Facebook Prophet: Providing different data sets to build a better model - python

My data frame looks like that. My goal is to predict event_id 3 based on data of event_id 1 & event_id 2
ds tickets_sold y event_id
3/12/19 90 90 1
3/13/19 40 130 1
3/14/19 13 143 1
3/15/19 8 151 1
3/16/19 13 164 1
3/17/19 14 178 1
3/20/19 10 188 1
3/20/19 15 203 1
3/20/19 13 216 1
3/21/19 6 222 1
3/22/19 11 233 1
3/23/19 12 245 1
3/12/19 30 30 2
3/13/19 23 53 2
3/14/19 43 96 2
3/15/19 24 120 2
3/16/19 3 123 2
3/17/19 5 128 2
3/20/19 3 131 2
3/20/19 25 156 2
3/20/19 64 220 2
3/21/19 6 226 2
3/22/19 4 230 2
3/23/19 63 293 2
I want to predict sales for the next 10 days of that data:
ds tickets_sold y event_id
3/24/19 20 20 3
3/25/19 30 50 3
3/26/19 20 70 3
3/27/19 12 82 3
3/28/19 12 94 3
3/29/19 12 106 3
3/30/19 12 118 3
So far my model is that one. However, I am not telling the model that these are two separate events. However, it would be useful to consider all data from different events as they belong to the same organizer and therefore provide more information than just one event. Is that kind of fitting possible for Prophet?
# Load data
df = pd.read_csv('event_data_prophet.csv')
df.drop(columns=['tickets_sold'], inplace=True, axis=0)
df.head()
# The important things to note are that cap must be specified for every row in the dataframe,
# and that it does not have to be constant. If the market size is growing, then cap can be an increasing sequence.
df['cap'] = 500
# growth: String 'linear' or 'logistic' to specify a linear or logistic trend.
m = Prophet(growth='linear')
m.fit(df)
# periods is the amount of days that I look in the future
future = m.make_future_dataframe(periods=20)
future['cap'] = 500
future.tail()
forecast = m.predict(future)
forecast[['ds', 'yhat', 'yhat_lower', 'yhat_upper']].tail()
fig1 = m.plot(forecast)


Start dates of events seem to cause peaks. You can use holidays for this by setting the starting date of each event as a holiday. This informs prophet about the events (and their peaks). I noticed event 1 and 2 are overlapping. I think you have multiple options here to deal with this. You need to ask yourself what the predictive value of each event is related to event3. You don't have too much data, that will be the main issue. If they have equal value, you could change the date of one event. For example 11 days earlier. The unequal value scenario could mean you drop 1 event.
events = pd.DataFrame({
'holiday': 'events',
'ds': pd.to_datetime(['2019-03-24', '2019-03-12', '2019-03-01']),
'lower_window': 0,
'upper_window': 1,
})
m = Prophet(growth='linear', holidays=events)
m.fit(df)
Also I noticed you forecast on the cumsum. I think your events are stationary therefor prophet probably benefits from forecasting on the daily ticket sales rather than the cumsum.

Related

Numpy Vectorized Window Operations

I'm interested in figuring out how to do vectorized computations in a numpy array / pandas dataframe where each new cell is updated with local information.
For example, lets say I'm a weatherman interested in making predictions about the weather. My prediction algorithm will be the mean of the past 3 days. While this prediction is simple, I'd like to be able to do this with an arbitrary function.
Example data:
day temp
1 70
2 72
3 68
4 67
...
After a transformation should become
day temp prediction
1 70 None (no previous data)
2 72 70 (only one data point)
3 68 71 (two data points)
4 67 70
5 70 69
...
I'm only interested in the prediction column, so no need to make an attempt to join the data back together after achieving the prediction! Thanks!

Use rolling with a window of 3 and the min_periods of 1
df['prediction'] = df['temp'].rolling(window = 3, min_periods = 1).mean().shift()
df
day temp prediction
0 1 70 NaN
1 2 72 70
2 3 68 71
3 4 67 70
4 5 70 69

Calculate Multiple Column Growth in Python Dataframe

The data I used look like this
data
Subject 2000_X1 2000_X2 2001_X1 2001_X2 2002_X1 2002_X2
1 100 50 120 45 110 50
2 95 40 100 45 105 50
3 110 45 100 45 110 40
I want to calculate each variable growth for each year so the result will look like this
Subject 2001_X1_gro 2001_X2_gro 2002_X1_gro 2002_X2_gro
1 0.2 -0.1 -0.08333 0.11111
2 0.052632 0.125 0.05 0.11111
3 -0.09091 0 0.1 -0.11111
I already do it manually for each variable for each year with code like this
data[2001_X1_gro]= (data[2001_X1]-data[2000_X1])/data[2000_X1]
data[2002_X1_gro]= (data[2002_X1]-data[2001_X1])/data[2001_X1]
data[2001_X2_gro]= (data[2001_X2]-data[2000_X2])/data[2000_X2]
data[2002_X2_gro]= (data[2002_X2]-data[2001_X2])/data[2001_X2]
Is there a way to do it more efficient escpecially if I have more year and/or more variable?

import pandas as pd
df = pd.read_csv('data.txt', sep=',', header=0)
Input
Subject 2000_X1 2000_X2 2001_X1 2001_X2 2002_X1 2002_X2
0 1 100 50 120 45 110 50
1 2 95 40 100 45 105 50
2 3 110 45 100 45 110 40
Next, a loop is created and the columns are filled:
qqq = '_gro'
new_name = ''
year = ''
for i in range(1, len(df.columns) - 2):
year = str(int(df.columns[i][:4]) + 1) + df.columns[i][4:]
new_name = year + qqq
df[new_name] = (df[year] - df[df.columns[i]])/df[df.columns[i]]
print(df)
Output
Subject 2000_X1 2000_X2 2001_X1 2001_X2 2002_X1 2002_X2 2001_X1_gro \
0 1 100 50 120 45 110 50 0.200000
1 2 95 40 100 45 105 50 0.052632
2 3 110 45 100 45 110 40 -0.090909
2001_X2_gro 2002_X1_gro 2002_X2_gro
0 -0.100 -0.083333 0.111111
1 0.125 0.050000 0.111111
2 0.000 0.100000 -0.111111
In the loop, the year is extracted from the column name, converted to int, 1 is added to it. The value is again converted to a string, the prefix '_Xn' is added. A new_name variable is created, to which the string '_gro ' is also appended. A column is created and filled with calculated values.
If you want to count, for example, for three years, then you need to add not 1, but 3. This is with the condition that your data will be ordered. And note that the loop does not go through all the elements: for i in range(1, len(df.columns) - 2):. In this case, it skips the Subject column and stops short of the last two values. That is, you need to know where to stop it.

Rearranging data from a table in python to correlate columns

I got a table that look like this:
code
year
month
Value A
Value B
1
2020
1
120
100
1
2020
2
130
90
1
2020
3
90
89
1
2020
4
67
65
...
...
...
...
...
100
2020
10
90
90
100
2020
11
115
100
100
2020
12
150
135
I would like to know if there's a way to rearrange the data to find the correlation between A and B for every distinct code.
What I'm thinking is, for example, getting an array for every code, like:
[(A1,A2,A3...,A12),(B1,B2,B3...,B12)]
where A and B is the values for the respective month, and then I could see the correlation between these two columns. Is there a way to make this dynamic?

IIUC, you don't need to re-arrange to get the correlation for each "code". Instead, try with groupby:
>>> df.groupby("code").apply(lambda x: x["Value A"].corr(x["Value B"]))
code
1 0.830163
100 0.977093
dtype: float64

Pandas: calculating mean value of multiple columns using datetime and Grouper removes columns or doesn't return correct Dataframe

As part of a larger task, I want to calculate the monthly mean values for each specific station. This is already difficult to do, but I am getting close.
The dataframe has many columns, but ultimately I only use the following information:
Date Value Station_Name
0 2006-01-03 18 2
1 2006-01-04 12 2
2 2006-01-05 11 2
3 2006-01-06 10 2
4 2006-01-09 22 2
... ... ...
3510 2006-12-23 47 45
3511 2006-12-24 46 45
3512 2006-12-26 35 45
3513 2006-12-27 35 45
3514 2006-12-30 28 45
I am running into two issues, using:
df.groupby(['Station_Name', pd.Grouper(freq='M')])['Value'].mean()
It results in something like:
Station_Name Date
2 2003-01-31 29.448387
2003-02-28 30.617857
2003-03-31 28.758065
2003-04-30 28.392593
2003-05-31 30.318519
...
45 2003-09-30 16.160000
2003-10-31 18.906452
2003-11-30 26.296667
2003-12-31 30.306667
2004-01-31 29.330000
Which I can't seem to use as a regular dataframe, and the datetime is messed up as it doesn't show the monthly mean but gives the last day back. Also the station name is a single index, and not for the whole column. Plus the mean value doesn't have a "column name" at all. This isn't a dataframe, but a pandas.core.series.Series. I can't convert this again because it's not correct, and using the .to_frame() method shows that it is still indeed a Dataframe. I don't get this part.
I found that in order to return a normal dataframe, to use
as_index = False
In the groupby method. But this results in the months not being shown:
df.groupby(['station_name', pd.Grouper(freq='M')], as_index = False)['Value'].mean()
Gives:
Station_Name Value
0 2 29.448387
1 2 30.617857
2 2 28.758065
3 2 28.392593
4 2 30.318519
... ... ...
142 45 16.160000
143 45 18.906452
144 45 26.296667
145 45 30.306667
146 45 29.330000
I can't just simply add the month later, as not every station has an observation in every month.
I've tried using other methods, such as
df.resample("M").mean()
But it doesn't seem possible to do this on multiple columns. It returns the mean value of everything.
Edit: This is ultimately what I would want.
Station_Name Date Value
0 2 2003-01 29.448387
1 2 2003-02 30.617857
2 2 2003-03 28.758065
3 2 2003-04 28.392593
4 2 2003-05 30.318519
... ... ...
142 45 2003-08 16.160000
143 45 2003-09 18.906452
144 45 2003-10 26.296667
145 45 2003-11 30.306667
146 45 2003-12 29.330000

ok , how baout this :
df = df.groupby(['Station_Name',df['Date'].dt.to_period('M')])['Value'].mean().reset_index()
outut:
>>
Station_Name Date Value
0 2 2006-01 14.6
1 45 2006-12 38.2

Pandas groupby two columns and only keep records satisfying condition based on count

Trying to filter out a number of actions a user has done if the number of actions reaches a threshold.
Here is the data set: (Only Few records)
user_id,session_id,item_id,rating,length,time
123,36,28,3.5,6243.0,2015-03-07 22:44:40
123,36,29,2.5,4884.0,2015-03-07 22:44:14
123,36,30,3.5,6846.0,2015-03-07 22:44:28
123,36,54,6.5,10281.0,2015-03-07 22:43:56
123,36,61,3.5,7639.0,2015-03-07 22:43:44
123,36,62,7.5,18640.0,2015-03-07 22:43:34
123,36,63,8.5,7189.0,2015-03-07 22:44:06
123,36,97,2.5,7627.0,2015-03-07 22:42:53
123,36,98,4.5,9000.0,2015-03-07 22:43:04
123,36,99,7.5,7514.0,2015-03-07 22:43:13
223,63,30,8.0,5412.0,2015-03-22 01:42:10
123,36,30,5.5,8046.0,2015-03-07 22:42:05
223,63,32,8.5,4872.0,2015-03-22 01:42:03
123,36,32,7.5,11914.0,2015-03-07 22:41:54
225,63,35,7.5,6491.0,2015-03-22 01:42:19
123,36,35,5.5,7202.0,2015-03-07 22:42:15
123,36,36,6.5,6806.0,2015-03-07 22:42:43
123,36,37,2.5,6810.0,2015-03-07 22:42:34
225,63,41,5.0,15026.0,2015-03-22 01:42:37
225,63,45,6.5,8532.0,2015-03-07 22:42:25
I can groupby the data using user_id and session_id and get a count of items a user has rated in a session:
df.groupby(['user_id', 'session_id']).agg({'item_id':'count'}).rename(columns={'item_id': 'count'})
List of items that user has rated in a session can be obtained:
df.groupby(['user_id','session_id'])['item_id'].apply(list)
The goal is to get following if a user has rated more than 3 items in session, I want to pick only the first three items (keep only first three per user per session) from the original data frame. Maybe use the time to sort the items?
First tried to obtain which sessions contain more than 3, somewhat struggling to go beyond.
df.groupby(['user_id', 'session_id'])['item_id'].apply(
lambda x: (x > 3).count())
Example: from original df, user 123 should have first three records belong to session 36

It seems like you want to use groupby with head:
In [8]: df.groupby([df.user_id, df.session_id]).head(3)
Out[8]:
user_id session_id item_id rating length time
0 123 36 28 3.5 6243.0 2015-03-07 22:44:40
1 123 36 29 2.5 4884.0 2015-03-07 22:44:14
2 123 36 30 3.5 6846.0 2015-03-07 22:44:28
10 223 63 30 8.0 5412.0 2015-03-22 01:42:10
12 223 63 32 8.5 4872.0 2015-03-22 01:42:03
14 225 63 35 7.5 6491.0 2015-03-22 01:42:19
18 225 63 41 5.0 15026.0 2015-03-22 01:42:37
19 225 63 45 6.5 8532.0 2015-03-07 22:42:25

One way is to use sort_values followed by groupby.cumcount. A method I find useful is to extract any series or MultiIndex data before applying any filtering.
The below example filters for minimum user_id / session_id combination of 3 items and only takes the first 3 in each group.
sizes = df.groupby(['user_id', 'session_id']).size()
counter = df.groupby(['user_id', 'session_id']).cumcount() + 1 # counting begins at 0
indices = df.set_index(['user_id', 'session_id']).index
df = df.sort_values('time')
res = df[(indices.map(sizes.get) >= 3) & (counter <=3)]
print(res)
user_id session_id item_id rating length time
0 123 36 28 3.5 6243.0 2015-03-07 22:44:40
1 123 36 29 2.5 4884.0 2015-03-07 22:44:14
2 123 36 30 3.5 6846.0 2015-03-07 22:44:28
14 225 63 35 7.5 6491.0 2015-03-22 01:42:19
18 225 63 41 5.0 15026.0 2015-03-22 01:42:37
19 225 63 45 6.5 8532.0 2015-03-07 22:42:25

Categories

Resources