When you do not use pandas often, searching for documentation is a time consuming step.

Core concepts

  • DataFrame (usually instantiated as df) is a table with row and column names/indices
    • row index default to 0:n, may be specified otherwise.
  • avoid chained(indexing then) assignment! e.g. df["col2"][df["col1"] > 4] = 0 will fail!

Import export

  • import from
    • files
      • table, csv: pd.read_csv …
      • json: pd.read_json
        • take note of the orientations: ‘split’, ‘records’, ‘index’, ‘columns’, ‘values’, see doc
    • memory:
      • pd.DataFrame.from_dict or from_records
  • export
    • df.to and check auto-complete

Using

  • quick view:
    • df.head(n=5)
    • df.columns for the column names
    • df.index for row indices
    • df.info df.describe for statistic (mean, std, etc)
  • access:
    • df.col_label or df['col_label'] to get one column
    • loc for label, iloc for integer index (like usual matrix)!
    • ignore .at, it is for single value access 
      df[["col1", "col2"]]                        # select multiple columns
df["col1"], df.col1                         # select the specific columns
df.loc[["row1", "row2"], ["col1", "col2"]]  # select specific rows and columns
df.loc["row4":"row7", "col3":"col5"]        # select row and column ranges
df.iloc[[1, 2, 5, 8], 5:8]                  # select by indices01il
  • others
    • df.sort_index or df.sort_values

Plotting

  • matplotlib matlab like, not the easiest, low level
  • seaborn wrapper of matplotlib, still requires matplotlib for tweaks e.g. axis etc
  • ggplot: from R, quite some learning curve
  • bokeh mcm nice and clean, interactive, but must be on notebook (more or less)
  • pygal svg, no panda integration, interactive-ish graph
  • plotly: interactive, high lvl

Sources