of the wheel is hard-coded so that the player will always win an new outfit, that way I can introduce them to the purpose of earning nibbles (to get new outfits):

[![2015-02-23_10-49-22](https://www.mikecann.blog/wp-content/uploads/2015/02/2015-02-23_10-49-22-1024x763.png)](https://www.mikecann.blog/wp-content/uploads/2015/02/2015-02-23_10-49-22.png)

[![2015-02-23_10-49-48](https://www.mikecann.blog/wp-content/uploads/2015/02/2015-02-23_10-49-48-1024x766.png)](https://www.mikecann.blog/wp-content/uploads/2015/02/2015-02-23_10-49-48.png)

So now the player has one purpose to play the game, earn nibbles so they can get new outfits. I however wasn't happy that this was enough, I felt that the game needed another hook to encourage play.

I struggled for hours last week to come up with a way to do this. I really liked the way "Jet Pack Joyride" has missions which encourage you to go back and keep playing so you can complete a mission:

[![Jetpack_Joyride_missions_thumb](https://www.mikecann.blog/wp-content/uploads/2015/02/Jetpack_Joyride_missions_thumb.png)](https://www.mikecann.blog/wp-content/uploads/2015/02/Jetpack_Joyride_missions_thumb.png)

I thought about this however and struggled to think of a way to make that work with Mr Nibble's simple mechanics, it was just too complex. After a while I had a brainwave, why don't I just set the player a "challenge" which is similar to a mission but I could change the rules somehow. This way I could keep the player coming back and challenge them to go further without over-complicating things.

Now every now and then the player is given an option to do a "challenge run":

[![2015-02-23_10-55-18](https://www.mikecann.blog/wp-content/uploads/2015/02/2015-02-23_10-55-18-1024x762.png)](https://www.mikecann.blog/wp-content/uploads/2015/02/2015-02-23_10-55-18.png)

[![2015-02-23_10-55-44](https://www.mikecann.blog/wp-content/uploads/2015/02/2015-02-23_10-55-44-1024x767.png)](https://www.mikecann.blog/wp-content/uploads/2015/02/2015-02-23_10-55-44.png)
(ignore the terrible visuals for now, they will change later)

So the player is given an explicit challenge, in this case reach at least 200cm, and they are given a reward. To unlock it they must watch a video but its totally optional, so if they dont want to watch a video and earn some nibbles then they dont have to.

As the player completes challenges new ones are offered but they get harder and harder but at the same time the rewards get greater and greater.

[![2015-02-23_10-59-03](https://www.mikecann.blog/wp-content/uploads/2015/02/2015-02-23_10-59-03-1024x765.png)](https://www.mikecann.blog/wp-content/uploads/2015/02/2015-02-23_10-59-03.png)

I feel its a neat solution to the problem and I think its compelling enough that players wont mind watching a 15-30 second ad each time.

If the player fails the challenge then they are given the option to retry (by watching another ad):

[![2015-02-23_11-01-34](https://www.mikecann.blog/wp-content/uploads/2015/02/2015-02-23_11-01-34-1024x767.png)](https://www.mikecann.blog/wp-content/uploads/2015/02/2015-02-23_11-01-34.png)

At the moment there is only one type of challenge; "distance run", but I hope to have more types before the game launches. Some ideas are "fruit collection" where the player must collect a certain number of special nibbles (that are placed in strategic positions).

In addition to the challenges I have added a highscores button, it doesn't do much at the moment but I hope to have that up and running this week.

Anyways, thats it for now. As usual if you would like to play the game send me an email: mike.cann@gmail.com and ill sort you out, or hop on the facebook page for the game: [www.facebook.com/mrnibblesforever](www.facebook.com/mrnibblesforever)

This week im hoping to get the game feature complete, fix bugs and start getting new audio and art assets in there!

mr nibbles forever

mr nibbles forever

mr nibbles forever