Generate a new data set

[ofithcheallaigh]

So far I have been training using separate data sets.

Now, I will generate a complete large data set. This will be build from all the balanced data sets, so the large data set will also be balanced.

Column headers will stay the same, there will be no need to reference the obstacle.

Todo

• [x] Think about best way to generate new data sets that will contain data for 3 objects and the no object data set, but rotating the left out data set.
• [x] Implement strategy and generate data sets.

[ofithcheallaigh]

Rather than generating new data sets in .csv files, I am doing this in code. This allows me some amount of flexibility. First, the data is read in, in the normal way:

# Read in the data
dataset1 = pd.read_csv("MATLAB\Data\grid0_closeddoor_clearhallway.csv")
dataset2 = pd.read_csv("MATLAB\Data\grid0_displaystand_clearhallway.csv")
dataset3 = pd.read_csv("MATLAB\Data\grid0_largebin_clearhallway.csv")
dataset4 = pd.read_csv("MATLAB\Data\grid0_storagebox_clearhallway.csv")

From there, I can select the permeation I want:

data = np.vstack((dataset2,dataset3,dataset4))
new_data = ((dataset1))

The new_data is data that the model will not be trained on.

This can be changed manually for each test.

This doesn't need to be automated, but could be something to look at later, if there is time.