2

I'm stuck with a simple DataFrame.reindex().interpolate() problem because the dataframes I am using don't have a datetime index.

I have DataFrame1: t that looks like this:

In[1]: import pandas as pd
t = pd.DataFrame({'D18O': [-0.47, -0.12,  0.55,  0.72,  1.8 ,  1.1 ,  0.43, -0.29, -0.55,
       -0.6 , -0.32,  0.28,  0.72,  1.1 ,  1.34,  1.32,  1.11,  0.46,
        0.09,  0.02]})

Out[2]: 
1    -0.47
2    -0.12
3     0.55
4     0.72
5     1.80
6     1.10
7     0.43
8    -0.29
9    -0.55
10   -0.60
11   -0.32
12    0.28
13    0.72
14    1.10
15    1.34
16    1.32
17    1.11
18    0.46
19    0.09
20    0.02
Name: D18O, dtype: float64

I want to "stretch" it to 430 rows by evenly spacing each row and linearly interpolating values in between. This is because my DataFrame2: env has 430 rows and I want to do some later analysis that needs both frames to have the same dimension. 

In[2]: env.index
Out[49]: RangeIndex(start=0, stop=430, step=1)

I've tried reindexing and interpolating in many combinations but just can't find the right method. I think the problem is, that 430 is not evenly divisable by 19/20.

new_idx = np.linspace(t.index[0], t.index[-1], env.shape[0])
t.reindex(new_idx).interpolate()

I thought this qould work, but because the indexes are not even it skips most of the values in t and leaves me with a nearly empty new dataframe.

For the reindexing step I expect something like:

In[3]: t['D18O']
Out[3]: 
0          0.47
2.13157     NaN
2.26315     NaN
...         ...
21.5      -0.12
22.63157    NaN
23.76315    NaN
...         ...
...         ...
430        0.02
Name: D18O, dtype: float64

The indexes don't really matter, as long as the values are evenly spaced and the number of rows matches the number of rows in env.

flurble
  • 1,086
  • 7
  • 21

2 Answers2

1

You can use parameter ffill with limit in DataFrame.reindex, but thre is problem with duplicates first value, so possible solution is add first helper value close 0 to index, reindex, remove it by iloc and last interpolate:

r = pd.RangeIndex(0, 430, 1)

t.loc[-0.001] = 0
t = t.sort_index()
new_idx = np.linspace(t.index[0], t.index[-1], len(r))
print (t.reindex(new_idx, method='ffill', limit=1).iloc[1:].interpolate())

               D18O
0.043291  -0.470000
0.087583  -0.454091
0.131874  -0.438182
0.176166  -0.422273
0.220457  -0.406364
0.264748  -0.390455
0.309040  -0.374545
0.353331  -0.358636
0.397622  -0.342727
0.441914  -0.326818
0.486205  -0.310909
0.530497  -0.295000
0.574788  -0.279091
0.619079  -0.263182
0.663371  -0.247273
0.707662  -0.231364
0.751953  -0.215455
...
...
jezrael
  • 822,522
  • 95
  • 1,334
  • 1,252
  • 1
    This works perfectly. If I plot the original and interpolated `t` they show the exact same graph. Nonetheless, I don't really understand the relationship between inserting the index, forward filling and excluding the first value. I thought `method='ffill'` would fill every value between two entries? – flurble Apr 18 '19 at 11:18
  • @cripcate - Exactly, if remove `limit=1` it replace all `NaN`s values between, limit i used for only first one NaN replacement. – jezrael Apr 18 '19 at 11:21
  • Ah okay, so it only fills one value. I still don't really get why the helper value is needed though, sorry. And shouldn't I do `t.loc[-0.0001] = t[0]` instead of `t.loc[-0.0001] = t[0]`? Sorry, bit confused here. – flurble Apr 18 '19 at 11:32
  • @cripcate - I think it working same if `t[0]` or `0`, because this value is not used for interpolate. Test it by removeing `t.loc[-0.001] = 0` and `t.reindex(new_idx, method='ffill', limit=1).interpolate()` – jezrael Apr 18 '19 at 11:35
  • for me this works perfectly, but only when I dismiss inserting the helper value – tbrodbeck Apr 03 '22 at 01:19
0

I used a more general way of interpolating data to a certain index now. I just wanna list my approach for future references:

import numpy as np
import pandas as pd 
from scipy.interpolate import interp1d

# Example data 5 numeric columns
i = pd.RangeIndex(0, 430, 1)
df1 = pd.DataFrame([-0.47, -0.12, 0.55, 0.72, 1.8, 1.1, 0.43, -0.29, 
                    -0.55, -0.6, -0.32, 0.28, 0.72, 1.1 , 1.34, 1.32,
                    1.11, 0.46, 0.09, 0.02], [-0.47, -0.12, 0.55, 0.72, 1.8, 1.1, 0.43, -0.29, 
                    -0.55, -0.6, -0.32, 0.28, 0.72, 1.1 , 1.34, 1.32,
                    1.11, 0.46, 0.09, 0.02], [-0.47, -0.12, 0.55, 0.72, 1.8, 1.1, 0.43, -0.29, 
                    -0.55, -0.6, -0.32, 0.28, 0.72, 1.1 , 1.34, 1.32,
                    1.11, 0.46, 0.09, 0.02])

# Select numeric columns
nums = df1.select_dtypes([np.number])
old_idx = df.index
# Calculate new index
len_idx = env.shape[0]
mi, ma = old_idx.min(), old_idx.max()
new_idx = np.linspace(mi, ma, len_idx)

# Plot to compare interpolation to original values
fig, ax = plt.subplots(1, 1)
ax.plot(old_idx, df1.iloc[:, 0], 'k--')

def interpol(column):
    ```Interpolation function```    
    interpolant = interp1d(old_idx, column)
    interpolated = interpolant(new_idx)
    return interpolated

# Interpolate data to match index length of enviromental data
inter_nums = pd.DataFrame(index=new_idx)
for col in nums:
    inter = interpol(nums[col])
    inter_nums[col] = inter

# Plot after interpolation. Same curve? good!      
ax.plot(inter_nums_iloc[:; 0], c='r')
flurble
  • 1,086
  • 7
  • 21