Repository for a Hackatron on 19 November, powered by NUWE.io and organized by General Electric Europe.
The objetive of this Hackatron was predict from some given images what type of deforestation shows.
The given files are not in this repository. There are 3 files to download (train.csv, test.csv and a compressed file with the images).
This repository contains:
folder of notebook
folder or datas
folder of Images
readme file
2 submissions files (one loaded as .json and another loaded as .csv)
presentation file to explain the project
main as .py o .ipynb (in this case I called it notebook_1)
For this Hackatron I decided to test how good I can work with a Neural Convolutional Network.
- First: I took a look for the quality of train datas.
In this situation I looked for duplicates situations and how balanced was the target feature.
- Second: I prepared the Neural Network.
This Neural Network had 8 layers.
- Third: I predicted target for test datas and saved in a .csv file and .json file.
I didn't work enough time in my training with this kind of prediction models so I made the decision to try it in this Hackatron.
The result is not quite good, because I only had time to prepare 1 model and this kind of Neural Network spend too much time.
These points are related to files to submit:
Time to prepare presentation, search of information about the thema.
Time to get a better analysis and visualization.
This point is related to improve results:
Get a better knowledge for predicting models
Sure, I will have much to improve, but this is just the beginning.