Preprocessing of raw power data into IF inputs

neuralwatt-helper.py

import pandas as pd

root = './neuralwatt/'
in_data = 'inference.nvidia_smi_log.none_h100_llama3.3_20250316_230401.csv'

# clean and prep data
data = pd.read_csv(root+in_data)
data = data[['timestamp', 'power.draw(W)']] # extract relevant features
data.replace(' ', 'T', regex=True, inplace=True) # format timestamp
data.replace('/', '-', regex=True, inplace=True) # format timestamp

# timestamp strings to Pandas TimeStamp objects
timestamps= []
for timestamp in data['timestamp']:
    timestamps.append(pd.Timestamp(timestamp))
data['timestamp']= timestamps

# make timestamp the df index
data.set_index(['timestamp'], inplace=True)

# resample to 1 second intervals
resampled = data[['power.draw(W)']].resample('s').sum()

# create IF formatted input data from resampled dataframe

inputs = []
for idx in range(0, len(resampled), 1):
    inputs.append(f"- timestamp: {resampled.index[idx]}")
    inputs.append("  duration: 1")
    inputs.append(f"  cpu-power: {resampled['power.draw(W)'][idx]}")

# write inputs to text file
with open(root+"if-inputs.txt", "w") as txt_file:
    for line in inputs:
        txt_file.write(line + "\n")