熊猫：建立价格差异矩阵？

Question

I am trying to build price differences between bitcoins and exchanges, for example I have a dataframe,

    Exchange coin           lastUpdate    price   volume
0   Bitfinex  BTC  2019-06-23 06:23:27    10646  24299.4
1   Bitfinex  ETH  2019-06-23 06:23:13   308.47   225945
2   Bitfinex  LTC  2019-06-23 06:23:18   140.41   215698
3   Bitstamp  BTC  2019-06-23 06:23:21  10546.4  9620.04
4   Bitstamp  ETH  2019-06-23 06:22:48   305.15  46062.6
5   Bitstamp  LTC  2019-06-23 06:22:46   139.22  85160.5
6     CCCAGG  BTC  2019-06-23 06:23:23  10580.4  79049.8
7     CCCAGG  ETH  2019-06-23 06:23:20   306.74   681056
8     CCCAGG  LTC  2019-06-23 06:23:24   139.71   752875
9   Coinbase  BTC  2019-06-23 06:23:17  10557.5  23731.2
10  Coinbase  ETH  2019-06-23 06:23:11   306.09   247213
11  Coinbase  LTC  2019-06-23 06:23:13   139.49   381421

And I am trying to get all of the prices differences between the coin and all the exchanges it trades on,

I want it to look like,

price_combos                        diff
Price Diff: BTC - Bitfinex-Bitstamp 14.06
Price Diff: BTC - Bitfinex-CCCAGG   14.32
Price Diff: BTC - Bitstamp-CCCAGG   0.26
Price Diff: BTC - Coinbase-Bitfinex -17.99
Price Diff: BTC - Coinbase-Bitstamp -3.93
Price Diff: BTC - Coinbase-CCCAGG   -3.67

And then repeat for each coin.

Edit: Added price to combinations, note that the diff is from a different set of data so it won't match the actual diff from the first dataframe.

Answer 1

We can approach this problem as following:

1. We do an outer merge on each coin itself so it gives us all the combinations back.
2. We filter out the rows with ne (not equal) where the exchange is the same (we don't want to compare those).
3. Create our Price diff column by subtracting the prices

# Step 1 outer merge
df2 = df[['Exchange', 'coin', 'price']].merge(df[['Exchange', 'coin', 'price']], 
                                              on='coin', 
                                              how='outer', 
                                              suffixes=['', '_2'])

# Step 2 filter out same exchange
df2 = df2[df2['Exchange'].ne(df2['Exchange_2'])]

# Step 3 create Price Diff column
df2['Price Diff'] = df2['price'] = df2['price_2']

---

    Exchange coin     price Exchange_2   price_2  Price Diff
1   Bitfinex  BTC  10546.40   Bitstamp  10546.40    10546.40
2   Bitfinex  BTC  10580.40     CCCAGG  10580.40    10580.40
3   Bitfinex  BTC  10557.50   Coinbase  10557.50    10557.50
4   Bitstamp  BTC  10646.00   Bitfinex  10646.00    10646.00
6   Bitstamp  BTC  10580.40     CCCAGG  10580.40    10580.40
7   Bitstamp  BTC  10557.50   Coinbase  10557.50    10557.50
8     CCCAGG  BTC  10646.00   Bitfinex  10646.00    10646.00
9     CCCAGG  BTC  10546.40   Bitstamp  10546.40    10546.40
11    CCCAGG  BTC  10557.50   Coinbase  10557.50    10557.50
12  Coinbase  BTC  10646.00   Bitfinex  10646.00    10646.00
13  Coinbase  BTC  10546.40   Bitstamp  10546.40    10546.40
14  Coinbase  BTC  10580.40     CCCAGG  10580.40    10580.40
17  Bitfinex  ETH    305.15   Bitstamp    305.15      305.15
18  Bitfinex  ETH    306.74     CCCAGG    306.74      306.74
19  Bitfinex  ETH    306.09   Coinbase    306.09      306.09
20  Bitstamp  ETH    308.47   Bitfinex    308.47      308.47
22  Bitstamp  ETH    306.74     CCCAGG    306.74      306.74
23  Bitstamp  ETH    306.09   Coinbase    306.09      306.09
24    CCCAGG  ETH    308.47   Bitfinex    308.47      308.47
25    CCCAGG  ETH    305.15   Bitstamp    305.15      305.15
27    CCCAGG  ETH    306.09   Coinbase    306.09      306.09
28  Coinbase  ETH    308.47   Bitfinex    308.47      308.47
29  Coinbase  ETH    305.15   Bitstamp    305.15      305.15
30  Coinbase  ETH    306.74     CCCAGG    306.74      306.74
33  Bitfinex  LTC    139.22   Bitstamp    139.22      139.22
34  Bitfinex  LTC    139.71     CCCAGG    139.71      139.71
35  Bitfinex  LTC    139.49   Coinbase    139.49      139.49
36  Bitstamp  LTC    140.41   Bitfinex    140.41      140.41
38  Bitstamp  LTC    139.71     CCCAGG    139.71      139.71
39  Bitstamp  LTC    139.49   Coinbase    139.49      139.49
40    CCCAGG  LTC    140.41   Bitfinex    140.41      140.41
41    CCCAGG  LTC    139.22   Bitstamp    139.22      139.22
43    CCCAGG  LTC    139.49   Coinbase    139.49      139.49
44  Coinbase  LTC    140.41   Bitfinex    140.41      140.41
45  Coinbase  LTC    139.22   Bitstamp    139.22      139.22
46  Coinbase  LTC    139.71     CCCAGG    139.71      139.71

Answer 2

You should have a look at the itertools module (doc) . There are a lot of nice functions for iterations.

Here you're exactly looking for the combination function.

Once you have the combinations, that becomes simple:

# Import modules
import pandas as pd
import itertools as iter

# Your data
df = pd.DataFrame([
    ["Bitfinex",  "BTC", "2019-06-23 06:23:27",  10646, 24299.4],
    ["Bitfinex",  "ETH", "2019-06-23 06:23:13",  308.47,  225945],
    ["Bitfinex",  "LTC", "2019-06-23 06:23:18",  140.41,  215698],
    ["Bitstamp",  "BTC", "2019-06-23 06:23:21", 10546.4, 9620.04],
    ["Bitstamp",  "ETH", "2019-06-23 06:22:48",  305.15, 46062.6],
    ["Bitstamp", "LTC", "2019-06-23 06:22:46", 139.22, 85160.5],
    ["CCCAGG",  "BTC", "2019-06-23 06:23:23", 10580.4, 79049.8],
    ["CCCAGG",  "ETH", "2019-06-23 06:23:20", 306.74,  681056],
    ["CCCAGG",  "LTC", "2019-06-23 06:23:24", 139.71, 752875],
    ["Coinbase",  "BTC", "2019-06-23 06:23:17", 10557.5, 23731.2],
    ["Coinbase", "ETH", "2019-06-23 06:23:11", 306.09, 247213],
    ["Coinbase", "LTC", "2019-06-23 06:23:13", 139.49,  381421],
], columns=["Exchange", "coin", "lastUpdate", "price", "volume"])


# Print all combinations for one coin
def print_combi(df, coin):
    # subset dataframe with matching rows
    sub_df = df[df["coin"] == coin]
    # Create all combinations for the exchange columns
    list_combi = [cb for cb in iter.combinations(sub_df.Exchange, 2)]

    # Print the expected output
    for combi in list_combi:
        print("Price diff: {0} - {1}-{2}".format(coin, combi[0], combi[1]))

print_combi(df, 'BTC')
# Price diff: BTC - Bitfinex-Bitstamp
# Price diff: BTC - Bitfinex-CCCAGG
# Price diff: BTC - Bitfinex-Coinbase
# Price diff: BTC - Bitstamp-CCCAGG
# Price diff: BTC - Bitstamp-Coinbase
# Price diff: BTC - CCCAGG-Coinbase

EDIT1:

Return a dataframe. The diff column is from the data used in the snippet above.

def combo_money_df(df, coin):
    # subset the dataframe
    sub_df = df[df["coin"] == coin]

    new_data = []
    # For each subset
    for combi in iter.combinations(sub_df.index, 2):
        # Select corresponding row
        row_1 = sub_df.loc[combi[0]]
        row_2 = sub_df.loc[combi[1]]
        # Create new rows
        new_data.append([row_1.Exchange + "-" + row_2.Exchange, row_1.price - row_2.price])
    # Return a dataframe object
    return pd.DataFrame(new_data, columns=["price_combo", "diff"])

print(combo_money_df(df, "BTC"))
#          price_combo  diff
# 0  Bitfinex-Bitstamp  99.6
# 1    Bitfinex-CCCAGG  65.6
# 2  Bitfinex-Coinbase  88.5
# 3    Bitstamp-CCCAGG -34.0
# 4  Bitstamp-Coinbase -11.1
# 5    CCCAGG-Coinbase  22.9