A friend of mine had an audio file with many clicks in it. He asked me if I could remove them. I tried with the Audacity click removal tool, but it didn't work: it removed the clicks, but created a lot of artifacts in the audio, so I thought it would be a nice challenge to try to do it with Python.

The waveform looked like this:

I did it using the pydub package. The basic idea is detecting the clicks and replacing them with the previous sample. I used two different criteria to detect the clicks:

• If the average of the previous samples is below a threshold and the current sample is above another threshold.

• If the difference between the previous sample and the current one is above a threshold.

The code is not very clean, but it worked. I had to play with the thresholds and the width of the samples to get the best results. I also had to play with the width of the clicks to avoid removing too much audio …

I haven't found any example online on how to compute the MASE (Mean Absolute Scaled Error) for historical forecasts in darts, so I thought I'd share this example:

import numpy as np
from darts import TimeSeries
from darts.models import NaiveSeasonal
from darts.metrics import mase

my_data = [40, 42, 41, 45, 45, 47, 50]
my_np = np.array(my_data)
my_ts = TimeSeries.from_values(my_np)
my_model = NaiveSeasonal(K=1)
start = 3
my_forecast = my_model.historical_forecasts(series=my_ts, start=start, forecast_horizon=1)
my_mase = mase(actual_series=my_ts, pred_series=my_forecast, insample=my_ts[:start])
print(f"my_mase: {my_mase}")
my_ts.plot(label="original", marker="o")
my_forecast.plot(label="historical_forecasts", marker="x")

Notice that the insample parameter is set to the original time series up to the start of the historical forecast.

Many people is interested in trying AI (Artificial Intelligence) chat bots locally, but they don't know how to start. These are some projects that I've found that can be used to test different LLMs (Language Learning Models) and chat bots:

• Text generation web UI. It tries to be the AUTOMATIC1111/stable-diffusion-webui of text generation. Based on Gradio, it provides a web interface to test different LLMs.

• Serge. It's an interface based on Svelte as web framework and llama.cpp for running mmodels. It's dockerized. It also provides an API.

• openplayground. Works with local models and remote APIs. Can be installed as a Python package or run with Docker. It's implemented as a Flask application with a React frontend.

• ChatAI. It is a desktop application for Windows and Ubuntu developed in Python with PyQt. It uses a Google T5-Flan model and it is based on A. I. Runner, a framework to run AI models.

• ColossalChat. ColossalAI provides a set of tools to develop deep learning models, and it includes ColossalChat, which is a chat bot based …

Some times, I get this error:

Unable to connect to VS Code server: Error in request.
Error: connect ENOENT [...]

I always have to search for the solution, so I decided to write it down here. From this issue, I found that running this solves the problem:

VSCODE_IPC_HOOK_CLI=$(lsof | grep $USER | grep vscode-ipc | awk '{print $(NF-1)}' | head -n 1)

If you want to have custom metrics per each stage in an executor, you have to tag the stages (with tagWithCurrentStageIndex()) and add bogus thresholds for each metric that you want with the tag stage:i (being i the number of each stage). The bogus threshold has to be different for gauges (max>0) and rates (rate>=0).

This example file shows how to do it:

import http from 'k6/http';
import { tagWithCurrentStageIndex } from 'https://jslib.k6.io/k6-utils/1.3.0/index.js';

const stages = [
    { target: 1, duration: '10s' },
    { target: 5, duration: '10s' },
];

export const options = {
  scenarios: {
    contacts: {
      executor: 'ramping-arrival-rate',
      timeUnit: '1s',
      preAllocatedVUs: 10,
      maxVUs: 200,
      stages: stages,
    },
  },
  // Uncomment the next line if you want the count statistic
  // summaryTrendStats: ['avg', 'min', 'med', 'max', 'p(90)', 'p(95)', 'p(99)', 'count'],
  thresholds: {
    // Intentionally empty. We'll programatically define our bogus
    // thresholds (to generate the sub-metrics) below. In your real-world
    // load test, you can add any real threshoulds you want here.
  }
}

function addThreshold(thresholdName, threshold) {
    if (!options.thresholds[thresholdName]) {
        options.thresholds[thresholdName] = [];
    }

    // 'max>=0' is a …