Case study on preferred tools in ICLR 2020

check_code_present

code_present

check_dirs

root = "."
dirlist = [item for item in os.listdir(root) if os.path.isdir(os.path.join(root, item))]
print(dirlist)

check_num_dir

len(dirlist)

check_top_github_users

code_links_df.loc[code_links_df.domains == "github.com"].links.apply(
    lambda x: urlparse(x)[2].split("/")[1]
).value_counts().head(10)

clean_repo_links

temp_link = ""


def clean_github_link(link):
    link = link.strip()
    if not link[-4:] == ".git":
        return link + ".git"
    else:
        return link


github_repo_links = (
    code_links_df.loc[code_links_df.domains == "github.com"]
    .links.apply(clean_github_link)
    .values
)

clean_tools_names

def cleaner(tool_list):
    cleaned_list = ""
    try:
        cleaned_list = []
        for tool in tool_list:
            cleaned_tool = re.findall("^\w+", tool)
            if not cleaned_tool:
                pass
            else:
                cleaned_list.append(cleaned_tool[0])
        cleaned_list = ",".join(cleaned_list)
        return cleaned_list
    except:
        tool_list = ",".join(tool_list)
        "unclean_list".join(tool_list)
        return tool_list


all_tools["all_tool_names_cleaned"] = all_tools.all_tool_names.str.split(",").apply(
    cleaner
)

create_link_df

code_links_df = pd.DataFrame({"links": code_links})

download_github_repos

# takes about 24 minutes to download
for link in github_repo_links:
    !git clone $link --depth 1 --quiet

find_sucessfully_parsed

has_req_cnt = no_req_cnt = 0
for repo in dirlist:
    path = "/content/" + repo
    if os.path.exists(path + "/requirements.txt"):
        has_req_cnt += 1
    else:
        no_req_cnt += 1

generate_requirements

# takes about 10 minutes to run
for repo in dirlist:
    path = "/content/" + repo
    if os.path.exists(path + "/requirements.txt"):
        pass
    else:
        !pipreqs $path

get_code_links

code_present = 0
code_links = []
for note in accepted_submissions:
    try:
        code_links.append(note.content["code"])
        # print("code found")
        code_present += 1
    except:
        print("Unexpected error:", sys.exc_info()[0])

get_domains

code_links_df["domains"] = code_links_df.links.apply(lambda x: urlparse(x)[1])

get_num_unique_tools

all_tools.all_tool_names_cleaned.str.split(",", expand=True).stack().unique().shape

get_score_gym

give_score("gym")

get_score_networkx

give_score("networkx")

get_score_tensorboard

give_score("tensorboard")

get_score_torch_tf_keras

give_score("torch")
print()
give_score("tensorflow", offset=12)
print()
give_score("keras")

get_score_transformers

give_score("transformers")

get_tool_count

all_tools.all_tool_names_cleaned.str.contains("torch").sum()

give_score_func

def give_score(tool_name, offset=0):
    num = all_tools.all_tool_names_cleaned.str.contains(tool_name).sum()
    num += offset
    print(
        "Count of {} is {} and total usage is {}%".format(
            tool_name, num, round((num / (all_tools.shape[0]+offset)) * 100, 4)
        )
    )

imports

%matplotlib inline

import os
import re
import sys
import requests
import openreview
import pandas as pd
import matplotlib.pyplot as plt

from random import choice
from wordcloud import WordCloud
from urllib.parse import urlparse

installation

!pip install openreview-py
!pip install pipreqs

load_all_tools_df

all_tools = pd.read_csv("all_tools.csv")

make_all_tools_df

all_tools = pd.DataFrame(
    {"all_repo_names": all_repo_names, "all_tool_names": all_tool_names}
)
all_tools.head()

openreview_api

client = openreview.Client(baseurl="https://openreview.net")

blind_notes = {
    note.id: note
    for note in openreview.tools.iterget_notes(
        client,
        invitation="ICLR.cc/2020/Conference/-/Blind_Submission",
        details="original",
    )
}

all_decision_notes = openreview.tools.iterget_notes(
    client, invitation="ICLR.cc/2020/Conference/Paper.*/-/Decision"
)

accepted_submissions = [
    blind_notes[decision_note.forum]
    for decision_note in all_decision_notes
    if "Accept" in decision_note.content["decision"]
]

len(accepted_submissions)

plot_top_10_tools

all_tools.all_tool_names_cleaned.str.split(",", expand=True).stack().value_counts()[
    :10
].plot(kind="bar")

print_head_all_tools_df_1

all_tools.head()

print_head_all_tools_df_2

all_tools.head()

print_links_value_counts

code_links_df.domains.value_counts()

print_num_parsed

has_req_cnt, no_req_cnt

print_random_link

urlparse(choice(code_links))

print_shape_all_tools_df

all_tools.shape

print_top_50_tools

all_tools.all_tool_names_cleaned.str.split(",", expand=True).stack().value_counts()[:50]

read_tools

all_repo_names = []
all_tool_names = []
for repo in dirlist:
    try:
        repo_name = repo
        with open("/content/" + repo + "/" + "requirements.txt", "r") as f:
            tools = f.readlines()
        tool_names = ",".join(tools).lower()

        all_repo_names.append(repo_name)
        all_tool_names.append(tool_names)
    except:
        print("Unexpected error for ", repo, sys.exc_info()[0])

read_tools_file

all_tools = pd.read_csv("all_tools.csv")

remove_extra_dirs

dirlist.remove('.config')
dirlist.remove('sample_data')

save_all_tools_df

all_tools.to_csv("all_tools.csv", index=False)

wordcloud_top_100_tools

all_tool_string = ",".join(all_tools.all_tool_names_cleaned)

wordcloud = WordCloud(background_color="white", max_words=100)
wordcloud.generate(all_tool_string)

plt.figure(figsize=(10, 20))
plt.imshow(wordcloud)
plt.axis("off")
plt.show()