gajomi

"""
Quick tutorial on comparing loss function for a fitted markov chain model to `baselines`.
Illustrates the effect of randomness in data.
Shows principle behind following gradients of loss to optimize the parameters
"""

import torch
from torch import tensor
from torch import nn
from torch.nn import Softmax

gajomi

Instructions:

(1) Load ICD category code descriptions into dataframe. Do not store the file on disk.
  url: https://github.com/kamillamagna/ICD-10-CSV/blob/master/categories.csv?raw=true
(2) Load ICD block descriptions into a dataframe. Do no store any files to disk
  url: https://www.aapc.com/icd-10/
  hint: consider using `pd.read_html`
(3) Count all the icd10 codes characterized as diseases according to the ICD block description
(4) From the codes in step (3) isolate though refering to 'viral' or 'virus' and concatenate, separating by semicolon

gajomi

import matplotlib.pyplot as plt
import torch
import numpy as np
from transformers import *

def get_albert_model(albert_model_name = 'albert-large-v2'):
    """get an albert model from name for thei experiment"""
    model_class, tokenizer_class, pretrained_weights = (AlbertModel, AlbertTokenizer, albert_model_name)
    tokenizer = tokenizer_class.from_pretrained(pretrained_weights)
    model = model_class.from_pretrained(pretrained_weights, output_hidden_states=True, output_attentions=True)

gajomi

gajomi

gajomi

import urllib.request, json
import pandas as pd

urlstr = "https://api.inaturalist.org/v1/observations?captive=false&taxon_name=Rubus&year=2018&lat=47.6&lng=-122.3&radius=50&order=desc&order_by=created_at"
with urllib.request.urlopen(urlstr) as url:
    data = json.loads(url.read().decode())
    
def parse_observation(observation):
    data = {}
    data['species'] = observation['taxon']['name']

gajomi

gajomi

gajomi

macro stats(setupexpr,testexpr)
    quote
        N = 2^8
        results = zeros(N,3)
        for i = 1:N
            $setupexpr
            data = @timed $testexpr
            results[i,1] = data[2]
            results[i,2] = Float64(data[3])
            results[i,3] = data[4]

gajomi

macro stats(setupexpr,testexpr)
    quote
        N = 2^8
        results = zeros(N,3)
        for i = 1:N
            $setupexpr
            data = @timed $testexpr
            results[i,1] = data[2]
            results[i,2] = Float64(data[3])
            results[i,3] = data[4]