无法对Pandas DataFrame排序及read_csv参数相关疑问求助

Hey there, let's break down your questions one by one and fix that frustrating KeyError: 'timestamp' issue!

1. Does your read_csv call meet your requirements? #

For the most part, yes—but we can make it more robust. Let's walk through your parameters:

• usecols=[0,1,2,3,4,5,6]: Correctly selects the first 7 columns of your CSV (timestamp to Volume), but using column names instead of indices is safer (more on that next).
• header=0: Properly tells pandas that the first row of your CSV is the header (column names).
• index_col=0: Sets the first column (timestamp) as the DataFrame's index—great for time-series data, as it aligns with the old example's approach.
• parse_dates=True: Automatically parses your index (timestamp) into datetime objects, which is exactly what you want for sorting and time-based operations.

2. Can you select columns by name instead of index? #

Absolutely! This is actually better practice because it avoids errors if your CSV's column order ever changes. Just replace the numeric usecols list with a list of column names matching your CSV:

usecols=['timestamp', 'open', 'high', 'low', 'close', 'adjusted_close', 'Volume']

3. What does the names parameter correspond to? #

This is a common point of confusion:

• If header=0 (your case, since your CSV has a header row), names replaces the existing column names from the CSV. For example, if your CSV's 7th column is named Volume, setting names=[..., 'volume'] will rename that column to lowercase volume.
• If your CSV had no header row, you'd use header=None and names to define column names from scratch.
• In your original code, since you set index_col=0, the timestamp column becomes the index—not a regular column—so renaming it with names doesn't affect the index label (though it's unnecessary here).

4. Is sort_values(by='timestamp') equivalent to the old .sort()? #

Almost, but with a key difference:

• In older pandas (Python 2.x era), .sort() sorted the DataFrame by its index by default. Since the old example used the datetime column as the index, .sort() was sorting by date.
• In your pandas 0.22.0, .sort() was replaced by two methods:
  • .sort_index(): Sorts by the DataFrame's index (this is the direct equivalent of the old .sort() in your case, since timestamp is your index).
  • .sort_values(): Sorts by one or more column values. Since you set timestamp as the index, it's no longer a regular column—hence the KeyError when you try to reference it here.

5. How to fix the KeyError: 'timestamp'? #

The root cause is that you're trying to sort by timestamp as a column, but you set it as the DataFrame's index with index_col=0. Here are two easy fixes:

Fix 1: Use .sort_index() (recommended for your setup)

Since timestamp is your index, sorting by the index is the most efficient approach and matches the old example's behavior:

self.symbol_data[s] = pd.read_csv(
    os.path.join(self.csv_dir, '%s.csv' % s),
    usecols=['timestamp', 'open', 'high', 'low', 'close', 'adjusted_close', 'Volume'],
    header=0,
    index_col='timestamp',  # Use column name instead of index for clarity
    parse_dates=True
).sort_index()

Fix 2: Keep timestamp as a regular column

If you prefer to keep timestamp as a column instead of the index, remove index_col=0 and explicitly parse it as a datetime:

self.symbol_data[s] = pd.read_csv(
    os.path.join(self.csv_dir, '%s.csv' % s),
    usecols=['timestamp', 'open', 'high', 'low', 'close', 'adjusted_close', 'Volume'],
    header=0,
    parse_dates=['timestamp']
).sort_values(by='timestamp')

Bonus Troubleshooting Tip

If you're still getting the error, double-check your CSV's header row! Maybe the column is named Timestamp (capital T) instead of timestamp, or there's a typo. You can quickly verify the header with:

print(pd.read_csv(os.path.join(self.csv_dir, '%s.csv' % s), nrows=0).columns)

内容的提问来源于stack exchange，提问作者Bazman