Subspace Reticulum Dev Log - Version 1.0.5

Hey, spacefarers and code wranglers! Welcome to the latest dev log for Subspace Reticulum, a neural-network-driven, retro-inspired space shooter. Version 1.0.5 is live, and it’s packed with new toys, visual polish, and a squashed bug that’s been haunting us since the Python days. Let’s dive into what’s new in the galaxy!

What's New

• Bullets Join the Arsenal: Say hello to bullets, your new best friend for chipping away at enemy ships! They’re not as beefy as bombs, but pair them with a well-timed explosive, and you’ll be blasting foes to stardust. Perfect for strategic combos.
• Shields Up!: Ships now have shields for extra protection. These bad boys absorb damage, giving you a fighting chance against your foes. Stay sharp, though—shields don’t regenerate like your energy.
• Prizes, Prizes Everywhere: Randomly spawning prizes now dot the map, and destroyed enemies drop loot too! Snag Shields, Energy Refills, or bonus points to keep your ship in top shape. Who doesn’t love a good treasure hunt?
• Visual Glow-Up: Been tinkering with the game’s look. Projectiles now have a sleeker, flashier design, and the UI got a facelift for better readability and style. It’s not a full overhaul yet, but it’s getting there!

Challenges Overcome

• The Great Rotation Bug Fix: Facepalm moment—this MonoGame port wasn’t feeding ship rotations to the neural network. This was a oversight when porting from the Python prototype that slipped through the cracks. After some proper C# type-casting, it’s fixed! The neural net now performs as expected, matching the Python version’s prowess. Lesson learned: always double-check your data types when jumping languages. This fix makes AI-controlled ships way smarter, so brace for tougher dogfights!
• New Model : I needed to rebuild the neural network to account for the additional input parameters. This process is generally rather cumbersome but it ended up in a new model architecture that I think is actually significantly better. I may experiment with a mixture of both at some point, but this version only uses the newest version of the neural network. I need to do some experimentation to see the strengths and weaknesses of both models. The new model allows for better scaling. This current network is only 3 layers and seems to achieve a higher level of skill over my prior 5 layer version. My prior testing was a bit tainted with the Rotation Bug, but even in my Python version that didn't have the same issue, the new version seems to be a winner.

What's Next

• More Prize Goodies: Working on cooking up new prize types like weapon mods and special equipment. Thinking about adding a shop for upgrades.
• 1v1 Duel Mode: A new game mode is on the horizon: a high-stakes duel between you and a single, souped-up bot. With more resources for the neural net, I'll begin experimenting with larger models with longer memory for a truly cunning opponent.
• Persistent Server Vibes: Exploring ideas to simulate a persistent server feel, so the game feels alive even when you’re offline. No promises, but it’s an interesting dream worth chasing.
• UI & Menu Revamp: The menus and UI are due for a full redo to make navigation smoother and slicker, and pave the way for game mode selection. Stay tuned for a more polished experience.

This version marks the milestone of accomplishing what I really envisioned for a first version of this game. The next update could take some time, as I plan to rebuild and optimize as much as possible, then implementing more features. A lot of the game code is rather sloppy as I was mostly focused on just getting things to work no matter how it looked. 
Thanks to anyone that gives this a try. If you have any issues, let me know.