How to merge two CSV files by value in column using pandas PYTHON

问题

I have 2 csv files price and performance.

Here is the data layout of each

Price:

Performance:

I import them into python using:

import pandas as pd

price = pd.read_csv("cpu.csv")
performance = pd.read_csv("geekbench.csv")

This works as intended, however I am unsure on how to create a new csv file with matches between Price[brand + model] and Performance[name]

I want to take:

Cores, tdp and price from Price
Score, multicore_score and name from Performance

Create a new csv file using these parameters above. Problems I've been having a finding a good way to match which ignores minor differences such as capitalization I was looking into algorithms such as fuzzy string matching but was not sure what the best option is.

This is my current attempt which throws errors;

for i in range(len(price.index)):
    brand = (price.iloc[i, 0])
    model = (price.iloc[i, 1])
    print(model)
    print(performance)
    print(performance.query('name == brand+model'))

Thanks

回答1:

I suggest the following :

import nltk
import pandas as pd
tokenizer = nltk.RegexpTokenizer(r'\w+')
price = pd.DataFrame({"brand": ["AMD", "AMD", "AMD", "AMD"],
                      "model" : ["2650", "3800", "5150", "4200"],
                      "cores" : [2,4,4,4],
                      "tdp" : [25,25,25,25]})
performance = pd.DataFrame({"name": ["AMD Athlon 64 3200+",
                                     "AMD Athlon 64 X2 3800+",
                                     "AMD Athlon 64 X2 4000+",
                                     "AMD Athlon 64 X2 4200+"],
                            "score" : [6,5,6,18]})
# I break down the name in performance and suppress capital letters
performance["tokens"] = (performance["name"].str.lower()
                         .apply(tokenizer.tokenize))
# And the same for price
price["tokens"] = price.loc[:,"brand"].values + " " + \
                   price.loc[:,"model"].values
price["tokens"] = (price["tokens"].str.lower()
                         .apply(tokenizer.tokenize))
# cartesian product

price["key"] = 1
performance["key"] = 1
df = pd.merge(price,performance, on = "key")
# define my criteria for match
n_match = 2

df['intersection'] =\
    [len(list(set(a).intersection(set(b))))
     for a, b in zip(df.tokens_x,
                     df.tokens_y)]
df = df.loc[df["intersection"]>=n_match,:]

I redefined your datasets so that in this example we would have some matches. Here is what I have as a result:

   brand model  cores  ...  score                     tokens_y  intersection
5    AMD  3800      4  ...      5  [amd, athlon, 64, x2, 3800]             2
15   AMD  4200      4  ...     18  [amd, athlon, 64, x2, 4200]             2
[2 rows x 10 columns]

You can redefine your criteria for n_match I put two because it seemed that it was what was required by the dataset. Hope it helps

回答2:

You can merge the 2 frames after creating a 'name' column in price and then merging it with Performane on that 'name'.

Price['name'] = Price.brand + ' ' + Price.model.astype(str)
Price.merge(Performance, on='name')

However, the resulting frame will probably be empty, since at least the sample data rows you are showing in your question won't match. This is not a linguistic problem with e.g. capaitalization, but simply missing information. Only after you defining a relationship rule in "real language" you will be able to code it in Python.

来源：https://stackoverflow.com/questions/60943829/how-to-merge-two-csv-files-by-value-in-column-using-pandas-python

标签

python

pandas

fuzzywuzzy