Drinking Prediction Model

How to change drinking prediction model of this project

We can do this to improve the performance of our drinking prediction model.

First of all, you only need to edit '/Project/Python Server/Python Server/train.py'

And, By looking at the items below, you can easily modify your drinking prediction code.

How to modify the data load location

• A function related to this is data_load().
• You can change the path of search("../../../Data", paths)
• And this project recognizes the labeling information using words in the file name.
• Therefore, you can extend to more class prediection by modifying path.find("class")

def data_load():
    paths = []
    search("../../../Data", paths)

    for path in paths:
        if (path.find("drink") != -1):
            read("drink", path)
        elif (path.find("exercise") != -1 or path.find("work out") != -1):
            read("workout", path)
        else:
            read("nothing", path)

How to delete the data clipping function

• Data clipping is commonly used to solve problems with insufficient datasets.
• However, this process can cause overfitting problems.
• So if you want to disable the clipping function, delete the code below.

In function read(htype, name)

   ...
 
    for i in range(0,1000,10):
        if (i > max_time - 15):
            break;
        for j in range(15,121,15):
            try:
                print("{} ({}~{})".format(name,i , i+j))
                data[htype].append({'name':"{} ({}~{})".format(name,i , i+j), 'data': create_clip(temp, i, i + j)})
            except:
                print("Error")
            if (i + j > max_time):
                break;

The original dataset is loaded by this.

data[htype].append({'name':name, 'data': create_clip(temp, 0, max_time)})

How to change the prediction model

• This is implemented in the train() function.
• If the model you want to change follows sklearn's interface, simply change the code below.

model = DecisionTreeClassifier(max_depth=7)

• But you can also take advantage of new libraries like TensorFlow.
  • In that case, you need to build a new model in that area.
  • In this process, you need a new preprocessing code for your model!!
  • These models should also be modified with the predict() code.
  • In the code below, write your prediction code according to the model you have implemented.

def predict(data):
    global min_max_scaler, model
    x = GetFeature(None, data);
    if (x is None):
        return "Need more data"
    x_min_max_scaled = min_max_scaler.transform([x])
    return model.predict(x_min_max_scaled)[0]