Welcome to the second part of the post. In this half, we will be implementing RandomForest regressor on our own dataset. For that first, you will need to have a dataset. The one that I used is a weather dataset from my city (Kathmandu) which is available at this link.
Once you download the dataset, you can see that there are only four columns in the dataset which are ‘DATE’, ‘TAVG’, ‘TMIN’, and ‘TMAX’. This dataset concludes the min, max, and avg temperature for each day from Jan-2000 to Jan-2019. For our simplicity, let’s mark the max. temperature (TMAX) as our label for the dataset, and the rest will be the features.
Now, open your jupyter notebook or any python IDE/Editor and import numpy and pandas.
1
2
import numpy as np
import pandas as pd
Use pandas to read the downloaded dataset. Make sure you put the dataset in an appropriate location for the import
1
2
3
4
5
df_raw = pd.read_csv('path/to/ktm_temps.csv',
parse_dates=['DATE'])
# Checkout the dataset
print(df_raw.head())
print(df_raw.shape)
If you look closely on the csv file, you can see a lot of values from the columns are missing. So, we need to do something for the missing values. There are certainly some good techniques to deal with missing values when training for RandomForest, but for now we will just replace the values with the mean of their column. For that, just type in the following command.
1
df_raw.fillna(df_raw.mean(), inplace=True)
Now checkout the dataframe, you will see that all NaN values are gone for good.
Feature Extraction from our dataset
We can see that we only have three feature columns in our dataset. In machine learning, it is generally seen that greater the number of valuable features, better the model performs. We can see that there is a ‘DATE’ column in our dataset. We can extract this feature to add some more insights to our data. For example, instead of only looking at the date, we can look for the day of the week or the month of the year or the day of the year. These extracted features from one variable can make our model perform better. So, this is exactly what we are going to do in the next step.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
# Take the 'DATE' column
date_column = df_raw.iloc[:, 0]
# Create a new dataframe with additional features
features = pd.DataFrame({
'year': date_column.dt.year,
'month': date_column.dt.month,
'dayofyear': date_column.dt.dayofyear,
'weekofyear': date_column.dt.weekofyear,
'dayofweek': date_column.dt.dayofweek,
})
# Add the original two features to the new dataframe
features['TAVG'] = df_raw['TAVG']
features['TMIN'] = df_raw['TMIN']
# Check out the newly formed dataset
features.head()
Since we decided the max. temperature of the day to be our label, we need to extract the labels from the dataset. We also convert both features and labels into numpy arrays for further processing.
1
2
3
labels = df_raw['TMAX']
labels = np.array(labels)
features = np.array(features)
Now, we have our dataset clean and ready. It is now time to split the dataset into training and test sets. For this we will use an inbuilt method from sklearn called train_test_split
1
2
3
4
5
6
7
from sklearn.model_selection import train_test_split
# Split the data into training and testing sets
train_features, test_features, train_labels, test_labels = train_test_split(features,
labels,
test_size=0.3,
random_state=42)
Don’t forget to checkout the shapes of and behavior of the splitted sets.
Training the model
After having a separate training and testing set, we are now ready to train our RandomForestRegressor model. For this, we will import the model from sklearn.ensemble and fit the model with the splitted training set.
1
2
3
4
5
6
7
8
9
from sklearn.ensemble import RandomForestRegressor
# Instantiate the model
# n_estimators refers to the number of crappy trees for the forest
rf = RandomForestRegressor(n_estimators=400,
random_state=42)
# Train the model
rf.fit(train_features, train_labels)
Once the model is trained successfully, it is now time to evaluate our model and see how it performs on our test set.
1
2
3
4
5
6
7
8
predictions = rf.predict(test_features)
# Calculate the absolute errors
errors = abs(predictions - test_labels)
# Mean abs error
mean_abs_error = round(np.mean(errors), 2)
print(mean_abs_error)
1
2
3
4
5
6
# Mean Absolute Percentage Error (MAPE)
mape = 100 * (errors / test_labels)
# Accuracy
accuracy = 100 - np.mean(mape)
print(accuracy)
In my case, the accuracy obtained for the dataset was around 96%. We can see that RandomForest performed really well on our dataset. You can play with the hyperparameters to tune it and see how it performs on other values. You can even use a different dataset to analyse the same model and its metrics.
You can find the complete notebook code from here.
This concludes the second and final part of the post. Hope that you got to learn something from this post. If you have any suggestions or queries, please feel free to comment below. I would really appreciate your feedback, and as always thanks for reading.