Using logistic regression to predict the parameter value

0
votes

I have written vary basic sklearn code using logistic regression to predict the value.

Training data looks like -

https://gist.github.com/anonymous/563591e0395e8d988277d3ce63d7438f

date    hr_of_day   vals
01/05/2014  9   929
01/05/2014  10  942
01/05/2014  11  968
01/05/2014  12  856
01/05/2014  13  835
01/05/2014  14  885
01/05/2014  15  945
01/05/2014  16  924
01/05/2014  17  914
01/05/2014  18  744
01/05/2014  19  377
01/05/2014  20  219
01/05/2014  21  106

and I have selected first 8 items from training data to just validate the classifier which is

I want to predict the value of vals, in testing data, I have put it as 0. Is that correct?

date    hr_of_day   vals
2014-05-01  0   0
2014-05-01  1   0
2014-05-01  2   0
2014-05-01  3   0
2014-05-01  4   0
2014-05-01  5   0
2014-05-01  6   0
2014-05-01  7   0

My model code, works fine. But my result looks trange. I was expecting value of vals in result. Rather then that, I am getting large matrix with all element value as 0.00030676.

I appreciate if someone can give details or help me to play better with this result.

import pandas as pd
from sklearn import datasets
from sklearn import metrics
from sklearn.linear_model import LogisticRegression
from datetime import datetime, date, timedelta
Train = pd.read_csv("data_scientist_assignment.tsv", sep='\t', parse_dates=['date'])
Train['timestamp'] = Train.date.values.astype(pd.np.int64)
x1=["timestamp", "hr_of_day"]
test=pd.read_csv("test.tsv", sep='\t', parse_dates=['date'])
test['timestamp'] = test.date.values.astype(pd.np.int64)
print(Train.columns)
print(test.columns)
model = LogisticRegression()
model.fit(Train[x1], Train["vals"])
print(model)
print model.score(Train[x1], Train["vals"])
print model.predict_proba(test[x1])

results looks like this:

In [92]: print(model)
LogisticRegression(C=1.0, class_weight=None, dual=False, fit_intercept=True,
          intercept_scaling=1, max_iter=100, multi_class='ovr', n_jobs=1,
          penalty='l2', random_state=None, solver='liblinear', tol=0.0001,
          verbose=0, warm_start=False)

In [93]: print model.score(Train[x1], Train["vals"])
0.00520833333333

In [94]: 

In [94]: print model.predict_proba(test[x1])
[[ 0.00030676  0.00030676  0.00030676 ...,  0.00030889  0.00030885
   0.00030902]
 [ 0.00030676  0.00030676  0.00030676 ...,  0.00030889  0.00030885
   0.00030902]
 [ 0.00030676  0.00030676  0.00030676 ...,  0.00030889  0.00030885
   0.00030902]
 ..., 
 [ 0.00030676  0.00030676  0.00030676 ...,  0.00030889  0.00030885
   0.00030902]
 [ 0.00030676  0.00030676  0.00030676 ...,  0.00030889  0.00030885
   0.00030902]
 [ 0.00030676  0.00030676  0.00030676 ...,  0.00030889  0.00030885
   0.00030902]]
2
pythonmachine-learningscikit-learnclassificationlogistic-regression
What is it that you don't understand? "I don't understand" doesn't make a question. - Julien
@JulienBernu: edited the question. Actually in result it should predict the value of vals rather then that, I am getting large matrix with all values set to 0.00030676 - user123
You are using predict_proba, that is why you are getting prediction probability. Instead try using model.predict(). - Ishank Gulati
1) your problem looks more like regression to me, but you are using Logistic Regression (which, despite its name, it is a classifier), so you might be using wrong techniques (i might be wrong tho) and 2) as pointed out before, replace predict_proba with just predict. - Imanol Luengo

2 Answers

3
votes
  1. You are using predict_proba() which gives class probabilities, instead of that you should use predict() function.
  2. You are using a wrong model. The target variable in your data has continuous data, therefore you will have to use linear regression. Logistic Regression actually works as a classifier and classification tasks require discrete data(as in fixed number of labels).
1
votes

Use following code to get predicted labels:

predicted_labels= model.predict(test[x1])

Also try following example to understand logistic regression in sklearn:

# Logistic Regression
from sklearn import datasets
from sklearn import metrics
from sklearn.linear_model import LogisticRegression
# load the iris datasets
dataset = datasets.load_iris()
# fit a logistic regression model to the data
model = LogisticRegression()
model.fit(dataset.data, dataset.target)
print(model)
# make predictions
expected = dataset.target
predicted = model.predict(dataset.data)
# summarize the fit of the model
print(metrics.classification_report(expected, predicted))
print(metrics.confusion_matrix(expected, predicted))

Example source: http://machinelearningmastery.com/get-your-hands-dirty-with-scikit-learn-now/