Login
main >   complex_prediction >  


ML to Predict Formulas:

Still in progress

Four labels:

  1. more: Structured: Predict equation from text formula. (important)
  2. less: Noise: Leave Nonsense. (won't equate)
  3. same: Label well trained: Existing equation. (stronger)
  4. unknown: Unstructured: Not sure (unstructured)

Features are words / formulas. Can only add a feature if you know how it applies to the other features.

  1. Good
  2. Bad
  3. Amount
  4. Frequency
  5. Distance
  6. Direction
  7. Time
  8. ...

First training formulas are obvious

import numpy as np

dataset = {}
dataset['target_name'] = np.array(['More', 'Less', 'Same', 'Unknown'])
# which means targets are identified by positions 0, 1, 2, 3

#Currently has 3 features, add more features when value is able to apply to other features
dataset['feature_name'] = np.array(['Good', 'Bad', 'Amount'])
dataset['feature_value'] = np.array([
                            [9999**9, -9999**9, 9999**9], # going to be target more thus 0
                            [-9999**9, 9999**9, 0], # going to be target less thus 1
                            [9999**9, 9999**9, 9999**9], # going to be target same thus 2
                            [1, -1, 1], # going to be target unknown thus 3
])

#So feature_values determined the following targets
dataset['target'] = np.array([0, 1, 2, 3])


################
#Now lets train a model
from sklearn.neighbors import KNeighborsClassifier
knn = KNeighborsClassifier(n_neighbors=1)

knn.fit(dataset['feature_value'], dataset['target'])

###############
#Now lets predict
#10 Good, 0 Bad, 1 Amount
what_is = np.array([[10, 0, 1]])

prediction = knn.predict(what_is)

print(dataset['target_name'][prediction])
prediction.predict_proba(what_is)
#Awnser is: Unknown
#Certainty: [[0. 0. 0. 1.]]

hidden1

hidden2